[
{
"source_record_id": "S5-170",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 17,
"source_number": 170,
"title": "Evaluating Uncertainty Calibration in Probabilistic Time Series Foundation Models",
"source_url": "https://neurips.cc/virtual/2026/poster/138974",
"raw_line": "| 1 | [P170 Evaluating Uncertainty Calibration in Probabilistic Time Series Foundation Models](https://neurips.cc/virtual/2026/poster/138974) | 71.5 | 100% | 30% | 66.7–76.2 | 0 | arXiv:2611.03592v1（2026-11-02）；与 NeurIPS 2026 接收稿是否逐字一致未核对 | **重要问题**：条件校准以非递减等序回归估计；下降或非单调条件覆盖可能被压平，低估总体QCE。有限样本同数据拟合评价还存在上偏；人口目标是保守下界，不是每次有限样本估计均为下界。；
**重要问题**：附录B.4置换检验要求序列在零假设下可交换，序列级bootstrap假定序列间独立；这些条件在所有真实数据上是否成立未独立核实。；
**重要问题**：作者给出的代码仓库在本次访问返回404，数据页面返回401；作者所述资源当前可供外部取得的完整程度未知。 | [评审](论文/P170/评审.md) · [评分数据](论文/P170/评审.json) · 无单独复核文件 |",
"raw_cells": [
"1",
"[P170 Evaluating Uncertainty Calibration in Probabilistic Time Series Foundation Models](https://neurips.cc/virtual/2026/poster/138974)",
"71.5",
"100%",
"30%",
"66.7–76.2",
"0",
"arXiv:2611.03592v1（2026-11-02）；与 NeurIPS 2026 接收稿是否逐字一致未核对",
"**重要问题**：条件校准以非递减等序回归估计；下降或非单调条件覆盖可能被压平，低估总体QCE。有限样本同数据拟合评价还存在上偏；人口目标是保守下界，不是每次有限样本估计均为下界。；
**重要问题**：附录B.4置换检验要求序列在零假设下可交换，序列级bootstrap假定序列间独立；这些条件在所有真实数据上是否成立未独立核实。；
**重要问题**：作者给出的代码仓库在本次访问返回404，数据页面返回401；作者所述资源当前可供外部取得的完整程度未知。",
"[评审](论文/P170/评审.md) · [评分数据](论文/P170/评审.json) · 无单独复核文件"
],
"normalized_title": "evaluatinguncertaintycalibrationinprobabilistictimeseriesfoundationmodels",
"normalized_url": "neurips.cc/virtual/2026/poster/138974",
"record_type": "legacy_scored",
"legacy_rank": 1,
"legacy_score": "71.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "30%",
"legacy_range": "66.7–76.2",
"citation_verbatim": "0",
"legacy_review_version": "arXiv:2611.03592v1（2026-11-02）；与 NeurIPS 2026 接收稿是否逐字一致未核对",
"legacy_risks_verbatim": "**重要问题**：条件校准以非递减等序回归估计；下降或非单调条件覆盖可能被压平，低估总体QCE。有限样本同数据拟合评价还存在上偏；人口目标是保守下界，不是每次有限样本估计均为下界。；
**重要问题**：附录B.4置换检验要求序列在零假设下可交换，序列级bootstrap假定序列间独立；这些条件在所有真实数据上是否成立未独立核实。；
**重要问题**：作者给出的代码仓库在本次访问返回404，数据页面返回401；作者所述资源当前可供外部取得的完整程度未知。",
"legacy_material_links_verbatim": "[评审](论文/P170/评审.md) · [评分数据](论文/P170/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-214",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 18,
"source_number": 214,
"title": "Position: Current Benchmarking Hinders Real Progress in Deep Learning for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/moretti26a.html",
"raw_line": "| 2 | [P214 Position: Current Benchmarking Hinders Real Progress in Deep Learning for Time Series Forecasting](https://proceedings.mlr.press/v306/moretti26a.html) | 70.5 | 100% | 35% | 66.5–75.0 | 1 | 分析正文为 arXiv 2512.22702v2 (2026-05-30)；ICML 2026 PMLR 306 官方 PDF 已下载，尚未逐字比对 | **重要问题**：主要对照覆盖4个基准和点预测；结果方向随数据集变化，跨领域和概率预测的作用尚未系统验证。；
**重要问题**：局部参数与空间注意力的替换会改变容量或组件，表1、2固定两侧相同超参；不同调参适配对结果的影响仍有歧义。 | [评审](论文/P214/评审.md) · [评分数据](论文/P214/评审.json) · 无单独复核文件 |",
"raw_cells": [
"2",
"[P214 Position: Current Benchmarking Hinders Real Progress in Deep Learning for Time Series Forecasting](https://proceedings.mlr.press/v306/moretti26a.html)",
"70.5",
"100%",
"35%",
"66.5–75.0",
"1",
"分析正文为 arXiv 2512.22702v2 (2026-05-30)；ICML 2026 PMLR 306 官方 PDF 已下载，尚未逐字比对",
"**重要问题**：主要对照覆盖4个基准和点预测；结果方向随数据集变化，跨领域和概率预测的作用尚未系统验证。；
**重要问题**：局部参数与空间注意力的替换会改变容量或组件，表1、2固定两侧相同超参；不同调参适配对结果的影响仍有歧义。",
"[评审](论文/P214/评审.md) · [评分数据](论文/P214/评审.json) · 无单独复核文件"
],
"normalized_title": "positioncurrentbenchmarkinghindersrealprogressindeeplearningfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/moretti26a.html",
"record_type": "legacy_scored",
"legacy_rank": 2,
"legacy_score": "70.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "66.5–75.0",
"citation_verbatim": "1",
"legacy_review_version": "分析正文为 arXiv 2512.22702v2 (2026-05-30)；ICML 2026 PMLR 306 官方 PDF 已下载，尚未逐字比对",
"legacy_risks_verbatim": "**重要问题**：主要对照覆盖4个基准和点预测；结果方向随数据集变化，跨领域和概率预测的作用尚未系统验证。；
**重要问题**：局部参数与空间注意力的替换会改变容量或组件，表1、2固定两侧相同超参；不同调参适配对结果的影响仍有歧义。",
"legacy_material_links_verbatim": "[评审](论文/P214/评审.md) · [评分数据](论文/P214/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-040",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 19,
"source_number": 40,
"title": "The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/c3f3c690b7a99fba16d0efd35cb83b2c-Abstract-Datasets_and_Benchmarks_Track.html",
"raw_line": "| 3 | [P040 The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c3f3c690b7a99fba16d0efd35cb83b2c-Abstract-Datasets_and_Benchmarks_Track.html) | 64.5 | 100% | 45% | 58.2–71.0 | 67 | NeurIPS 2024 Datasets and Benchmarks 正式版；31页；PDF版本日期未明确标注 | **重要问题**：多变量转单变量时用检测器与已有标签的分数筛选各指标前40%交集，之后又人工和算法联合修订；已知检测器更容易识别的序列可能更易保留，原始标签来源和逐例人工判断不完全可追溯，所见排名可能依赖所选问题分布。；
**重要问题**：VUS-PR的领先结论依赖已选合成场景、时移和随机评分比较；不同迟报代价、标注和误报预算可能改变指标适配性。 | [评审](论文/P040/评审.md) · [评分数据](论文/P040/评审.json) · 无单独复核文件 |",
"raw_cells": [
"3",
"[P040 The Elephant in the Room: Towards A Reliable Time-Series Anomaly Detection Benchmark](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c3f3c690b7a99fba16d0efd35cb83b2c-Abstract-Datasets_and_Benchmarks_Track.html)",
"64.5",
"100%",
"45%",
"58.2–71.0",
"67",
"NeurIPS 2024 Datasets and Benchmarks 正式版；31页；PDF版本日期未明确标注",
"**重要问题**：多变量转单变量时用检测器与已有标签的分数筛选各指标前40%交集，之后又人工和算法联合修订；已知检测器更容易识别的序列可能更易保留，原始标签来源和逐例人工判断不完全可追溯，所见排名可能依赖所选问题分布。；
**重要问题**：VUS-PR的领先结论依赖已选合成场景、时移和随机评分比较；不同迟报代价、标注和误报预算可能改变指标适配性。",
"[评审](论文/P040/评审.md) · [评分数据](论文/P040/评审.json) · 无单独复核文件"
],
"normalized_title": "theelephantintheroomtowardsareliabletimeseriesanomalydetectionbenchmark",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/c3f3c690b7a99fba16d0efd35cb83b2c-abstract-datasets_and_benchmarks_track.html",
"record_type": "legacy_scored",
"legacy_rank": 3,
"legacy_score": "64.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "58.2–71.0",
"citation_verbatim": "67",
"legacy_review_version": "NeurIPS 2024 Datasets and Benchmarks 正式版；31页；PDF版本日期未明确标注",
"legacy_risks_verbatim": "**重要问题**：多变量转单变量时用检测器与已有标签的分数筛选各指标前40%交集，之后又人工和算法联合修订；已知检测器更容易识别的序列可能更易保留，原始标签来源和逐例人工判断不完全可追溯，所见排名可能依赖所选问题分布。；
**重要问题**：VUS-PR的领先结论依赖已选合成场景、时移和随机评分比较；不同迟报代价、标注和误报预算可能改变指标适配性。",
"legacy_material_links_verbatim": "[评审](论文/P040/评审.md) · [评分数据](论文/P040/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-115",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 20,
"source_number": 115,
"title": "Continuous Ensemble Weather Forecasting with Diffusion models",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/424403b4d2553f16d8ad11635a8214b3-Abstract-Conference.html",
"raw_line": "| 4 | [P115 Continuous Ensemble Weather Forecasting with Diffusion models](https://proceedings.iclr.cc/paper_files/paper/2025/hash/424403b4d2553f16d8ad11635a8214b3-Abstract-Conference.html) | 61.2 | 100% | 25% | 57.5–64.0 | 4 | ICLR 2025 官方会议论文 PDF，25 页；SHA-256 e0ea4c6d136636280fac38f19c69ba11cd5b80f0e9f9797cba0f5219a3c0bd2e | **重要问题**：§4.1 从各单时刻评分函数充分匹配推向生成函数分布应反映真实解分布；单时刻边缘并未确定跨时间联合律。反例为常数标准正态过程与正弦—余弦高斯过程，二者各时刻边缘相同、联合相关不同。；
**重要问题**：算法 1 采样标准正态 Z，算法 4 初始化写 z0=σ²(s0)Z，而训练扰动 ε∼N(0,σ²I) 中 σ 是标准差。若算法 1 的 Z 直接输入算法 4，采样 σmax=80 时初始噪声标准差为 6400，同噪声等级的训练扰动标准差为 80。已视觉核对公式；算法间是否另有变换及实际实现未知。；
**重要问题**：真实 GenCast 未直接参加比较；改编 DYffusion 的条件历史较短，AR-24h 是较粗时间分辨率参考。第 5 天 ws10 的 SSR 中，Graph-EFM 为 1.01、ARCI 为 0.96，前者更接近 1；6 小时和 1 小时结果分别使用 50 和 10 个集合成员。；
**重要问题**：测试仅覆盖 2018 年，初始化逐小时且预测窗重叠；附录 C 对 ARCI-24/6h 的五次重训、每次 10 成员报告标准差，其中可能混合训练及有限集合采样波动。已读会议 PDF 未见其他模型和模型间差异的配对置信区间，也未量化多年或极端天气稳定性。 | [评审](论文/P115/评审.md) · [评分数据](论文/P115/评审.json) · [复核](论文/P115/复核.md) |",
"raw_cells": [
"4",
"[P115 Continuous Ensemble Weather Forecasting with Diffusion models](https://proceedings.iclr.cc/paper_files/paper/2025/hash/424403b4d2553f16d8ad11635a8214b3-Abstract-Conference.html)",
"61.2",
"100%",
"25%",
"57.5–64.0",
"4",
"ICLR 2025 官方会议论文 PDF，25 页；SHA-256 e0ea4c6d136636280fac38f19c69ba11cd5b80f0e9f9797cba0f5219a3c0bd2e",
"**重要问题**：§4.1 从各单时刻评分函数充分匹配推向生成函数分布应反映真实解分布；单时刻边缘并未确定跨时间联合律。反例为常数标准正态过程与正弦—余弦高斯过程，二者各时刻边缘相同、联合相关不同。；
**重要问题**：算法 1 采样标准正态 Z，算法 4 初始化写 z0=σ²(s0)Z，而训练扰动 ε∼N(0,σ²I) 中 σ 是标准差。若算法 1 的 Z 直接输入算法 4，采样 σmax=80 时初始噪声标准差为 6400，同噪声等级的训练扰动标准差为 80。已视觉核对公式；算法间是否另有变换及实际实现未知。；
**重要问题**：真实 GenCast 未直接参加比较；改编 DYffusion 的条件历史较短，AR-24h 是较粗时间分辨率参考。第 5 天 ws10 的 SSR 中，Graph-EFM 为 1.01、ARCI 为 0.96，前者更接近 1；6 小时和 1 小时结果分别使用 50 和 10 个集合成员。；
**重要问题**：测试仅覆盖 2018 年，初始化逐小时且预测窗重叠；附录 C 对 ARCI-24/6h 的五次重训、每次 10 成员报告标准差，其中可能混合训练及有限集合采样波动。已读会议 PDF 未见其他模型和模型间差异的配对置信区间，也未量化多年或极端天气稳定性。",
"[评审](论文/P115/评审.md) · [评分数据](论文/P115/评审.json) · [复核](论文/P115/复核.md)"
],
"normalized_title": "continuousensembleweatherforecastingwithdiffusionmodels",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/424403b4d2553f16d8ad11635a8214b3-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 4,
"legacy_score": "61.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "25%",
"legacy_range": "57.5–64.0",
"citation_verbatim": "4",
"legacy_review_version": "ICLR 2025 官方会议论文 PDF，25 页；SHA-256 e0ea4c6d136636280fac38f19c69ba11cd5b80f0e9f9797cba0f5219a3c0bd2e",
"legacy_risks_verbatim": "**重要问题**：§4.1 从各单时刻评分函数充分匹配推向生成函数分布应反映真实解分布；单时刻边缘并未确定跨时间联合律。反例为常数标准正态过程与正弦—余弦高斯过程，二者各时刻边缘相同、联合相关不同。；
**重要问题**：算法 1 采样标准正态 Z，算法 4 初始化写 z0=σ²(s0)Z，而训练扰动 ε∼N(0,σ²I) 中 σ 是标准差。若算法 1 的 Z 直接输入算法 4，采样 σmax=80 时初始噪声标准差为 6400，同噪声等级的训练扰动标准差为 80。已视觉核对公式；算法间是否另有变换及实际实现未知。；
**重要问题**：真实 GenCast 未直接参加比较；改编 DYffusion 的条件历史较短，AR-24h 是较粗时间分辨率参考。第 5 天 ws10 的 SSR 中，Graph-EFM 为 1.01、ARCI 为 0.96，前者更接近 1；6 小时和 1 小时结果分别使用 50 和 10 个集合成员。；
**重要问题**：测试仅覆盖 2018 年，初始化逐小时且预测窗重叠；附录 C 对 ARCI-24/6h 的五次重训、每次 10 成员报告标准差，其中可能混合训练及有限集合采样波动。已读会议 PDF 未见其他模型和模型间差异的配对置信区间，也未量化多年或极端天气稳定性。",
"legacy_material_links_verbatim": "[评审](论文/P115/评审.md) · [评分数据](论文/P115/评审.json) · [复核](论文/P115/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-016",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 21,
"source_number": 16,
"title": "GraFITi: Graphs for Forecasting Irregularly Sampled Time Series",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/29560",
"raw_line": "| 5 | [P016 GraFITi: Graphs for Forecasting Irregularly Sampled Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/29560) | 61.0 | 100% | 45% | 56.5–66.0 | 32 | AAAI 2024 出版版，38(15):16255–16263 | **重要问题**：按式(5)–(7)、(13)及算法1，L=1时查询边更新只读取初始通道、查询时间和查询边特征，故历史测量值不会影响该输出；论文搜索范围包含L=1。；
**重要问题**：论文称适配的规则序列模型比任何 IMTS 模型都差；表2中 USHCN Informer+ 0.320 低于 GRU-ODE-Bayes 0.401 和 Neural Flow 0.414。；
**重要问题**：v2附录B对规则序列基线使用默认参数，而GraFITi搜索参数；各模型完整搜索预算与适配细节未列齐，出版版又无简单时间恒定对照。；
**重要问题**：MIMIC-III 人工完全异步时，GraFITi MSE 为0.931，LinODEnet为0.894；表4同时改变稀疏度及可用信息，无法单独识别断连的因果贡献。 | [评审](论文/P016/评审.md) · [评分数据](论文/P016/评审.json) · 无单独复核文件 |",
"raw_cells": [
"5",
"[P016 GraFITi: Graphs for Forecasting Irregularly Sampled Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/29560)",
"61.0",
"100%",
"45%",
"56.5–66.0",
"32",
"AAAI 2024 出版版，38(15):16255–16263",
"**重要问题**：按式(5)–(7)、(13)及算法1，L=1时查询边更新只读取初始通道、查询时间和查询边特征，故历史测量值不会影响该输出；论文搜索范围包含L=1。；
**重要问题**：论文称适配的规则序列模型比任何 IMTS 模型都差；表2中 USHCN Informer+ 0.320 低于 GRU-ODE-Bayes 0.401 和 Neural Flow 0.414。；
**重要问题**：v2附录B对规则序列基线使用默认参数，而GraFITi搜索参数；各模型完整搜索预算与适配细节未列齐，出版版又无简单时间恒定对照。；
**重要问题**：MIMIC-III 人工完全异步时，GraFITi MSE 为0.931，LinODEnet为0.894；表4同时改变稀疏度及可用信息，无法单独识别断连的因果贡献。",
"[评审](论文/P016/评审.md) · [评分数据](论文/P016/评审.json) · 无单独复核文件"
],
"normalized_title": "grafitigraphsforforecastingirregularlysampledtimeseries",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/29560",
"record_type": "legacy_scored",
"legacy_rank": 5,
"legacy_score": "61.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "56.5–66.0",
"citation_verbatim": "32",
"legacy_review_version": "AAAI 2024 出版版，38(15):16255–16263",
"legacy_risks_verbatim": "**重要问题**：按式(5)–(7)、(13)及算法1，L=1时查询边更新只读取初始通道、查询时间和查询边特征，故历史测量值不会影响该输出；论文搜索范围包含L=1。；
**重要问题**：论文称适配的规则序列模型比任何 IMTS 模型都差；表2中 USHCN Informer+ 0.320 低于 GRU-ODE-Bayes 0.401 和 Neural Flow 0.414。；
**重要问题**：v2附录B对规则序列基线使用默认参数，而GraFITi搜索参数；各模型完整搜索预算与适配细节未列齐，出版版又无简单时间恒定对照。；
**重要问题**：MIMIC-III 人工完全异步时，GraFITi MSE 为0.931，LinODEnet为0.894；表4同时改变稀疏度及可用信息，无法单独识别断连的因果贡献。",
"legacy_material_links_verbatim": "[评审](论文/P016/评审.md) · [评分数据](论文/P016/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-065",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 22,
"source_number": 65,
"title": "Neural Conformal Control for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34029",
"raw_line": "| 6 | [P065 Neural Conformal Control for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34029) | 60.8 | 95% | 42% | 52.5–68.5 | 6 | AAAI 2025 正式版与 arXiv 2412.18144v1 补充附录 | **重要问题**：定理1只陈述 Δq 的单侧上界，附录A把该条件用于推出双侧渐近等式；窗口求和亦由近似写成等式。需补充下界或次线性等适当条件方能完成该推导。；
**重要问题**：主实验基线经排序，NCC 的测试期调整和训练/多视图机制不同；多视图在流感、天气、电力的 CS/WIS 存在取舍，不能解释为对各项指标均提升。；
**重要问题**：NCC 声称使用 scorecaster 会破坏 C-PID 理论保障；C-PID 原文定理 1 在满足有界及积分器饱和条件时允许任意 scorecaster。两者笼统说法存在张力，不能据此推断 NCC 使用的特定实现满足原定理全部条件。 | [评审](论文/P065/评审.md) · [评分数据](论文/P065/评审.json) · 无单独复核文件 |",
"raw_cells": [
"6",
"[P065 Neural Conformal Control for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34029)",
"60.8",
"95%",
"42%",
"52.5–68.5",
"6",
"AAAI 2025 正式版与 arXiv 2412.18144v1 补充附录",
"**重要问题**：定理1只陈述 Δq 的单侧上界，附录A把该条件用于推出双侧渐近等式；窗口求和亦由近似写成等式。需补充下界或次线性等适当条件方能完成该推导。；
**重要问题**：主实验基线经排序，NCC 的测试期调整和训练/多视图机制不同；多视图在流感、天气、电力的 CS/WIS 存在取舍，不能解释为对各项指标均提升。；
**重要问题**：NCC 声称使用 scorecaster 会破坏 C-PID 理论保障；C-PID 原文定理 1 在满足有界及积分器饱和条件时允许任意 scorecaster。两者笼统说法存在张力，不能据此推断 NCC 使用的特定实现满足原定理全部条件。",
"[评审](论文/P065/评审.md) · [评分数据](论文/P065/评审.json) · 无单独复核文件"
],
"normalized_title": "neuralconformalcontrolfortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34029",
"record_type": "legacy_scored",
"legacy_rank": 6,
"legacy_score": "60.8",
"legacy_coverage": "95%",
"legacy_tentative_share": "42%",
"legacy_range": "52.5–68.5",
"citation_verbatim": "6",
"legacy_review_version": "AAAI 2025 正式版与 arXiv 2412.18144v1 补充附录",
"legacy_risks_verbatim": "**重要问题**：定理1只陈述 Δq 的单侧上界，附录A把该条件用于推出双侧渐近等式；窗口求和亦由近似写成等式。需补充下界或次线性等适当条件方能完成该推导。；
**重要问题**：主实验基线经排序，NCC 的测试期调整和训练/多视图机制不同；多视图在流感、天气、电力的 CS/WIS 存在取舍，不能解释为对各项指标均提升。；
**重要问题**：NCC 声称使用 scorecaster 会破坏 C-PID 理论保障；C-PID 原文定理 1 在满足有界及积分器饱和条件时允许任意 scorecaster。两者笼统说法存在张力，不能据此推断 NCC 使用的特定实现满足原定理全部条件。",
"legacy_material_links_verbatim": "[评审](论文/P065/评审.md) · [评分数据](论文/P065/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-030",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 23,
"source_number": 30,
"title": "Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling",
"source_url": "https://proceedings.mlr.press/v235/marisca24a.html",
"raw_line": "| 7 | [P030 Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling](https://proceedings.mlr.press/v235/marisca24a.html) | 60.2 | 100% | 55% | 54.7–66.0 | 10 | ICML 2024 / PMLR 235 会议论文版（本地PDF创建时间为2024-06-02；这不是出版日期） | **重要问题**：第2页式(2)在每个节点、时间项内部除以该项有效通道数；全通道缺失时为0/0。式中对节点和时间的求和没有再归一化；有效通道数不同时，各有效值的权重也不同于统一平均。；
**重要问题**：表1、2、5、6有±但表注、实验正文和附录B未解释其含义及训练重复次数；每个数据集使用固定种子的缺失掩码。；
**重要问题**：EngRAD由ECMWF IFS模式生成且原始数据登记缺失率为0，块缺失由实验模拟；论文以五个气象量为目标报告单一MAE，未逐变量报告表2的结果。 | [评审](论文/P030/评审.md) · [评分数据](论文/P030/评审.json) · [复核](论文/P030/复核.md) |",
"raw_cells": [
"7",
"[P030 Graph-based Forecasting with Missing Data through Spatiotemporal Downsampling](https://proceedings.mlr.press/v235/marisca24a.html)",
"60.2",
"100%",
"55%",
"54.7–66.0",
"10",
"ICML 2024 / PMLR 235 会议论文版（本地PDF创建时间为2024-06-02；这不是出版日期）",
"**重要问题**：第2页式(2)在每个节点、时间项内部除以该项有效通道数；全通道缺失时为0/0。式中对节点和时间的求和没有再归一化；有效通道数不同时，各有效值的权重也不同于统一平均。；
**重要问题**：表1、2、5、6有±但表注、实验正文和附录B未解释其含义及训练重复次数；每个数据集使用固定种子的缺失掩码。；
**重要问题**：EngRAD由ECMWF IFS模式生成且原始数据登记缺失率为0，块缺失由实验模拟；论文以五个气象量为目标报告单一MAE，未逐变量报告表2的结果。",
"[评审](论文/P030/评审.md) · [评分数据](论文/P030/评审.json) · [复核](论文/P030/复核.md)"
],
"normalized_title": "graphbasedforecastingwithmissingdatathroughspatiotemporaldownsampling",
"normalized_url": "proceedings.mlr.press/v235/marisca24a.html",
"record_type": "legacy_scored",
"legacy_rank": 7,
"legacy_score": "60.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "54.7–66.0",
"citation_verbatim": "10",
"legacy_review_version": "ICML 2024 / PMLR 235 会议论文版（本地PDF创建时间为2024-06-02；这不是出版日期）",
"legacy_risks_verbatim": "**重要问题**：第2页式(2)在每个节点、时间项内部除以该项有效通道数；全通道缺失时为0/0。式中对节点和时间的求和没有再归一化；有效通道数不同时，各有效值的权重也不同于统一平均。；
**重要问题**：表1、2、5、6有±但表注、实验正文和附录B未解释其含义及训练重复次数；每个数据集使用固定种子的缺失掩码。；
**重要问题**：EngRAD由ECMWF IFS模式生成且原始数据登记缺失率为0，块缺失由实验模拟；论文以五个气象量为目标报告单一MAE，未逐变量报告表2的结果。",
"legacy_material_links_verbatim": "[评审](论文/P030/评审.md) · [评分数据](论文/P030/评审.json) · [复核](论文/P030/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-153",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 24,
"source_number": 153,
"title": "Unleashing The Power of Pre-Trained Language Models for Irregularly Sampled Time Series",
"source_url": "https://doi.org/10.1145/3711896.3737171",
"raw_line": "| 8 | [P153 Unleashing The Power of Pre-Trained Language Models for Irregularly Sampled Time Series](https://doi.org/10.1145/3711896.3737171) | 60.2 | 100% | 40% | 55.0–65.2 | 6 | arXiv:2408.08328v2（2025-06-05）；PDF带KDD 2025会议标识，未与ACM终稿逐页比对 | **重要问题**：原文对预测任务按样本随机划分；USHCN有1,114站和26,736个25个月样本、数据来自1996—1999年，平均每站24个样本。同站时间窗可能跨集合，文章未交代按站点、时间分组或归一化拟合范围。此处的潜在数据泄漏尚未经划分索引确认。；
**重要问题**：系列表示的主模型额外引入变量间PLM，并改变池化对象、层级及输出结构；论文使用平均池化。随机对照使用随机初始化可训练权重，而预训练主干主要冻结。表示、结构和优化条件存在混杂。 | [评审](论文/P153/评审.md) · [评分数据](论文/P153/评审.json) · [复核](论文/P153/复核.md) |",
"raw_cells": [
"8",
"[P153 Unleashing The Power of Pre-Trained Language Models for Irregularly Sampled Time Series](https://doi.org/10.1145/3711896.3737171)",
"60.2",
"100%",
"40%",
"55.0–65.2",
"6",
"arXiv:2408.08328v2（2025-06-05）；PDF带KDD 2025会议标识，未与ACM终稿逐页比对",
"**重要问题**：原文对预测任务按样本随机划分；USHCN有1,114站和26,736个25个月样本、数据来自1996—1999年，平均每站24个样本。同站时间窗可能跨集合，文章未交代按站点、时间分组或归一化拟合范围。此处的潜在数据泄漏尚未经划分索引确认。；
**重要问题**：系列表示的主模型额外引入变量间PLM，并改变池化对象、层级及输出结构；论文使用平均池化。随机对照使用随机初始化可训练权重，而预训练主干主要冻结。表示、结构和优化条件存在混杂。",
"[评审](论文/P153/评审.md) · [评分数据](论文/P153/评审.json) · [复核](论文/P153/复核.md)"
],
"normalized_title": "unleashingthepowerofpretrainedlanguagemodelsforirregularlysampledtimeseries",
"normalized_url": "doi.org/10.1145/3711896.3737171",
"record_type": "legacy_scored",
"legacy_rank": 8,
"legacy_score": "60.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "55.0–65.2",
"citation_verbatim": "6",
"legacy_review_version": "arXiv:2408.08328v2（2025-06-05）；PDF带KDD 2025会议标识，未与ACM终稿逐页比对",
"legacy_risks_verbatim": "**重要问题**：原文对预测任务按样本随机划分；USHCN有1,114站和26,736个25个月样本、数据来自1996—1999年，平均每站24个样本。同站时间窗可能跨集合，文章未交代按站点、时间分组或归一化拟合范围。此处的潜在数据泄漏尚未经划分索引确认。；
**重要问题**：系列表示的主模型额外引入变量间PLM，并改变池化对象、层级及输出结构；论文使用平均池化。随机对照使用随机初始化可训练权重，而预训练主干主要冻结。表示、结构和优化条件存在混杂。",
"legacy_material_links_verbatim": "[评审](论文/P153/评审.md) · [评分数据](论文/P153/评审.json) · [复核](论文/P153/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-037",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 25,
"source_number": 37,
"title": "Class-incremental Learning for Time Series: Benchmark and Evaluation",
"source_url": "https://doi.org/10.1145/3637528.3671581",
"raw_line": "| 9 | [P037 Class-incremental Learning for Time Series: Benchmark and Evaluation](https://doi.org/10.1145/3637528.3671581) | 60.0 | 95% | 16% | 55.5–63.5 | 19 | arXiv:2402.12035v2，本地 12 页 PDF；有 KDD 2024 会议标识，未逐项比对 ACM 出版版 | **重要问题**：小数据集的超参数选择使用整个任务流的验证数据；GRABMyo 和 WISDM 的训练与测试均包含相同受试者，本文有意避免受试者间分布偏移。；
**重要问题**：5 次实验同时变化类序和随机种子；表 3 称报告置信区间，全文未说明区间水平、计算方法或方法间配对差值区间。表 5 亦未单独报告配对差值不确定性。；
**重要问题**：ER 能保存原始数据，正则化方法无此条件；附录 A.2 的 ER/GR 早停耐心为 20，其他方法为 5；CLOPS 更换提取策略，FastICARL 去除量化。；
**重要问题**：生成回放避免保存原始历史样本，但论文未报告成员推断、重识别或其他隐私泄露测试，也未给形式化隐私保证。 | [评审](论文/P037/评审.md) · [评分数据](论文/P037/评审.json) · 无单独复核文件 |",
"raw_cells": [
"9",
"[P037 Class-incremental Learning for Time Series: Benchmark and Evaluation](https://doi.org/10.1145/3637528.3671581)",
"60.0",
"95%",
"16%",
"55.5–63.5",
"19",
"arXiv:2402.12035v2，本地 12 页 PDF；有 KDD 2024 会议标识，未逐项比对 ACM 出版版",
"**重要问题**：小数据集的超参数选择使用整个任务流的验证数据；GRABMyo 和 WISDM 的训练与测试均包含相同受试者，本文有意避免受试者间分布偏移。；
**重要问题**：5 次实验同时变化类序和随机种子；表 3 称报告置信区间，全文未说明区间水平、计算方法或方法间配对差值区间。表 5 亦未单独报告配对差值不确定性。；
**重要问题**：ER 能保存原始数据，正则化方法无此条件；附录 A.2 的 ER/GR 早停耐心为 20，其他方法为 5；CLOPS 更换提取策略，FastICARL 去除量化。；
**重要问题**：生成回放避免保存原始历史样本，但论文未报告成员推断、重识别或其他隐私泄露测试，也未给形式化隐私保证。",
"[评审](论文/P037/评审.md) · [评分数据](论文/P037/评审.json) · 无单独复核文件"
],
"normalized_title": "classincrementallearningfortimeseriesbenchmarkandevaluation",
"normalized_url": "doi.org/10.1145/3637528.3671581",
"record_type": "legacy_scored",
"legacy_rank": 9,
"legacy_score": "60.0",
"legacy_coverage": "95%",
"legacy_tentative_share": "16%",
"legacy_range": "55.5–63.5",
"citation_verbatim": "19",
"legacy_review_version": "arXiv:2402.12035v2，本地 12 页 PDF；有 KDD 2024 会议标识，未逐项比对 ACM 出版版",
"legacy_risks_verbatim": "**重要问题**：小数据集的超参数选择使用整个任务流的验证数据；GRABMyo 和 WISDM 的训练与测试均包含相同受试者，本文有意避免受试者间分布偏移。；
**重要问题**：5 次实验同时变化类序和随机种子；表 3 称报告置信区间，全文未说明区间水平、计算方法或方法间配对差值区间。表 5 亦未单独报告配对差值不确定性。；
**重要问题**：ER 能保存原始数据，正则化方法无此条件；附录 A.2 的 ER/GR 早停耐心为 20，其他方法为 5；CLOPS 更换提取策略，FastICARL 去除量化。；
**重要问题**：生成回放避免保存原始历史样本，但论文未报告成员推断、重识别或其他隐私泄露测试，也未给形式化隐私保证。",
"legacy_material_links_verbatim": "[评审](论文/P037/评审.md) · [评分数据](论文/P037/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-124",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 26,
"source_number": 124,
"title": "Fixing the Double Penalty in Data-Driven Weather Forecasting Through a Modified Spherical Harmonic Loss Function",
"source_url": "https://proceedings.mlr.press/v267/subich25a.html",
"raw_line": "| 10 | [P124 Fixing the Double Penalty in Data-Driven Weather Forecasting Through a Modified Spherical Harmonic Loss Function](https://proceedings.mlr.press/v267/subich25a.html) | 59.7 | 100% | 35% | 55.0–64.7 | 15 | ICML 2025 / PMLR 267 会刊版；本地 PDF，共 21 页 | **重要问题**：论文附录 B.4 式 (14) 将成员间绝对差项以加号计入所称 fair CRPS；WeatherBench 2 的对应有限集合公式为减号。图 4、11、表 2 的实际计算采用何种符号未知。；
**重要问题**：1250 km 由对照 5 天预报能量低于目标 25% 得到，160 km 由 AMSE 长时效预报振幅比超过 1 得到，两个端点使用不同阈值。；
**重要问题**：风暴和地面站对照使用原始检查点，样本限于 2022 年；图 12 在若干变量和时效显示 eRMSE 变差，表 2 报告中 MAE 在 12 项中 8 项优于 AMSE。 | [评审](论文/P124/评审.md) · [评分数据](论文/P124/评审.json) · 无单独复核文件 |",
"raw_cells": [
"10",
"[P124 Fixing the Double Penalty in Data-Driven Weather Forecasting Through a Modified Spherical Harmonic Loss Function](https://proceedings.mlr.press/v267/subich25a.html)",
"59.7",
"100%",
"35%",
"55.0–64.7",
"15",
"ICML 2025 / PMLR 267 会刊版；本地 PDF，共 21 页",
"**重要问题**：论文附录 B.4 式 (14) 将成员间绝对差项以加号计入所称 fair CRPS；WeatherBench 2 的对应有限集合公式为减号。图 4、11、表 2 的实际计算采用何种符号未知。；
**重要问题**：1250 km 由对照 5 天预报能量低于目标 25% 得到，160 km 由 AMSE 长时效预报振幅比超过 1 得到，两个端点使用不同阈值。；
**重要问题**：风暴和地面站对照使用原始检查点，样本限于 2022 年；图 12 在若干变量和时效显示 eRMSE 变差，表 2 报告中 MAE 在 12 项中 8 项优于 AMSE。",
"[评审](论文/P124/评审.md) · [评分数据](论文/P124/评审.json) · 无单独复核文件"
],
"normalized_title": "fixingthedoublepenaltyindatadrivenweatherforecastingthroughamodifiedsphericalharmoniclossfunction",
"normalized_url": "proceedings.mlr.press/v267/subich25a.html",
"record_type": "legacy_scored",
"legacy_rank": 10,
"legacy_score": "59.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "55.0–64.7",
"citation_verbatim": "15",
"legacy_review_version": "ICML 2025 / PMLR 267 会刊版；本地 PDF，共 21 页",
"legacy_risks_verbatim": "**重要问题**：论文附录 B.4 式 (14) 将成员间绝对差项以加号计入所称 fair CRPS；WeatherBench 2 的对应有限集合公式为减号。图 4、11、表 2 的实际计算采用何种符号未知。；
**重要问题**：1250 km 由对照 5 天预报能量低于目标 25% 得到，160 km 由 AMSE 长时效预报振幅比超过 1 得到，两个端点使用不同阈值。；
**重要问题**：风暴和地面站对照使用原始检查点，样本限于 2022 年；图 12 在若干变量和时效显示 eRMSE 变差，表 2 报告中 MAE 在 12 项中 8 项优于 AMSE。",
"legacy_material_links_verbatim": "[评审](论文/P124/评审.md) · [评分数据](论文/P124/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-066",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 27,
"source_number": 66,
"title": "Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34224",
"raw_line": "| 11 | [P066 Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine](https://ojs.aaai.org/index.php/AAAI/article/view/34224) | 59.7 | 100% | 35% | 54.9–64.4 | 9 | AAAI 2025 正式版；arXiv v1 另作早期版本参照 | **重要问题**：摘要称3至6个数据集使误判最佳的风险降低约40%；正文图4报告的是在有利子集下可取得前列的模型比例，未明确给出该风险定义与40%的计算。；
**重要问题**：46%/77%由13个方法与固定基准池的有利子集构造，不能解释为所有实验论文造假比例、方法普遍胜率或随机抽样概率。；
**重要问题**：论文正式版四数据集为46%/77%；论文所指当前仓库README的另一标题页面写54%/85%，尚无对应版本与配置说明。 | [评审](论文/P066/评审.md) · [评分数据](论文/P066/评审.json) · 无单独复核文件 |",
"raw_cells": [
"11",
"[P066 Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine](https://ojs.aaai.org/index.php/AAAI/article/view/34224)",
"59.7",
"100%",
"35%",
"54.9–64.4",
"9",
"AAAI 2025 正式版；arXiv v1 另作早期版本参照",
"**重要问题**：摘要称3至6个数据集使误判最佳的风险降低约40%；正文图4报告的是在有利子集下可取得前列的模型比例，未明确给出该风险定义与40%的计算。；
**重要问题**：46%/77%由13个方法与固定基准池的有利子集构造，不能解释为所有实验论文造假比例、方法普遍胜率或随机抽样概率。；
**重要问题**：论文正式版四数据集为46%/77%；论文所指当前仓库README的另一标题页面写54%/85%，尚无对应版本与配置说明。",
"[评审](论文/P066/评审.md) · [评分数据](论文/P066/评审.json) · 无单独复核文件"
],
"normalized_title": "cherrypickingintimeseriesforecastinghowtoselectdatasetstomakeyourmodelshine",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34224",
"record_type": "legacy_scored",
"legacy_rank": 11,
"legacy_score": "59.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "54.9–64.4",
"citation_verbatim": "9",
"legacy_review_version": "AAAI 2025 正式版；arXiv v1 另作早期版本参照",
"legacy_risks_verbatim": "**重要问题**：摘要称3至6个数据集使误判最佳的风险降低约40%；正文图4报告的是在有利子集下可取得前列的模型比例，未明确给出该风险定义与40%的计算。；
**重要问题**：46%/77%由13个方法与固定基准池的有利子集构造，不能解释为所有实验论文造假比例、方法普遍胜率或随机抽样概率。；
**重要问题**：论文正式版四数据集为46%/77%；论文所指当前仓库README的另一标题页面写54%/85%，尚无对应版本与配置说明。",
"legacy_material_links_verbatim": "[评审](论文/P066/评审.md) · [评分数据](论文/P066/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-086",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 28,
"source_number": 86,
"title": "CrossLinear: Plug-and-Play Cross-Correlation Embedding for Time Series Forecasting with Exogenous Variables",
"source_url": "https://doi.org/10.1145/3711896.3736899",
"raw_line": "| 12 | [P086 CrossLinear: Plug-and-Play Cross-Correlation Embedding for Time Series Forecasting with Exogenous Variables](https://doi.org/10.1145/3711896.3736899) | 59.2 | 100% | 55% | 53.7–65.0 | 18 | arXiv:2505.23116v1；本地 PDF；ACM 论文集版本未单独读取 | **重要问题**：主表及消融未报告重复运行次数、随机种子或误差区间；例如 ECL 多对一平均 MSE 的差距为 0.004，稳定性未知。；
**重要问题**：附录 B 给出 Sum 到 Cross Only 的卷积重参数化；单靠表 4 的结果无法区分参数化、初始化、训练动态和约束作用，也缺权重物理含义的独立验证。；
**重要问题**：插件在部分设置下出现退化：表 11 中 Traffic 96 步的 DLinear MSE 从 0.268 升至 0.338；ECL 的 DLinear 平均 MAE 从 0.457 升至 0.463。 | [评审](论文/P086/评审.md) · [评分数据](论文/P086/评审.json) · 无单独复核文件 |",
"raw_cells": [
"12",
"[P086 CrossLinear: Plug-and-Play Cross-Correlation Embedding for Time Series Forecasting with Exogenous Variables](https://doi.org/10.1145/3711896.3736899)",
"59.2",
"100%",
"55%",
"53.7–65.0",
"18",
"arXiv:2505.23116v1；本地 PDF；ACM 论文集版本未单独读取",
"**重要问题**：主表及消融未报告重复运行次数、随机种子或误差区间；例如 ECL 多对一平均 MSE 的差距为 0.004，稳定性未知。；
**重要问题**：附录 B 给出 Sum 到 Cross Only 的卷积重参数化；单靠表 4 的结果无法区分参数化、初始化、训练动态和约束作用，也缺权重物理含义的独立验证。；
**重要问题**：插件在部分设置下出现退化：表 11 中 Traffic 96 步的 DLinear MSE 从 0.268 升至 0.338；ECL 的 DLinear 平均 MAE 从 0.457 升至 0.463。",
"[评审](论文/P086/评审.md) · [评分数据](论文/P086/评审.json) · 无单独复核文件"
],
"normalized_title": "crosslinearplugandplaycrosscorrelationembeddingfortimeseriesforecastingwithexogenousvariables",
"normalized_url": "doi.org/10.1145/3711896.3736899",
"record_type": "legacy_scored",
"legacy_rank": 12,
"legacy_score": "59.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "53.7–65.0",
"citation_verbatim": "18",
"legacy_review_version": "arXiv:2505.23116v1；本地 PDF；ACM 论文集版本未单独读取",
"legacy_risks_verbatim": "**重要问题**：主表及消融未报告重复运行次数、随机种子或误差区间；例如 ECL 多对一平均 MSE 的差距为 0.004，稳定性未知。；
**重要问题**：附录 B 给出 Sum 到 Cross Only 的卷积重参数化；单靠表 4 的结果无法区分参数化、初始化、训练动态和约束作用，也缺权重物理含义的独立验证。；
**重要问题**：插件在部分设置下出现退化：表 11 中 Traffic 96 步的 DLinear MSE 从 0.268 升至 0.338；ECL 的 DLinear 平均 MAE 从 0.457 升至 0.463。",
"legacy_material_links_verbatim": "[评审](论文/P086/评审.md) · [评分数据](论文/P086/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-097",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 29,
"source_number": 97,
"title": "Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33908",
"raw_line": "| 13 | [P097 Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33908) | 59.2 | 100% | 35% | 55.5–63.5 | 74 | AAAI 2025 会议论文；补充理论与结果为 arXiv:2406.03751v2（2025-04-15） | **重要问题**：arXiv v2 附录 A 式 (13)→(14) 把分母由 a-b 变为 d(a-b) 时遗漏 d；式 (21)→(22) 的不等式也忽略不同列权重。；
**重要问题**：表 1 的 ETTh2/720 中 AMD MSE 0.402，TimeMixer 0.342；长期预测表现取决于数据与预测长度。；
**重要问题**：基线各模型调参预算和最佳输入长度选择程序未在已检查文字中完整交代；影响相对效果的可比性。 | [评审](论文/P097/评审.md) · [评分数据](论文/P097/评审.json) · 无单独复核文件 |",
"raw_cells": [
"13",
"[P097 Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33908)",
"59.2",
"100%",
"35%",
"55.5–63.5",
"74",
"AAAI 2025 会议论文；补充理论与结果为 arXiv:2406.03751v2（2025-04-15）",
"**重要问题**：arXiv v2 附录 A 式 (13)→(14) 把分母由 a-b 变为 d(a-b) 时遗漏 d；式 (21)→(22) 的不等式也忽略不同列权重。；
**重要问题**：表 1 的 ETTh2/720 中 AMD MSE 0.402，TimeMixer 0.342；长期预测表现取决于数据与预测长度。；
**重要问题**：基线各模型调参预算和最佳输入长度选择程序未在已检查文字中完整交代；影响相对效果的可比性。",
"[评审](论文/P097/评审.md) · [评分数据](论文/P097/评审.json) · 无单独复核文件"
],
"normalized_title": "adaptivemultiscaledecompositionframeworkfortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33908",
"record_type": "legacy_scored",
"legacy_rank": 13,
"legacy_score": "59.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "55.5–63.5",
"citation_verbatim": "74",
"legacy_review_version": "AAAI 2025 会议论文；补充理论与结果为 arXiv:2406.03751v2（2025-04-15）",
"legacy_risks_verbatim": "**重要问题**：arXiv v2 附录 A 式 (13)→(14) 把分母由 a-b 变为 d(a-b) 时遗漏 d；式 (21)→(22) 的不等式也忽略不同列权重。；
**重要问题**：表 1 的 ETTh2/720 中 AMD MSE 0.402，TimeMixer 0.342；长期预测表现取决于数据与预测长度。；
**重要问题**：基线各模型调参预算和最佳输入长度选择程序未在已检查文字中完整交代；影响相对效果的可比性。",
"legacy_material_links_verbatim": "[评审](论文/P097/评审.md) · [评分数据](论文/P097/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-114",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 30,
"source_number": 114,
"title": "ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33465",
"raw_line": "| 14 | [P114 ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33465) | 59.2 | 100% | 75% | 51.0–68.7 | 4 | AAAI-25 论文集会议 PDF；作者 2024-12-19 arXiv v1 扩展版附录补充 | **重要问题**：扩展版附录 A 报告十次重复的平均指标；所读会议正文和扩展版正文、附录文字及表格未见单次结果、标准差、置信区间或随机种子，重复时变化因素未说明；部分小幅差异缺少波动背景。；
**重要问题**：扩展版附录 A 给出 6:2:2 划分和整体 Z-score；所读文字未明确时间顺序、标准化参数拟合范围及窗口跨界处理，实际是否泄漏未知。 | [评审](论文/P114/评审.md) · [评分数据](论文/P114/评审.json) · [复核](论文/P114/复核.md) |",
"raw_cells": [
"14",
"[P114 ST-ReP: Learning Predictive Representations Efficiently for Spatial-Temporal Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33465)",
"59.2",
"100%",
"75%",
"51.0–68.7",
"4",
"AAAI-25 论文集会议 PDF；作者 2024-12-19 arXiv v1 扩展版附录补充",
"**重要问题**：扩展版附录 A 报告十次重复的平均指标；所读会议正文和扩展版正文、附录文字及表格未见单次结果、标准差、置信区间或随机种子，重复时变化因素未说明；部分小幅差异缺少波动背景。；
**重要问题**：扩展版附录 A 给出 6:2:2 划分和整体 Z-score；所读文字未明确时间顺序、标准化参数拟合范围及窗口跨界处理，实际是否泄漏未知。",
"[评审](论文/P114/评审.md) · [评分数据](论文/P114/评审.json) · [复核](论文/P114/复核.md)"
],
"normalized_title": "streplearningpredictiverepresentationsefficientlyforspatialtemporalforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33465",
"record_type": "legacy_scored",
"legacy_rank": 14,
"legacy_score": "59.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "75%",
"legacy_range": "51.0–68.7",
"citation_verbatim": "4",
"legacy_review_version": "AAAI-25 论文集会议 PDF；作者 2024-12-19 arXiv v1 扩展版附录补充",
"legacy_risks_verbatim": "**重要问题**：扩展版附录 A 报告十次重复的平均指标；所读会议正文和扩展版正文、附录文字及表格未见单次结果、标准差、置信区间或随机种子，重复时变化因素未说明；部分小幅差异缺少波动背景。；
**重要问题**：扩展版附录 A 给出 6:2:2 划分和整体 Z-score；所读文字未明确时间顺序、标准化参数拟合范围及窗口跨界处理，实际是否泄漏未知。",
"legacy_material_links_verbatim": "[评审](论文/P114/评审.md) · [评分数据](论文/P114/评审.json) · [复核](论文/P114/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-050",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 31,
"source_number": 50,
"title": "TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/29299",
"raw_line": "| 15 | [P050 TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning](https://ojs.aaai.org/index.php/AAAI/article/view/29299) | 58.7 | 95% | 26% | 52.7–63.7 | 123 | AAAI 2024 终稿（2024-03-24）；对照 arXiv v1（2023-12-25） | **重要问题**：附录表明部分基线直接采用此前论文数字，分类均值和排名排除部分缺失数据，影响比较的可比性与覆盖；未逐项核查历史版本。；
**重要问题**：主要组件消融只在 UEA 分类平均分给出；跨预测、异常等任务的机制归因仍缺直接证据。 | [评审](论文/P050/评审.md) · [评分数据](论文/P050/评审.json) · 无单独复核文件 |",
"raw_cells": [
"15",
"[P050 TimesURL: Self-Supervised Contrastive Learning for Universal Time Series Representation Learning](https://ojs.aaai.org/index.php/AAAI/article/view/29299)",
"58.7",
"95%",
"26%",
"52.7–63.7",
"123",
"AAAI 2024 终稿（2024-03-24）；对照 arXiv v1（2023-12-25）",
"**重要问题**：附录表明部分基线直接采用此前论文数字，分类均值和排名排除部分缺失数据，影响比较的可比性与覆盖；未逐项核查历史版本。；
**重要问题**：主要组件消融只在 UEA 分类平均分给出；跨预测、异常等任务的机制归因仍缺直接证据。",
"[评审](论文/P050/评审.md) · [评分数据](论文/P050/评审.json) · 无单独复核文件"
],
"normalized_title": "timesurlselfsupervisedcontrastivelearningforuniversaltimeseriesrepresentationlearning",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/29299",
"record_type": "legacy_scored",
"legacy_rank": 15,
"legacy_score": "58.7",
"legacy_coverage": "95%",
"legacy_tentative_share": "26%",
"legacy_range": "52.7–63.7",
"citation_verbatim": "123",
"legacy_review_version": "AAAI 2024 终稿（2024-03-24）；对照 arXiv v1（2023-12-25）",
"legacy_risks_verbatim": "**重要问题**：附录表明部分基线直接采用此前论文数字，分类均值和排名排除部分缺失数据，影响比较的可比性与覆盖；未逐项核查历史版本。；
**重要问题**：主要组件消融只在 UEA 分类平均分给出；跨预测、异常等任务的机制归因仍缺直接证据。",
"legacy_material_links_verbatim": "[评审](论文/P050/评审.md) · [评分数据](论文/P050/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-161",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 32,
"source_number": 161,
"title": "Dynamic Dual-Feedback Conformal Inference for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/155680",
"raw_line": "| 16 | [P161 Dynamic Dual-Feedback Conformal Inference for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/155680) | 58.4 | 95% | 63% | 46.5–70.2 | 0 | arXiv:2610.26987v1，2026-10-19，NeurIPS 2026 已录用；首次贡献公开日期未知 | **关键问题**：经独立核查，采用论文理论简化的分母2B=2，预置历史窗口构造在最多五步内违反命题4.3的阈值上界；定理4.6本身尚无直接反例。；
**重要问题**：定理4.6的上界含一般非消失偏差，因此不能将其描述为无条件长期误覆盖精确等于α。；
**重要问题**：真实序列单次表格未给配对不确定性区间；单元间差异和重叠测试样本影响普遍化解释。 | [评审](论文/P161/评审.md) · [评分数据](论文/P161/评审.json) · 无单独复核文件 |",
"raw_cells": [
"16",
"[P161 Dynamic Dual-Feedback Conformal Inference for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/155680)",
"58.4",
"95%",
"63%",
"46.5–70.2",
"0",
"arXiv:2610.26987v1，2026-10-19，NeurIPS 2026 已录用；首次贡献公开日期未知",
"**关键问题**：经独立核查，采用论文理论简化的分母2B=2，预置历史窗口构造在最多五步内违反命题4.3的阈值上界；定理4.6本身尚无直接反例。；
**重要问题**：定理4.6的上界含一般非消失偏差，因此不能将其描述为无条件长期误覆盖精确等于α。；
**重要问题**：真实序列单次表格未给配对不确定性区间；单元间差异和重叠测试样本影响普遍化解释。",
"[评审](论文/P161/评审.md) · [评分数据](论文/P161/评审.json) · 无单独复核文件"
],
"normalized_title": "dynamicdualfeedbackconformalinferencefortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/155680",
"record_type": "legacy_scored",
"legacy_rank": 16,
"legacy_score": "58.4",
"legacy_coverage": "95%",
"legacy_tentative_share": "63%",
"legacy_range": "46.5–70.2",
"citation_verbatim": "0",
"legacy_review_version": "arXiv:2610.26987v1，2026-10-19，NeurIPS 2026 已录用；首次贡献公开日期未知",
"legacy_risks_verbatim": "**关键问题**：经独立核查，采用论文理论简化的分母2B=2，预置历史窗口构造在最多五步内违反命题4.3的阈值上界；定理4.6本身尚无直接反例。；
**重要问题**：定理4.6的上界含一般非消失偏差，因此不能将其描述为无条件长期误覆盖精确等于α。；
**重要问题**：真实序列单次表格未给配对不确定性区间；单元间差异和重叠测试样本影响普遍化解释。",
"legacy_material_links_verbatim": "[评审](论文/P161/评审.md) · [评分数据](论文/P161/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-044",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 33,
"source_number": 44,
"title": "UniMTS: Unified Pre-training for Motion Time Series",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/c290d4373c495b2cad0625d6288260f0-Abstract-Conference.html",
"raw_line": "| 17 | [P044 UniMTS: Unified Pre-training for Motion Time Series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c290d4373c495b2cad0625d6288260f0-Abstract-Conference.html) | 57.7 | 100% | 40% | 53.5–61.7 | 15 | NeurIPS 2024 正式会议论文 PDF，含附录 A 和 NeurIPS checklist | **重要问题**：摘要声称全样本较最佳基线提升 9.2%；表 2 平均宏 F1 的最佳基线 BioBankSSL 为 82.7，UniMTS 为 87.5，对应相对提升约 5.8%。；
**重要问题**：§4.1 和附录 A.2–A.3 没有逐一交代 18 个数据集的受试者划分、窗口重叠、模型选择和统计配对单位；是否存在数据泄漏或其他实现问题未知。；
**重要问题**：随机旋转增强与组件消融显示平均效果，但论文未给出严格的任意旋转不变性证明，也未逐项隔离姿态、位置和类别变化；图编码替换改变架构。 | [评审](论文/P044/评审.md) · [评分数据](论文/P044/评审.json) · 无单独复核文件 |",
"raw_cells": [
"17",
"[P044 UniMTS: Unified Pre-training for Motion Time Series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c290d4373c495b2cad0625d6288260f0-Abstract-Conference.html)",
"57.7",
"100%",
"40%",
"53.5–61.7",
"15",
"NeurIPS 2024 正式会议论文 PDF，含附录 A 和 NeurIPS checklist",
"**重要问题**：摘要声称全样本较最佳基线提升 9.2%；表 2 平均宏 F1 的最佳基线 BioBankSSL 为 82.7，UniMTS 为 87.5，对应相对提升约 5.8%。；
**重要问题**：§4.1 和附录 A.2–A.3 没有逐一交代 18 个数据集的受试者划分、窗口重叠、模型选择和统计配对单位；是否存在数据泄漏或其他实现问题未知。；
**重要问题**：随机旋转增强与组件消融显示平均效果，但论文未给出严格的任意旋转不变性证明，也未逐项隔离姿态、位置和类别变化；图编码替换改变架构。",
"[评审](论文/P044/评审.md) · [评分数据](论文/P044/评审.json) · 无单独复核文件"
],
"normalized_title": "unimtsunifiedpretrainingformotiontimeseries",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/c290d4373c495b2cad0625d6288260f0-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 17,
"legacy_score": "57.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "53.5–61.7",
"citation_verbatim": "15",
"legacy_review_version": "NeurIPS 2024 正式会议论文 PDF，含附录 A 和 NeurIPS checklist",
"legacy_risks_verbatim": "**重要问题**：摘要声称全样本较最佳基线提升 9.2%；表 2 平均宏 F1 的最佳基线 BioBankSSL 为 82.7，UniMTS 为 87.5，对应相对提升约 5.8%。；
**重要问题**：§4.1 和附录 A.2–A.3 没有逐一交代 18 个数据集的受试者划分、窗口重叠、模型选择和统计配对单位；是否存在数据泄漏或其他实现问题未知。；
**重要问题**：随机旋转增强与组件消融显示平均效果，但论文未给出严格的任意旋转不变性证明，也未逐项隔离姿态、位置和类别变化；图编码替换改变架构。",
"legacy_material_links_verbatim": "[评审](论文/P044/评审.md) · [评分数据](论文/P044/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-100",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 34,
"source_number": 100,
"title": "xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34270",
"raw_line": "| 18 | [P100 xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition](https://ojs.aaai.org/index.php/AAAI/article/view/34270) | 57.2 | 100% | 70% | 47.5–68.2 | 76 | AAAI 2025 会议论文；附录使用 arXiv:2412.17323v3 | **重要问题**：会议论文称实验实现来自官方代码；当前 README 说采用其他方法发表的结果以避免重新实验。两项表述存在来源边界待澄清，尚未证明同一表格存在直接冲突。；
**重要问题**：附录 C 将分块 ADF 检验平均 p 值及阈值用于成分平稳性判断；平均 p 值本身并非经校准的整体检验，拒绝单位根也不足以单独确立一般平稳性。；
**重要问题**：主结果平均预测长度；已读主文和附录未定位种子重复与波动报告，小幅差异的稳定性缺少直接支持。 | [评审](论文/P100/评审.md) · [评分数据](论文/P100/评审.json) · 无单独复核文件 |",
"raw_cells": [
"18",
"[P100 xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition](https://ojs.aaai.org/index.php/AAAI/article/view/34270)",
"57.2",
"100%",
"70%",
"47.5–68.2",
"76",
"AAAI 2025 会议论文；附录使用 arXiv:2412.17323v3",
"**重要问题**：会议论文称实验实现来自官方代码；当前 README 说采用其他方法发表的结果以避免重新实验。两项表述存在来源边界待澄清，尚未证明同一表格存在直接冲突。；
**重要问题**：附录 C 将分块 ADF 检验平均 p 值及阈值用于成分平稳性判断；平均 p 值本身并非经校准的整体检验，拒绝单位根也不足以单独确立一般平稳性。；
**重要问题**：主结果平均预测长度；已读主文和附录未定位种子重复与波动报告，小幅差异的稳定性缺少直接支持。",
"[评审](论文/P100/评审.md) · [评分数据](论文/P100/评审.json) · 无单独复核文件"
],
"normalized_title": "xpatchdualstreamtimeseriesforecastingwithexponentialseasonaltrenddecomposition",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34270",
"record_type": "legacy_scored",
"legacy_rank": 18,
"legacy_score": "57.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "70%",
"legacy_range": "47.5–68.2",
"citation_verbatim": "76",
"legacy_review_version": "AAAI 2025 会议论文；附录使用 arXiv:2412.17323v3",
"legacy_risks_verbatim": "**重要问题**：会议论文称实验实现来自官方代码；当前 README 说采用其他方法发表的结果以避免重新实验。两项表述存在来源边界待澄清，尚未证明同一表格存在直接冲突。；
**重要问题**：附录 C 将分块 ADF 检验平均 p 值及阈值用于成分平稳性判断；平均 p 值本身并非经校准的整体检验，拒绝单位根也不足以单独确立一般平稳性。；
**重要问题**：主结果平均预测长度；已读主文和附录未定位种子重复与波动报告，小幅差异的稳定性缺少直接支持。",
"legacy_material_links_verbatim": "[评审](论文/P100/评审.md) · [评分数据](论文/P100/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-051",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 35,
"source_number": 51,
"title": "PATE: Proximity-Aware Time Series Anomaly Evaluation",
"source_url": "https://doi.org/10.1145/3637528.3671971",
"raw_line": "| 19 | [P051 PATE: Proximity-Aware Time Series Anomaly Evaluation](https://doi.org/10.1145/3637528.3671971) | 57.0 | 100% | 40% | 51.7–62.0 | 24 | arXiv:2405.12096v1，KDD 2024 Research Track 接收预印本；本次未逐页阅读正式出版版 | **重要问题**：提前缓冲区报警须在后续真实异常区间内另有命中才获提前奖励；单独的预警会按区间外误报处理，事后缓冲区命中却可获得权重。对干预后异常消退或只需要一次提前报警的用途，效用排序可能与该规则不同。；
**重要问题**：合成场景偏好由作者给定；真实数据中，用于直观论证指标有效性的证据主要是两个具体示例和选取的异常分数片段，表4另给四组基准的评分。正文和附录未给独立业务损失或专家排序作为指标正确性的外部参照。；
**重要问题**：缓冲区敏感性仅展示e=d的50、100、200、400、800五组跨数据集均值，分数随缓冲区增大而变化；各数据集、非对称缓冲、随机重复及不确定性未在正文和附录给出。 | [评审](论文/P051/评审.md) · [评分数据](论文/P051/评审.json) · 无单独复核文件 |",
"raw_cells": [
"19",
"[P051 PATE: Proximity-Aware Time Series Anomaly Evaluation](https://doi.org/10.1145/3637528.3671971)",
"57.0",
"100%",
"40%",
"51.7–62.0",
"24",
"arXiv:2405.12096v1，KDD 2024 Research Track 接收预印本；本次未逐页阅读正式出版版",
"**重要问题**：提前缓冲区报警须在后续真实异常区间内另有命中才获提前奖励；单独的预警会按区间外误报处理，事后缓冲区命中却可获得权重。对干预后异常消退或只需要一次提前报警的用途，效用排序可能与该规则不同。；
**重要问题**：合成场景偏好由作者给定；真实数据中，用于直观论证指标有效性的证据主要是两个具体示例和选取的异常分数片段，表4另给四组基准的评分。正文和附录未给独立业务损失或专家排序作为指标正确性的外部参照。；
**重要问题**：缓冲区敏感性仅展示e=d的50、100、200、400、800五组跨数据集均值，分数随缓冲区增大而变化；各数据集、非对称缓冲、随机重复及不确定性未在正文和附录给出。",
"[评审](论文/P051/评审.md) · [评分数据](论文/P051/评审.json) · 无单独复核文件"
],
"normalized_title": "pateproximityawaretimeseriesanomalyevaluation",
"normalized_url": "doi.org/10.1145/3637528.3671971",
"record_type": "legacy_scored",
"legacy_rank": 19,
"legacy_score": "57.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "51.7–62.0",
"citation_verbatim": "24",
"legacy_review_version": "arXiv:2405.12096v1，KDD 2024 Research Track 接收预印本；本次未逐页阅读正式出版版",
"legacy_risks_verbatim": "**重要问题**：提前缓冲区报警须在后续真实异常区间内另有命中才获提前奖励；单独的预警会按区间外误报处理，事后缓冲区命中却可获得权重。对干预后异常消退或只需要一次提前报警的用途，效用排序可能与该规则不同。；
**重要问题**：合成场景偏好由作者给定；真实数据中，用于直观论证指标有效性的证据主要是两个具体示例和选取的异常分数片段，表4另给四组基准的评分。正文和附录未给独立业务损失或专家排序作为指标正确性的外部参照。；
**重要问题**：缓冲区敏感性仅展示e=d的50、100、200、400、800五组跨数据集均值，分数随缓冲区增大而变化；各数据集、非对称缓冲、随机重复及不确定性未在正文和附录给出。",
"legacy_material_links_verbatim": "[评审](论文/P051/评审.md) · [评分数据](论文/P051/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-002",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 36,
"source_number": 2,
"title": "Calibration of Time-Series Forecasting: Detecting and Adapting Context-Driven Distribution Shift",
"source_url": "https://doi.org/10.1145/3637528.3671926",
"raw_line": "| 20 | [P002 Calibration of Time-Series Forecasting: Detecting and Adapting Context-Driven Distribution Shift](https://doi.org/10.1145/3637528.3671926) | 56.7 | 100% | 75% | 46.0–68.5 | 22 | KDD 2024 完整15页 PDF，首次公开 arXiv v1 2023-10-23，arXiv v2 2024-06-11；版本逐行比对未做 | **重要问题**：§4 的 P(Y|X) 混合公式以 P(C) 而非一般所需的 P(C|X) 加权；正文设想 C 同时影响 X，因此所列情形无需自动满足二者相等。；
**重要问题**：定理1–2比较固定设计全局和分组最小二乘的端点风险；并未给有限步 SOLID 对学习率及步数的全局最优性结论。；
**重要问题**：§5.1 用 MSE 支持残差高斯假设；MSE 最小化本身不要求噪声正态，所计算高斯 KL 是模型近似。 | [评审](论文/P002/评审.md) · [评分数据](论文/P002/评审.json) · 无单独复核文件 |",
"raw_cells": [
"20",
"[P002 Calibration of Time-Series Forecasting: Detecting and Adapting Context-Driven Distribution Shift](https://doi.org/10.1145/3637528.3671926)",
"56.7",
"100%",
"75%",
"46.0–68.5",
"22",
"KDD 2024 完整15页 PDF，首次公开 arXiv v1 2023-10-23，arXiv v2 2024-06-11；版本逐行比对未做",
"**重要问题**：§4 的 P(Y|X) 混合公式以 P(C) 而非一般所需的 P(C|X) 加权；正文设想 C 同时影响 X，因此所列情形无需自动满足二者相等。；
**重要问题**：定理1–2比较固定设计全局和分组最小二乘的端点风险；并未给有限步 SOLID 对学习率及步数的全局最优性结论。；
**重要问题**：§5.1 用 MSE 支持残差高斯假设；MSE 最小化本身不要求噪声正态，所计算高斯 KL 是模型近似。",
"[评审](论文/P002/评审.md) · [评分数据](论文/P002/评审.json) · 无单独复核文件"
],
"normalized_title": "calibrationoftimeseriesforecastingdetectingandadaptingcontextdrivendistributionshift",
"normalized_url": "doi.org/10.1145/3637528.3671926",
"record_type": "legacy_scored",
"legacy_rank": 20,
"legacy_score": "56.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "75%",
"legacy_range": "46.0–68.5",
"citation_verbatim": "22",
"legacy_review_version": "KDD 2024 完整15页 PDF，首次公开 arXiv v1 2023-10-23，arXiv v2 2024-06-11；版本逐行比对未做",
"legacy_risks_verbatim": "**重要问题**：§4 的 P(Y|X) 混合公式以 P(C) 而非一般所需的 P(C|X) 加权；正文设想 C 同时影响 X，因此所列情形无需自动满足二者相等。；
**重要问题**：定理1–2比较固定设计全局和分组最小二乘的端点风险；并未给有限步 SOLID 对学习率及步数的全局最优性结论。；
**重要问题**：§5.1 用 MSE 支持残差高斯假设；MSE 最小化本身不要求噪声正态，所计算高斯 KL 是模型近似。",
"legacy_material_links_verbatim": "[评审](论文/P002/评审.md) · [评分数据](论文/P002/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-061",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 37,
"source_number": 61,
"title": "Merlin: Multi-View Representation Learning for Robust Multivariate Time Series Forecasting with Unfixed Missing Rates",
"source_url": "https://doi.org/10.1145/3711896.3737046",
"raw_line": "| 21 | [P061 Merlin: Multi-View Representation Learning for Robust Multivariate Time Series Forecasting with Unfixed Missing Rates](https://doi.org/10.1145/3711896.3737046) | 56.7 | 100% | 55% | 51.2–62.7 | 11 | arXiv:2506.12459v1；带 KDD 2025 会议信息的 12 页稿，未与 ACM 出版版本逐字对照 | **重要问题**：所评 arXiv v1 报告五随机种子的均值，但主结果未给对应的标准差、区间或逐种子配对结果；部分优势较小。旧 ICLR 稿报告标准差，模型与比较设置不同。；
**重要问题**：v1 参考文献包含 ICML 2024 缺失数据预测工作和 GinAR+，但表 2、3、5 未将它们列为对照；各基线调参及主表训练细节报告有限。；
**重要问题**：v1 的主评测为合成随机点缺失与四档分段缺失；自然故障形成的缺失机制、未见缺失率及不同预测步长未在该版本结果中单独报告。ICLR 旧稿另有特定长预测和两种合成缺失测试。 | [评审](论文/P061/评审.md) · [评分数据](论文/P061/评审.json) · [复核](论文/P061/复核.md) |",
"raw_cells": [
"21",
"[P061 Merlin: Multi-View Representation Learning for Robust Multivariate Time Series Forecasting with Unfixed Missing Rates](https://doi.org/10.1145/3711896.3737046)",
"56.7",
"100%",
"55%",
"51.2–62.7",
"11",
"arXiv:2506.12459v1；带 KDD 2025 会议信息的 12 页稿，未与 ACM 出版版本逐字对照",
"**重要问题**：所评 arXiv v1 报告五随机种子的均值，但主结果未给对应的标准差、区间或逐种子配对结果；部分优势较小。旧 ICLR 稿报告标准差，模型与比较设置不同。；
**重要问题**：v1 参考文献包含 ICML 2024 缺失数据预测工作和 GinAR+，但表 2、3、5 未将它们列为对照；各基线调参及主表训练细节报告有限。；
**重要问题**：v1 的主评测为合成随机点缺失与四档分段缺失；自然故障形成的缺失机制、未见缺失率及不同预测步长未在该版本结果中单独报告。ICLR 旧稿另有特定长预测和两种合成缺失测试。",
"[评审](论文/P061/评审.md) · [评分数据](论文/P061/评审.json) · [复核](论文/P061/复核.md)"
],
"normalized_title": "merlinmultiviewrepresentationlearningforrobustmultivariatetimeseriesforecastingwithunfixedmissingrates",
"normalized_url": "doi.org/10.1145/3711896.3737046",
"record_type": "legacy_scored",
"legacy_rank": 21,
"legacy_score": "56.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "51.2–62.7",
"citation_verbatim": "11",
"legacy_review_version": "arXiv:2506.12459v1；带 KDD 2025 会议信息的 12 页稿，未与 ACM 出版版本逐字对照",
"legacy_risks_verbatim": "**重要问题**：所评 arXiv v1 报告五随机种子的均值，但主结果未给对应的标准差、区间或逐种子配对结果；部分优势较小。旧 ICLR 稿报告标准差，模型与比较设置不同。；
**重要问题**：v1 参考文献包含 ICML 2024 缺失数据预测工作和 GinAR+，但表 2、3、5 未将它们列为对照；各基线调参及主表训练细节报告有限。；
**重要问题**：v1 的主评测为合成随机点缺失与四档分段缺失；自然故障形成的缺失机制、未见缺失率及不同预测步长未在该版本结果中单独报告。ICLR 旧稿另有特定长预测和两种合成缺失测试。",
"legacy_material_links_verbatim": "[评审](论文/P061/评审.md) · [评分数据](论文/P061/评审.json) · [复核](论文/P061/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-017",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 38,
"source_number": 17,
"title": "Heterogeneity-Informed Meta-Parameter Learning for Spatiotemporal Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3637528.3671961",
"raw_line": "| 22 | [P017 Heterogeneity-Informed Meta-Parameter Learning for Spatiotemporal Time Series Forecasting](https://doi.org/10.1145/3637528.3671961) | 56.5 | 100% | 40% | 51.5–60.7 | 73 | arXiv v2，KDD 2024 论文的作者后续修订版；附加 PEMS03 | **重要问题**：2024-09-03 v2 表3增列PEMS03：HimNet MAE=15.11、MegaCRN=14.57，但§5.2仍写所有数据和指标优于所有基线；作者最初5数据集主张须与后加第6项分开。；
**重要问题**：主表和消融未报告多种子差异/置信区间；PEMS08 MAE 13.57 vs 13.58 差0.01，稳定优势无法从表格判断。；
**重要问题**：交通五基准同属加州道路，节点嵌入依赖固定传感器；跨城市、不同节点和其他任务的泛化声明未覆盖。；
**重要问题**：作者表3未列2023年STAEformer，该前作在自己的METRLA 60分钟表报MAE 3.34、本文报3.37；二者数据与调参是否严格一致未查，不能据此判定真实胜负。 | [评审](论文/P017/评审.md) · [评分数据](论文/P017/评审.json) · 无单独复核文件 |",
"raw_cells": [
"22",
"[P017 Heterogeneity-Informed Meta-Parameter Learning for Spatiotemporal Time Series Forecasting](https://doi.org/10.1145/3637528.3671961)",
"56.5",
"100%",
"40%",
"51.5–60.7",
"73",
"arXiv v2，KDD 2024 论文的作者后续修订版；附加 PEMS03",
"**重要问题**：2024-09-03 v2 表3增列PEMS03：HimNet MAE=15.11、MegaCRN=14.57，但§5.2仍写所有数据和指标优于所有基线；作者最初5数据集主张须与后加第6项分开。；
**重要问题**：主表和消融未报告多种子差异/置信区间；PEMS08 MAE 13.57 vs 13.58 差0.01，稳定优势无法从表格判断。；
**重要问题**：交通五基准同属加州道路，节点嵌入依赖固定传感器；跨城市、不同节点和其他任务的泛化声明未覆盖。；
**重要问题**：作者表3未列2023年STAEformer，该前作在自己的METRLA 60分钟表报MAE 3.34、本文报3.37；二者数据与调参是否严格一致未查，不能据此判定真实胜负。",
"[评审](论文/P017/评审.md) · [评分数据](论文/P017/评审.json) · 无单独复核文件"
],
"normalized_title": "heterogeneityinformedmetaparameterlearningforspatiotemporaltimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3637528.3671961",
"record_type": "legacy_scored",
"legacy_rank": 22,
"legacy_score": "56.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "51.5–60.7",
"citation_verbatim": "73",
"legacy_review_version": "arXiv v2，KDD 2024 论文的作者后续修订版；附加 PEMS03",
"legacy_risks_verbatim": "**重要问题**：2024-09-03 v2 表3增列PEMS03：HimNet MAE=15.11、MegaCRN=14.57，但§5.2仍写所有数据和指标优于所有基线；作者最初5数据集主张须与后加第6项分开。；
**重要问题**：主表和消融未报告多种子差异/置信区间；PEMS08 MAE 13.57 vs 13.58 差0.01，稳定优势无法从表格判断。；
**重要问题**：交通五基准同属加州道路，节点嵌入依赖固定传感器；跨城市、不同节点和其他任务的泛化声明未覆盖。；
**重要问题**：作者表3未列2023年STAEformer，该前作在自己的METRLA 60分钟表报MAE 3.34、本文报3.37；二者数据与调参是否严格一致未查，不能据此判定真实胜负。",
"legacy_material_links_verbatim": "[评审](论文/P017/评审.md) · [评分数据](论文/P017/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-001",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 39,
"source_number": 1,
"title": "RHiOTS: A Framework for Evaluating Hierarchical Time Series Forecasting Algorithms",
"source_url": "https://doi.org/10.1145/3637528.3672062",
"raw_line": "| 23 | [P001 RHiOTS: A Framework for Evaluating Hierarchical Time Series Forecasting Algorithms](https://doi.org/10.1145/3637528.3672062) | 56.3 | 95% | 79% | 43.5–70.5 | 6 | arXiv:2408.03399v1，2024-08-06；KDD出版2024-08-24，是否与终稿逐字一致未核 | **重要问题**：§2 定义 τ 为预测长度，但 MASE 式从 T+1 累加至 τ；§4.2 缩放系数中心写为 0，低强度极限不是原序列。两处涉及评估定义和扰动语义，实际软件计算方式未查。；
**重要问题**：三个数据集及有限模型和人为扰动无法直接覆盖全部现实变化；M5/Police挑选高计数序列，选择影响仅以未展示的预实验支持。 | [评审](论文/P001/评审.md) · [评分数据](论文/P001/评审.json) · 无单独复核文件 |",
"raw_cells": [
"23",
"[P001 RHiOTS: A Framework for Evaluating Hierarchical Time Series Forecasting Algorithms](https://doi.org/10.1145/3637528.3672062)",
"56.3",
"95%",
"79%",
"43.5–70.5",
"6",
"arXiv:2408.03399v1，2024-08-06；KDD出版2024-08-24，是否与终稿逐字一致未核",
"**重要问题**：§2 定义 τ 为预测长度，但 MASE 式从 T+1 累加至 τ；§4.2 缩放系数中心写为 0，低强度极限不是原序列。两处涉及评估定义和扰动语义，实际软件计算方式未查。；
**重要问题**：三个数据集及有限模型和人为扰动无法直接覆盖全部现实变化；M5/Police挑选高计数序列，选择影响仅以未展示的预实验支持。",
"[评审](论文/P001/评审.md) · [评分数据](论文/P001/评审.json) · 无单独复核文件"
],
"normalized_title": "rhiotsaframeworkforevaluatinghierarchicaltimeseriesforecastingalgorithms",
"normalized_url": "doi.org/10.1145/3637528.3672062",
"record_type": "legacy_scored",
"legacy_rank": 23,
"legacy_score": "56.3",
"legacy_coverage": "95%",
"legacy_tentative_share": "79%",
"legacy_range": "43.5–70.5",
"citation_verbatim": "6",
"legacy_review_version": "arXiv:2408.03399v1，2024-08-06；KDD出版2024-08-24，是否与终稿逐字一致未核",
"legacy_risks_verbatim": "**重要问题**：§2 定义 τ 为预测长度，但 MASE 式从 T+1 累加至 τ；§4.2 缩放系数中心写为 0，低强度极限不是原序列。两处涉及评估定义和扰动语义，实际软件计算方式未查。；
**重要问题**：三个数据集及有限模型和人为扰动无法直接覆盖全部现实变化；M5/Police挑选高计数序列，选择影响仅以未展示的预实验支持。",
"legacy_material_links_verbatim": "[评审](论文/P001/评审.md) · [评分数据](论文/P001/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-092",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 40,
"source_number": 92,
"title": "Proactive Model Adaptation Against Concept Drift for Online Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709210",
"raw_line": "| 24 | [P092 Proactive Model Adaptation Against Concept Drift for Online Time Series Forecasting](https://doi.org/10.1145/3690624.3709210) | 55.7 | 100% | 70% | 48.2–63.5 | 33 | arXiv:2412.08435v5，本地 13 页预印本；KDD 2025 会议相关版本 | **重要问题**：Optimal 的更新用 Y_{t−1}=[v_t,…,v_{t+H−1}]，与待预测的 Y_t 共享 H−1 个未来值；它是作者明示的不可部署对照。该差距混合泄漏信息与时间间隔，不能独立识别漂移影响。；
**重要问题**：附录 A 由连续性推出局部 L_upper/L_lower→1 的推导有反例：F(θ1,θ2)=(θ1,10θ2) 连续且线性，在任意半径内比值仍为 10；式 (6)–(7) 另为所需的接近性假设，从输出接近最优输出推到标签误差更小还需要条件。；
**重要问题**：表 3–5 给三次运行均值而未给分散度或时间相关误差区间，部分差异接近表格显示精度；主表比较也依赖重实现和 SOLID++ 改写。 | [评审](论文/P092/评审.md) · [评分数据](论文/P092/评审.json) · 无单独复核文件 |",
"raw_cells": [
"24",
"[P092 Proactive Model Adaptation Against Concept Drift for Online Time Series Forecasting](https://doi.org/10.1145/3690624.3709210)",
"55.7",
"100%",
"70%",
"48.2–63.5",
"33",
"arXiv:2412.08435v5，本地 13 页预印本；KDD 2025 会议相关版本",
"**重要问题**：Optimal 的更新用 Y_{t−1}=[v_t,…,v_{t+H−1}]，与待预测的 Y_t 共享 H−1 个未来值；它是作者明示的不可部署对照。该差距混合泄漏信息与时间间隔，不能独立识别漂移影响。；
**重要问题**：附录 A 由连续性推出局部 L_upper/L_lower→1 的推导有反例：F(θ1,θ2)=(θ1,10θ2) 连续且线性，在任意半径内比值仍为 10；式 (6)–(7) 另为所需的接近性假设，从输出接近最优输出推到标签误差更小还需要条件。；
**重要问题**：表 3–5 给三次运行均值而未给分散度或时间相关误差区间，部分差异接近表格显示精度；主表比较也依赖重实现和 SOLID++ 改写。",
"[评审](论文/P092/评审.md) · [评分数据](论文/P092/评审.json) · 无单独复核文件"
],
"normalized_title": "proactivemodeladaptationagainstconceptdriftforonlinetimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709210",
"record_type": "legacy_scored",
"legacy_rank": 24,
"legacy_score": "55.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "70%",
"legacy_range": "48.2–63.5",
"citation_verbatim": "33",
"legacy_review_version": "arXiv:2412.08435v5，本地 13 页预印本；KDD 2025 会议相关版本",
"legacy_risks_verbatim": "**重要问题**：Optimal 的更新用 Y_{t−1}=[v_t,…,v_{t+H−1}]，与待预测的 Y_t 共享 H−1 个未来值；它是作者明示的不可部署对照。该差距混合泄漏信息与时间间隔，不能独立识别漂移影响。；
**重要问题**：附录 A 由连续性推出局部 L_upper/L_lower→1 的推导有反例：F(θ1,θ2)=(θ1,10θ2) 连续且线性，在任意半径内比值仍为 10；式 (6)–(7) 另为所需的接近性假设，从输出接近最优输出推到标签误差更小还需要条件。；
**重要问题**：表 3–5 给三次运行均值而未给分散度或时间相关误差区间，部分差异接近表格显示精度；主表比较也依赖重实现和 SOLID++ 改写。",
"legacy_material_links_verbatim": "[评审](论文/P092/评审.md) · [评分数据](论文/P092/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-167",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 41,
"source_number": 167,
"title": "Ensemble Modeling for Time Series Forecasting: an Adaptive Robust Optimization Approach",
"source_url": "https://neurips.cc/virtual/2026/poster/149487",
"raw_line": "| 25 | [P167 Ensemble Modeling for Time Series Forecasting: an Adaptive Robust Optimization Approach](https://neurips.cc/virtual/2026/poster/149487) | 55.7 | 100% | 45% | 50.0–62.0 | 11 | arXiv:2304.04308v1，2023-04-09，两作者预印本；非 2026 会议版 | **重要问题**：第 18 页印刷的 Exp3 式把归一化分母放在 exp 内：当所有 regret 为 0 且 m>1 时，每个权重等于 1，权重和为 m，与第 27 页和为 1 的叙述冲突。；
**重要问题**：设备能耗数据每 10 分钟取样，论文称预测一小时内能耗，却未明确写出目标与过去误差的时间索引和标签揭示时点；默认规则使用上一时点误差。这是已核实的报告缺口。；
**重要问题**：理论等价式固定由已观测 X、y 形成的 Z 和 Ω_adapt，且使用整个扩展矩阵的范数球；该球允许非块对角扰动。作者将方法联系模型漂移，但该等价式没有重算受扰动轨迹产生的后续误差。；
**重要问题**：真实数据表格只报告单个测试划分的点指标，未报告逐风暴或时间块不确定性；气旋两海盆测试分别有 261 和 254 个时点，CVaR 5% 对应约 13 个样本的尾部权重。 | [评审](论文/P167/评审.md) · [评分数据](论文/P167/评审.json) · [复核](论文/P167/复核.md) |",
"raw_cells": [
"25",
"[P167 Ensemble Modeling for Time Series Forecasting: an Adaptive Robust Optimization Approach](https://neurips.cc/virtual/2026/poster/149487)",
"55.7",
"100%",
"45%",
"50.0–62.0",
"11",
"arXiv:2304.04308v1，2023-04-09，两作者预印本；非 2026 会议版",
"**重要问题**：第 18 页印刷的 Exp3 式把归一化分母放在 exp 内：当所有 regret 为 0 且 m>1 时，每个权重等于 1，权重和为 m，与第 27 页和为 1 的叙述冲突。；
**重要问题**：设备能耗数据每 10 分钟取样，论文称预测一小时内能耗，却未明确写出目标与过去误差的时间索引和标签揭示时点；默认规则使用上一时点误差。这是已核实的报告缺口。；
**重要问题**：理论等价式固定由已观测 X、y 形成的 Z 和 Ω_adapt，且使用整个扩展矩阵的范数球；该球允许非块对角扰动。作者将方法联系模型漂移，但该等价式没有重算受扰动轨迹产生的后续误差。；
**重要问题**：真实数据表格只报告单个测试划分的点指标，未报告逐风暴或时间块不确定性；气旋两海盆测试分别有 261 和 254 个时点，CVaR 5% 对应约 13 个样本的尾部权重。",
"[评审](论文/P167/评审.md) · [评分数据](论文/P167/评审.json) · [复核](论文/P167/复核.md)"
],
"normalized_title": "ensemblemodelingfortimeseriesforecastinganadaptiverobustoptimizationapproach",
"normalized_url": "neurips.cc/virtual/2026/poster/149487",
"record_type": "legacy_scored",
"legacy_rank": 25,
"legacy_score": "55.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "50.0–62.0",
"citation_verbatim": "11",
"legacy_review_version": "arXiv:2304.04308v1，2023-04-09，两作者预印本；非 2026 会议版",
"legacy_risks_verbatim": "**重要问题**：第 18 页印刷的 Exp3 式把归一化分母放在 exp 内：当所有 regret 为 0 且 m>1 时，每个权重等于 1，权重和为 m，与第 27 页和为 1 的叙述冲突。；
**重要问题**：设备能耗数据每 10 分钟取样，论文称预测一小时内能耗，却未明确写出目标与过去误差的时间索引和标签揭示时点；默认规则使用上一时点误差。这是已核实的报告缺口。；
**重要问题**：理论等价式固定由已观测 X、y 形成的 Z 和 Ω_adapt，且使用整个扩展矩阵的范数球；该球允许非块对角扰动。作者将方法联系模型漂移，但该等价式没有重算受扰动轨迹产生的后续误差。；
**重要问题**：真实数据表格只报告单个测试划分的点指标，未报告逐风暴或时间块不确定性；气旋两海盆测试分别有 261 和 254 个时点，CVaR 5% 对应约 13 个样本的尾部权重。",
"legacy_material_links_verbatim": "[评审](论文/P167/评审.md) · [评分数据](论文/P167/评审.json) · [复核](论文/P167/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-110",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 42,
"source_number": 110,
"title": "HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/32057",
"raw_line": "| 26 | [P110 HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/32057) | 55.0 | 100% | 45% | 49.2–59.7 | 13 | AAAI-25 会议论文集出版版，39(1):746–754；本地9页PDF | **重要问题**：已读正文与预印本附录文字和表格未交代数据时间划分边界、预处理参数拟合范围及选择模型或温度所用子集；由此产生评测可追溯性缺口。未观察到实际泄漏的直接证据。；
**重要问题**：表1 Electricity 的 HDT CRPS_sum 三个预测期为0.025、0.028、0.036，算术均值约0.0297，表中 Avg 为0.028；图3 Taxi 48步 HDT 的 NRMSE_sum 标为0.317，表1同名同预测期为0.264，文中未解释差异。；
**重要问题**：主指标评估变量求和序列；预印本附录表6的名义95%区间在 Taxi 48与96步的 HDT PICP 分别为54.03%与49.24%，距名义值分别40.97与45.76个百分点。论文未提供非交换共形覆盖保证。 | [评审](论文/P110/评审.md) · [评分数据](论文/P110/评审.json) · 无单独复核文件 |",
"raw_cells": [
"26",
"[P110 HDT: Hierarchical Discrete Transformer for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/32057)",
"55.0",
"100%",
"45%",
"49.2–59.7",
"13",
"AAAI-25 会议论文集出版版，39(1):746–754；本地9页PDF",
"**重要问题**：已读正文与预印本附录文字和表格未交代数据时间划分边界、预处理参数拟合范围及选择模型或温度所用子集；由此产生评测可追溯性缺口。未观察到实际泄漏的直接证据。；
**重要问题**：表1 Electricity 的 HDT CRPS_sum 三个预测期为0.025、0.028、0.036，算术均值约0.0297，表中 Avg 为0.028；图3 Taxi 48步 HDT 的 NRMSE_sum 标为0.317，表1同名同预测期为0.264，文中未解释差异。；
**重要问题**：主指标评估变量求和序列；预印本附录表6的名义95%区间在 Taxi 48与96步的 HDT PICP 分别为54.03%与49.24%，距名义值分别40.97与45.76个百分点。论文未提供非交换共形覆盖保证。",
"[评审](论文/P110/评审.md) · [评分数据](论文/P110/评审.json) · 无单独复核文件"
],
"normalized_title": "hdthierarchicaldiscretetransformerformultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/32057",
"record_type": "legacy_scored",
"legacy_rank": 26,
"legacy_score": "55.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "49.2–59.7",
"citation_verbatim": "13",
"legacy_review_version": "AAAI-25 会议论文集出版版，39(1):746–754；本地9页PDF",
"legacy_risks_verbatim": "**重要问题**：已读正文与预印本附录文字和表格未交代数据时间划分边界、预处理参数拟合范围及选择模型或温度所用子集；由此产生评测可追溯性缺口。未观察到实际泄漏的直接证据。；
**重要问题**：表1 Electricity 的 HDT CRPS_sum 三个预测期为0.025、0.028、0.036，算术均值约0.0297，表中 Avg 为0.028；图3 Taxi 48步 HDT 的 NRMSE_sum 标为0.317，表1同名同预测期为0.264，文中未解释差异。；
**重要问题**：主指标评估变量求和序列；预印本附录表6的名义95%区间在 Taxi 48与96步的 HDT PICP 分别为54.03%与49.24%，距名义值分别40.97与45.76个百分点。论文未提供非交换共形覆盖保证。",
"legacy_material_links_verbatim": "[评审](论文/P110/评审.md) · [评分数据](论文/P110/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-131",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 43,
"source_number": 131,
"title": "AutoSTF: Decoupled Neural Architecture Search for Cost-Effective Automated Spatio-Temporal Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709323",
"raw_line": "| 27 | [P131 AutoSTF: Decoupled Neural Architecture Search for Cost-Effective Automated Spatio-Temporal Forecasting](https://doi.org/10.1145/3690624.3709323) | 55.0 | 100% | 50% | 47.7–60.5 | 8 | arXiv:2409.16586v2，本地 PDF；会议列表身份匹配，会议正式 PDF 未取得，版本差异未知 | **重要问题**：主要结果、消融及混合搜索对照表未给出重复运行次数、标准差或置信区间；论文也未详细交代基线调参预算、具体时间切分边界、归一化和缺失值处理。实际波动和是否存在评测偏差未知。；
**重要问题**：附录 B.5 称敏感性分析默认 M=2；表 11 的 METR-LA M=1、PEMS04 M=3 指标分别与主结果相应数值相同，图 10 的 PEMS04 示例含三个空间 DAG。论文未明确主结果分块数的选择依据及选择所用划分。；
**重要问题**：图 5 给出每轮搜索耗时，未给出各方法搜索轮数及完整搜索、训练与预训练成本；AutoCTS+ 未纳入效率比较。 | [评审](论文/P131/评审.md) · [评分数据](论文/P131/评审.json) · [复核](论文/P131/复核.md) |",
"raw_cells": [
"27",
"[P131 AutoSTF: Decoupled Neural Architecture Search for Cost-Effective Automated Spatio-Temporal Forecasting](https://doi.org/10.1145/3690624.3709323)",
"55.0",
"100%",
"50%",
"47.7–60.5",
"8",
"arXiv:2409.16586v2，本地 PDF；会议列表身份匹配，会议正式 PDF 未取得，版本差异未知",
"**重要问题**：主要结果、消融及混合搜索对照表未给出重复运行次数、标准差或置信区间；论文也未详细交代基线调参预算、具体时间切分边界、归一化和缺失值处理。实际波动和是否存在评测偏差未知。；
**重要问题**：附录 B.5 称敏感性分析默认 M=2；表 11 的 METR-LA M=1、PEMS04 M=3 指标分别与主结果相应数值相同，图 10 的 PEMS04 示例含三个空间 DAG。论文未明确主结果分块数的选择依据及选择所用划分。；
**重要问题**：图 5 给出每轮搜索耗时，未给出各方法搜索轮数及完整搜索、训练与预训练成本；AutoCTS+ 未纳入效率比较。",
"[评审](论文/P131/评审.md) · [评分数据](论文/P131/评审.json) · [复核](论文/P131/复核.md)"
],
"normalized_title": "autostfdecoupledneuralarchitecturesearchforcosteffectiveautomatedspatiotemporalforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709323",
"record_type": "legacy_scored",
"legacy_rank": 27,
"legacy_score": "55.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "47.7–60.5",
"citation_verbatim": "8",
"legacy_review_version": "arXiv:2409.16586v2，本地 PDF；会议列表身份匹配，会议正式 PDF 未取得，版本差异未知",
"legacy_risks_verbatim": "**重要问题**：主要结果、消融及混合搜索对照表未给出重复运行次数、标准差或置信区间；论文也未详细交代基线调参预算、具体时间切分边界、归一化和缺失值处理。实际波动和是否存在评测偏差未知。；
**重要问题**：附录 B.5 称敏感性分析默认 M=2；表 11 的 METR-LA M=1、PEMS04 M=3 指标分别与主结果相应数值相同，图 10 的 PEMS04 示例含三个空间 DAG。论文未明确主结果分块数的选择依据及选择所用划分。；
**重要问题**：图 5 给出每轮搜索耗时，未给出各方法搜索轮数及完整搜索、训练与预训练成本；AutoCTS+ 未纳入效率比较。",
"legacy_material_links_verbatim": "[评审](论文/P131/评审.md) · [评分数据](论文/P131/评审.json) · [复核](论文/P131/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-033",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 44,
"source_number": 33,
"title": "Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/28707",
"raw_line": "| 28 | [P033 Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/28707) | 54.7 | 100% | 85% | 42.7–68.7 | 140 | AAAI 2024 会议正式版，2024-03-24；更早公开全文的时间未核实 | **重要问题**：表 2 PEMS07 的 STSGCN RMSE 为 29.03，小于 STPGNN 的 33.38，而正文称仅 PEMS-BAY MAPE 例外。STSGCN 2020 原文表 2 则写 39.03±0.27，可能是转录差异，仍需核对条件。；
**重要问题**：AAAI 2023 PDFormer 对传播时延与远程依赖建模，与本文问题接近，但未出现在本文比较列表。；
**重要问题**：学习到的相似度和少量地理可视化尚缺独立的交通聚散真值验证，关键节点重要性的机制解释存在替代可能。 | [评审](论文/P033/评审.md) · [评分数据](论文/P033/评审.json) · 无单独复核文件 |",
"raw_cells": [
"28",
"[P033 Spatio-Temporal Pivotal Graph Neural Networks for Traffic Flow Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/28707)",
"54.7",
"100%",
"85%",
"42.7–68.7",
"140",
"AAAI 2024 会议正式版，2024-03-24；更早公开全文的时间未核实",
"**重要问题**：表 2 PEMS07 的 STSGCN RMSE 为 29.03，小于 STPGNN 的 33.38，而正文称仅 PEMS-BAY MAPE 例外。STSGCN 2020 原文表 2 则写 39.03±0.27，可能是转录差异，仍需核对条件。；
**重要问题**：AAAI 2023 PDFormer 对传播时延与远程依赖建模，与本文问题接近，但未出现在本文比较列表。；
**重要问题**：学习到的相似度和少量地理可视化尚缺独立的交通聚散真值验证，关键节点重要性的机制解释存在替代可能。",
"[评审](论文/P033/评审.md) · [评分数据](论文/P033/评审.json) · 无单独复核文件"
],
"normalized_title": "spatiotemporalpivotalgraphneuralnetworksfortrafficflowforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/28707",
"record_type": "legacy_scored",
"legacy_rank": 28,
"legacy_score": "54.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "85%",
"legacy_range": "42.7–68.7",
"citation_verbatim": "140",
"legacy_review_version": "AAAI 2024 会议正式版，2024-03-24；更早公开全文的时间未核实",
"legacy_risks_verbatim": "**重要问题**：表 2 PEMS07 的 STSGCN RMSE 为 29.03，小于 STPGNN 的 33.38，而正文称仅 PEMS-BAY MAPE 例外。STSGCN 2020 原文表 2 则写 39.03±0.27，可能是转录差异，仍需核对条件。；
**重要问题**：AAAI 2023 PDFormer 对传播时延与远程依赖建模，与本文问题接近，但未出现在本文比较列表。；
**重要问题**：学习到的相似度和少量地理可视化尚缺独立的交通聚散真值验证，关键节点重要性的机制解释存在替代可能。",
"legacy_material_links_verbatim": "[评审](论文/P033/评审.md) · [评分数据](论文/P033/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-025",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 45,
"source_number": 25,
"title": "Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks",
"source_url": "https://doi.org/10.1145/3637528.3671665",
"raw_line": "| 29 | [P025 Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks](https://doi.org/10.1145/3637528.3671665) | 54.5 | 100% | 50% | 49.2–60.7 | 32 | arXiv:2308.16818v3，KDD 2024 排版；本地 PDF | **重要问题**：表 2 的株洲 C-MAPE：mTAND 3.93%，ASeer 4.10%，与摘要/结论的宽泛优势表述存在例外；§4.2 承认此例外。；
**重要问题**：F-MAE/F-RMSE 采用真实周期长度，所有指标掩码缺失真值；两城全数据缺失期比例为 44.2%/27.2%。；
**重要问题**：论文提供单组表格和图，没有报告多次训练的随机种子、波动或置信区间；两个城市 F-AAE 优势分别约 1.62% 和 2.62%。；
**重要问题**：RAINDROP 是较早的不规则序列异步图方法，但未出现在表 2 和附录 A.2 基线中；部分基线采用逐步解码，其他采用固定长度分块，论文未完整列出调参预算。 | [评审](论文/P025/评审.md) · [评分数据](论文/P025/评审.json) · 无单独复核文件 |",
"raw_cells": [
"29",
"[P025 Irregular Traffic Time Series Forecasting Based on Asynchronous Spatio-Temporal Graph Convolutional Networks](https://doi.org/10.1145/3637528.3671665)",
"54.5",
"100%",
"50%",
"49.2–60.7",
"32",
"arXiv:2308.16818v3，KDD 2024 排版；本地 PDF",
"**重要问题**：表 2 的株洲 C-MAPE：mTAND 3.93%，ASeer 4.10%，与摘要/结论的宽泛优势表述存在例外；§4.2 承认此例外。；
**重要问题**：F-MAE/F-RMSE 采用真实周期长度，所有指标掩码缺失真值；两城全数据缺失期比例为 44.2%/27.2%。；
**重要问题**：论文提供单组表格和图，没有报告多次训练的随机种子、波动或置信区间；两个城市 F-AAE 优势分别约 1.62% 和 2.62%。；
**重要问题**：RAINDROP 是较早的不规则序列异步图方法，但未出现在表 2 和附录 A.2 基线中；部分基线采用逐步解码，其他采用固定长度分块，论文未完整列出调参预算。",
"[评审](论文/P025/评审.md) · [评分数据](论文/P025/评审.json) · 无单独复核文件"
],
"normalized_title": "irregulartraffictimeseriesforecastingbasedonasynchronousspatiotemporalgraphconvolutionalnetworks",
"normalized_url": "doi.org/10.1145/3637528.3671665",
"record_type": "legacy_scored",
"legacy_rank": 29,
"legacy_score": "54.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "49.2–60.7",
"citation_verbatim": "32",
"legacy_review_version": "arXiv:2308.16818v3，KDD 2024 排版；本地 PDF",
"legacy_risks_verbatim": "**重要问题**：表 2 的株洲 C-MAPE：mTAND 3.93%，ASeer 4.10%，与摘要/结论的宽泛优势表述存在例外；§4.2 承认此例外。；
**重要问题**：F-MAE/F-RMSE 采用真实周期长度，所有指标掩码缺失真值；两城全数据缺失期比例为 44.2%/27.2%。；
**重要问题**：论文提供单组表格和图，没有报告多次训练的随机种子、波动或置信区间；两个城市 F-AAE 优势分别约 1.62% 和 2.62%。；
**重要问题**：RAINDROP 是较早的不规则序列异步图方法，但未出现在表 2 和附录 A.2 基线中；部分基线采用逐步解码，其他采用固定长度分块，论文未完整列出调参预算。",
"legacy_material_links_verbatim": "[评审](论文/P025/评审.md) · [评分数据](论文/P025/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-096",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 46,
"source_number": 96,
"title": "ST-MTM: Masked Time Series Modeling with Seasonal-Trend Decomposition for Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709254",
"raw_line": "| 30 | [P096 ST-MTM: Masked Time Series Modeling with Seasonal-Trend Decomposition for Time Series Forecasting](https://doi.org/10.1145/3690624.3709254) | 54.5 | 100% | 40% | 49.2–59.0 | 8 | arXiv:2507.00013v1；ACM会议终稿未取得，版本差异未知 | **重要问题**：表1称所有结果平均四个预测长度；附录A.2同时说明Ti-MAE缺Electricity的336、720和PEMS08的720。相应表格数字的汇总口径未交代，与其他模型四长度平均的可比性待核实。；
**重要问题**：图8右图10%缺失时PatchTST曲线低于ST-MTM，与§4.4.3称始终最低不一致。左图称ETTh1，但336输入处六模型点与表1–2的ETTh2值接近；右图称ETTh2，但0%缺失处六模型点与ETTh1值接近。标签互换是可能解释，数据来源和实际运行条件待核实。；
**重要问题**：图6中ST与随机或几何方法同时变更掩码类型、聚合以及部分遮蔽比例；图5仅移除一个损失项，缺少同架构两项均移除的对照。不同遮蔽比例、聚合方式或对比训练均可能解释部分差异。；
**重要问题**：每个预测长度重复五次，但已读论文主表及附录未报告逐次结果、标准差或置信区间；数据集切分边界、标准化拟合集和超参数选择细节也未在论文中明确给出。表3标题的196与§4.2的192不一致。；
**重要问题**：附录A.2称因显存和时间限制，在Electricity与PEMS08上缩小SimMTM模型并将其预训练设为10轮；§4.1.3称ST-MTM在这两个数据集也预训练10轮。两者模型容量不同，配置调整对相对性能的影响未知。 | [评审](论文/P096/评审.md) · [评分数据](论文/P096/评审.json) · [复核](论文/P096/复核.md) |",
"raw_cells": [
"30",
"[P096 ST-MTM: Masked Time Series Modeling with Seasonal-Trend Decomposition for Time Series Forecasting](https://doi.org/10.1145/3690624.3709254)",
"54.5",
"100%",
"40%",
"49.2–59.0",
"8",
"arXiv:2507.00013v1；ACM会议终稿未取得，版本差异未知",
"**重要问题**：表1称所有结果平均四个预测长度；附录A.2同时说明Ti-MAE缺Electricity的336、720和PEMS08的720。相应表格数字的汇总口径未交代，与其他模型四长度平均的可比性待核实。；
**重要问题**：图8右图10%缺失时PatchTST曲线低于ST-MTM，与§4.4.3称始终最低不一致。左图称ETTh1，但336输入处六模型点与表1–2的ETTh2值接近；右图称ETTh2，但0%缺失处六模型点与ETTh1值接近。标签互换是可能解释，数据来源和实际运行条件待核实。；
**重要问题**：图6中ST与随机或几何方法同时变更掩码类型、聚合以及部分遮蔽比例；图5仅移除一个损失项，缺少同架构两项均移除的对照。不同遮蔽比例、聚合方式或对比训练均可能解释部分差异。；
**重要问题**：每个预测长度重复五次，但已读论文主表及附录未报告逐次结果、标准差或置信区间；数据集切分边界、标准化拟合集和超参数选择细节也未在论文中明确给出。表3标题的196与§4.2的192不一致。；
**重要问题**：附录A.2称因显存和时间限制，在Electricity与PEMS08上缩小SimMTM模型并将其预训练设为10轮；§4.1.3称ST-MTM在这两个数据集也预训练10轮。两者模型容量不同，配置调整对相对性能的影响未知。",
"[评审](论文/P096/评审.md) · [评分数据](论文/P096/评审.json) · [复核](论文/P096/复核.md)"
],
"normalized_title": "stmtmmaskedtimeseriesmodelingwithseasonaltrenddecompositionfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709254",
"record_type": "legacy_scored",
"legacy_rank": 30,
"legacy_score": "54.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "49.2–59.0",
"citation_verbatim": "8",
"legacy_review_version": "arXiv:2507.00013v1；ACM会议终稿未取得，版本差异未知",
"legacy_risks_verbatim": "**重要问题**：表1称所有结果平均四个预测长度；附录A.2同时说明Ti-MAE缺Electricity的336、720和PEMS08的720。相应表格数字的汇总口径未交代，与其他模型四长度平均的可比性待核实。；
**重要问题**：图8右图10%缺失时PatchTST曲线低于ST-MTM，与§4.4.3称始终最低不一致。左图称ETTh1，但336输入处六模型点与表1–2的ETTh2值接近；右图称ETTh2，但0%缺失处六模型点与ETTh1值接近。标签互换是可能解释，数据来源和实际运行条件待核实。；
**重要问题**：图6中ST与随机或几何方法同时变更掩码类型、聚合以及部分遮蔽比例；图5仅移除一个损失项，缺少同架构两项均移除的对照。不同遮蔽比例、聚合方式或对比训练均可能解释部分差异。；
**重要问题**：每个预测长度重复五次，但已读论文主表及附录未报告逐次结果、标准差或置信区间；数据集切分边界、标准化拟合集和超参数选择细节也未在论文中明确给出。表3标题的196与§4.2的192不一致。；
**重要问题**：附录A.2称因显存和时间限制，在Electricity与PEMS08上缩小SimMTM模型并将其预训练设为10轮；§4.1.3称ST-MTM在这两个数据集也预训练10轮。两者模型容量不同，配置调整对相对性能的影响未知。",
"legacy_material_links_verbatim": "[评审](论文/P096/评审.md) · [评分数据](论文/P096/评审.json) · [复核](论文/P096/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-111",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 47,
"source_number": 111,
"title": "Multi-period Learning for Financial Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709422",
"raw_line": "| 31 | [P111 Multi-period Learning for Financial Time Series Forecasting](https://doi.org/10.1145/3690624.3709422) | 54.5 | 100% | 45% | 50.0–59.7 | 10 | arXiv:2511.08622v2（2026-01-31）；本地PDF，会议题名与作者已核对 | **重要问题**：按式(2)、(6)、α=2、目标64块，n=5、10、30、60时步长为0；n=120和150时分别得到120和75块。第4.1.4节同时列出这些短期长度及每周期64块。实际是否使用另一规则未知。；
**重要问题**：论文未逐一给出主表各基线的完整窗口和调参预算；图7的基金与天气均用MLF五窗口128、384、512、768、1024，分别与短期通用窗口5、10、30、60、120、150及长期通用窗口128、256、512、768、1024、2048不同。2816为五窗口长度之和且含重叠。；
**重要问题**：本地PDF正文及附录未交代基金表中±的定义和重复次数；部署表给出五周WMAPE和GMV改善，未交代随机分组、同期对照及波动估计。；
**重要问题**：IRF/LWI/MA消融混合结构容量、优化和尺度；图8热图之间色标不同，表1按真实值事后划分最优窗口；LWI权重与各窗口准确度的关系未单独检验。；
**重要问题**：论文介绍2015–2023基金样本；当前README称因保密仅公开2021–2023年子集。本次网页工具未打开论文及README的数据文件夹。 | [评审](论文/P111/评审.md) · [评分数据](论文/P111/评审.json) · [复核](论文/P111/复核.md) |",
"raw_cells": [
"31",
"[P111 Multi-period Learning for Financial Time Series Forecasting](https://doi.org/10.1145/3690624.3709422)",
"54.5",
"100%",
"45%",
"50.0–59.7",
"10",
"arXiv:2511.08622v2（2026-01-31）；本地PDF，会议题名与作者已核对",
"**重要问题**：按式(2)、(6)、α=2、目标64块，n=5、10、30、60时步长为0；n=120和150时分别得到120和75块。第4.1.4节同时列出这些短期长度及每周期64块。实际是否使用另一规则未知。；
**重要问题**：论文未逐一给出主表各基线的完整窗口和调参预算；图7的基金与天气均用MLF五窗口128、384、512、768、1024，分别与短期通用窗口5、10、30、60、120、150及长期通用窗口128、256、512、768、1024、2048不同。2816为五窗口长度之和且含重叠。；
**重要问题**：本地PDF正文及附录未交代基金表中±的定义和重复次数；部署表给出五周WMAPE和GMV改善，未交代随机分组、同期对照及波动估计。；
**重要问题**：IRF/LWI/MA消融混合结构容量、优化和尺度；图8热图之间色标不同，表1按真实值事后划分最优窗口；LWI权重与各窗口准确度的关系未单独检验。；
**重要问题**：论文介绍2015–2023基金样本；当前README称因保密仅公开2021–2023年子集。本次网页工具未打开论文及README的数据文件夹。",
"[评审](论文/P111/评审.md) · [评分数据](论文/P111/评审.json) · [复核](论文/P111/复核.md)"
],
"normalized_title": "multiperiodlearningforfinancialtimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709422",
"record_type": "legacy_scored",
"legacy_rank": 31,
"legacy_score": "54.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "50.0–59.7",
"citation_verbatim": "10",
"legacy_review_version": "arXiv:2511.08622v2（2026-01-31）；本地PDF，会议题名与作者已核对",
"legacy_risks_verbatim": "**重要问题**：按式(2)、(6)、α=2、目标64块，n=5、10、30、60时步长为0；n=120和150时分别得到120和75块。第4.1.4节同时列出这些短期长度及每周期64块。实际是否使用另一规则未知。；
**重要问题**：论文未逐一给出主表各基线的完整窗口和调参预算；图7的基金与天气均用MLF五窗口128、384、512、768、1024，分别与短期通用窗口5、10、30、60、120、150及长期通用窗口128、256、512、768、1024、2048不同。2816为五窗口长度之和且含重叠。；
**重要问题**：本地PDF正文及附录未交代基金表中±的定义和重复次数；部署表给出五周WMAPE和GMV改善，未交代随机分组、同期对照及波动估计。；
**重要问题**：IRF/LWI/MA消融混合结构容量、优化和尺度；图8热图之间色标不同，表1按真实值事后划分最优窗口；LWI权重与各窗口准确度的关系未单独检验。；
**重要问题**：论文介绍2015–2023基金样本；当前README称因保密仅公开2021–2023年子集。本次网页工具未打开论文及README的数据文件夹。",
"legacy_material_links_verbatim": "[评审](论文/P111/评审.md) · [评分数据](论文/P111/评审.json) · [复核](论文/P111/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-026",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 48,
"source_number": 26,
"title": "AutoXPCR: Automated Multi-Objective Model Selection for Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3637528.3672057",
"raw_line": "| 32 | [P026 AutoXPCR: Automated Multi-Objective Model Selection for Time Series Forecasting](https://doi.org/10.1145/3637528.3672057) | 54.2 | 100% | 65% | 45.2–62.5 | 12 | arXiv:2312.13038v1，2023-12-20，12 页预印本；评分仅针对该版本 | **重要问题**：本地 arXiv v1 摘要为 114 个数据集版本、90% 相对质量；ACM 正式版摘要为 108 个数据集、85% 经验最佳质量。两版结果的具体口径和修改原因尚未核实。；
**重要问题**：论文报告按原始数据集分组五折，但没有完整交代时间切分、元学习器选择是否嵌套以及前 k 模型的选择与最终测试数据如何分离。；
**重要问题**：预印本把 AutoForecast 描述为仅重视 MASE 的情形；其原论文含通用与时间序列两类元学习器，预印本没有交代所比较实现是否保留完整结构；AutoKeras 也没有可比综合分。；
**重要问题**：图 2 衡量测试前 k 推荐模型的相对能耗；预印本另称部署 1254 个 DNN 约耗两周，但没有给出每个新任务分摊离线建库成本后的能耗比较。；
**重要问题**：114 个数据集版本来自 19 个原始数据集；图 4 展示元学习器的属性与特征重要度，未报告面向用户的解释效果或预测 DNN 内部机制验证。 | [评审](论文/P026/评审.md) · [评分数据](论文/P026/评审.json) · 无单独复核文件 |",
"raw_cells": [
"32",
"[P026 AutoXPCR: Automated Multi-Objective Model Selection for Time Series Forecasting](https://doi.org/10.1145/3637528.3672057)",
"54.2",
"100%",
"65%",
"45.2–62.5",
"12",
"arXiv:2312.13038v1，2023-12-20，12 页预印本；评分仅针对该版本",
"**重要问题**：本地 arXiv v1 摘要为 114 个数据集版本、90% 相对质量；ACM 正式版摘要为 108 个数据集、85% 经验最佳质量。两版结果的具体口径和修改原因尚未核实。；
**重要问题**：论文报告按原始数据集分组五折，但没有完整交代时间切分、元学习器选择是否嵌套以及前 k 模型的选择与最终测试数据如何分离。；
**重要问题**：预印本把 AutoForecast 描述为仅重视 MASE 的情形；其原论文含通用与时间序列两类元学习器，预印本没有交代所比较实现是否保留完整结构；AutoKeras 也没有可比综合分。；
**重要问题**：图 2 衡量测试前 k 推荐模型的相对能耗；预印本另称部署 1254 个 DNN 约耗两周，但没有给出每个新任务分摊离线建库成本后的能耗比较。；
**重要问题**：114 个数据集版本来自 19 个原始数据集；图 4 展示元学习器的属性与特征重要度，未报告面向用户的解释效果或预测 DNN 内部机制验证。",
"[评审](论文/P026/评审.md) · [评分数据](论文/P026/评审.json) · 无单独复核文件"
],
"normalized_title": "autoxpcrautomatedmultiobjectivemodelselectionfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3637528.3672057",
"record_type": "legacy_scored",
"legacy_rank": 32,
"legacy_score": "54.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "65%",
"legacy_range": "45.2–62.5",
"citation_verbatim": "12",
"legacy_review_version": "arXiv:2312.13038v1，2023-12-20，12 页预印本；评分仅针对该版本",
"legacy_risks_verbatim": "**重要问题**：本地 arXiv v1 摘要为 114 个数据集版本、90% 相对质量；ACM 正式版摘要为 108 个数据集、85% 经验最佳质量。两版结果的具体口径和修改原因尚未核实。；
**重要问题**：论文报告按原始数据集分组五折，但没有完整交代时间切分、元学习器选择是否嵌套以及前 k 模型的选择与最终测试数据如何分离。；
**重要问题**：预印本把 AutoForecast 描述为仅重视 MASE 的情形；其原论文含通用与时间序列两类元学习器，预印本没有交代所比较实现是否保留完整结构；AutoKeras 也没有可比综合分。；
**重要问题**：图 2 衡量测试前 k 推荐模型的相对能耗；预印本另称部署 1254 个 DNN 约耗两周，但没有给出每个新任务分摊离线建库成本后的能耗比较。；
**重要问题**：114 个数据集版本来自 19 个原始数据集；图 4 展示元学习器的属性与特征重要度，未报告面向用户的解释效果或预测 DNN 内部机制验证。",
"legacy_material_links_verbatim": "[评审](论文/P026/评审.md) · [评分数据](论文/P026/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-004",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 49,
"source_number": 4,
"title": "Multi-Patch Prediction: Adapting Language Models for Time Series Representation Learning",
"source_url": "https://proceedings.mlr.press/v235/bian24a.html",
"raw_line": "| 33 | [P004 Multi-Patch Prediction: Adapting Language Models for Time Series Representation Learning](https://proceedings.mlr.press/v235/bian24a.html) | 54.0 | 100% | 30% | 50.2–58.2 | 21 | ICML 2024 PMLR 235 会议论文 PDF；PDF 元数据创建于 2024-05-26，确切发布日未核实 | **重要问题**：§4.2 最近历史锚点写为 x[t+L-H:t+L-1]。当 §5.1–5.2 的 L=336、H=720 时，公式所指区间从 t-384 开始，比回看窗口多 384 个较早时间点；论文未说明此设置下是否改为截断、重复或其他处理。；
**重要问题**：表 5 的 PatchTST Masking+SD 在 ETTh1 的五个长度上 MSE 均小于 Masking+Sta2；表 15 的 ETTh1-96 序列头 0.375、逐块头 0.498。表 6 ETTm2 自身消融中序列头 0.284、逐块头 0.294。；
**重要问题**：表 9 的 ETTm2-96 默认 MSE 为 0.171，表 13 的默认列为 0.212；表 3 对少样本 ETTh2 的 DLinear/PatchTST 分别写 0.694/0.827，表 10 的相应逐长度行均值约为 0.828/0.439。表 3 称四个长度取均值，表 10 的 ETTh1/ETTh2 在 720 长度无结果。；
**重要问题**：长期预测在多个数据集的训练集上进行共享持续预训练；多数基线按各预测长度单独训练。少样本阶段使用其他 ETT 组进行预训练，目标数据集只用 5% 训练时间点。 | [评审](论文/P004/评审.md) · [评分数据](论文/P004/评审.json) · 无单独复核文件 |",
"raw_cells": [
"33",
"[P004 Multi-Patch Prediction: Adapting Language Models for Time Series Representation Learning](https://proceedings.mlr.press/v235/bian24a.html)",
"54.0",
"100%",
"30%",
"50.2–58.2",
"21",
"ICML 2024 PMLR 235 会议论文 PDF；PDF 元数据创建于 2024-05-26，确切发布日未核实",
"**重要问题**：§4.2 最近历史锚点写为 x[t+L-H:t+L-1]。当 §5.1–5.2 的 L=336、H=720 时，公式所指区间从 t-384 开始，比回看窗口多 384 个较早时间点；论文未说明此设置下是否改为截断、重复或其他处理。；
**重要问题**：表 5 的 PatchTST Masking+SD 在 ETTh1 的五个长度上 MSE 均小于 Masking+Sta2；表 15 的 ETTh1-96 序列头 0.375、逐块头 0.498。表 6 ETTm2 自身消融中序列头 0.284、逐块头 0.294。；
**重要问题**：表 9 的 ETTm2-96 默认 MSE 为 0.171，表 13 的默认列为 0.212；表 3 对少样本 ETTh2 的 DLinear/PatchTST 分别写 0.694/0.827，表 10 的相应逐长度行均值约为 0.828/0.439。表 3 称四个长度取均值，表 10 的 ETTh1/ETTh2 在 720 长度无结果。；
**重要问题**：长期预测在多个数据集的训练集上进行共享持续预训练；多数基线按各预测长度单独训练。少样本阶段使用其他 ETT 组进行预训练，目标数据集只用 5% 训练时间点。",
"[评审](论文/P004/评审.md) · [评分数据](论文/P004/评审.json) · 无单独复核文件"
],
"normalized_title": "multipatchpredictionadaptinglanguagemodelsfortimeseriesrepresentationlearning",
"normalized_url": "proceedings.mlr.press/v235/bian24a.html",
"record_type": "legacy_scored",
"legacy_rank": 33,
"legacy_score": "54.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "30%",
"legacy_range": "50.2–58.2",
"citation_verbatim": "21",
"legacy_review_version": "ICML 2024 PMLR 235 会议论文 PDF；PDF 元数据创建于 2024-05-26，确切发布日未核实",
"legacy_risks_verbatim": "**重要问题**：§4.2 最近历史锚点写为 x[t+L-H:t+L-1]。当 §5.1–5.2 的 L=336、H=720 时，公式所指区间从 t-384 开始，比回看窗口多 384 个较早时间点；论文未说明此设置下是否改为截断、重复或其他处理。；
**重要问题**：表 5 的 PatchTST Masking+SD 在 ETTh1 的五个长度上 MSE 均小于 Masking+Sta2；表 15 的 ETTh1-96 序列头 0.375、逐块头 0.498。表 6 ETTm2 自身消融中序列头 0.284、逐块头 0.294。；
**重要问题**：表 9 的 ETTm2-96 默认 MSE 为 0.171，表 13 的默认列为 0.212；表 3 对少样本 ETTh2 的 DLinear/PatchTST 分别写 0.694/0.827，表 10 的相应逐长度行均值约为 0.828/0.439。表 3 称四个长度取均值，表 10 的 ETTh1/ETTh2 在 720 长度无结果。；
**重要问题**：长期预测在多个数据集的训练集上进行共享持续预训练；多数基线按各预测长度单独训练。少样本阶段使用其他 ETT 组进行预训练，目标数据集只用 5% 训练时间点。",
"legacy_material_links_verbatim": "[评审](论文/P004/评审.md) · [评分数据](论文/P004/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-095",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 50,
"source_number": 95,
"title": "Unlocking the Power of LSTM for Long Term Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33303",
"raw_line": "| 34 | [P095 Unlocking the Power of LSTM for Long Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33303) | 53.7 | 100% | 35% | 49.2–58.2 | 106 | AAAI 2025 会议论文终稿；作者仓库附录 2024-12-19 版本 | **重要问题**：正文声称首次将 CI 应用于 RNN 预测；SegRNN 2023 v1 已采用 CI、片段迭代和 GRU。该冲突影响宽泛首创表述，特定 sLSTM 组合仍需单独评价。；
**重要问题**：特定随机过程、噪声和遗忘门条件下的结论被扩展为较宽泛的短记忆论述；分块对长期依赖的解决未由所列定理直接建立。 | [评审](论文/P095/评审.md) · [评分数据](论文/P095/评审.json) · 无单独复核文件 |",
"raw_cells": [
"34",
"[P095 Unlocking the Power of LSTM for Long Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33303)",
"53.7",
"100%",
"35%",
"49.2–58.2",
"106",
"AAAI 2025 会议论文终稿；作者仓库附录 2024-12-19 版本",
"**重要问题**：正文声称首次将 CI 应用于 RNN 预测；SegRNN 2023 v1 已采用 CI、片段迭代和 GRU。该冲突影响宽泛首创表述，特定 sLSTM 组合仍需单独评价。；
**重要问题**：特定随机过程、噪声和遗忘门条件下的结论被扩展为较宽泛的短记忆论述；分块对长期依赖的解决未由所列定理直接建立。",
"[评审](论文/P095/评审.md) · [评分数据](论文/P095/评审.json) · 无单独复核文件"
],
"normalized_title": "unlockingthepoweroflstmforlongtermtimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33303",
"record_type": "legacy_scored",
"legacy_rank": 34,
"legacy_score": "53.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "49.2–58.2",
"citation_verbatim": "106",
"legacy_review_version": "AAAI 2025 会议论文终稿；作者仓库附录 2024-12-19 版本",
"legacy_risks_verbatim": "**重要问题**：正文声称首次将 CI 应用于 RNN 预测；SegRNN 2023 v1 已采用 CI、片段迭代和 GRU。该冲突影响宽泛首创表述，特定 sLSTM 组合仍需单独评价。；
**重要问题**：特定随机过程、噪声和遗忘门条件下的结论被扩展为较宽泛的短记忆论述；分块对长期依赖的解决未由所列定理直接建立。",
"legacy_material_links_verbatim": "[评审](论文/P095/评审.md) · [评分数据](论文/P095/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-133",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 51,
"source_number": 133,
"title": "Efficient Large-Scale Traffic Forecasting with Transformers: A Spatial Data Management Perspective",
"source_url": "https://doi.org/10.1145/3690624.3709177",
"raw_line": "| 35 | [P133 Efficient Large-Scale Traffic Forecasting with Transformers: A Spatial Data Management Perspective](https://doi.org/10.1145/3690624.3709177) | 53.7 | 100% | 50% | 48.0–58.7 | 60 | arXiv:2412.09972v2；本地 11 页 PDF，非已核对内容的 ACM 出版版 | **重要问题**：S01 第 5 页称 M=C×2^(⌊log N⌋−log C)≥N。以二叉树常用的 log₂ 解释，表 2 和 §5.1.4 的 SD 参数 N=716、C=2 给出 M=512<716；这与填充后至少保留 N 个位置的要求冲突。；
**重要问题**：S01 表 5 的 GLA 约 10× 来自约 41/4 的总小时数，单轮为 1483/295≈5.03；CA 的 4× 为 32/8 批量比且未报告显存峰值。GBA 改善行的每轮 1.8× 与 STWave 的 926/262≈3.53 等明细不一致。；
**重要问题**：论文给出按时间 6:2:2 划分，但未详细说明式 5 相似度使用的数据时间范围、缺失处理、跨边界窗口、超参数选择所用数据、MAPE 零值或掩码规则和随机重复；LargeST 原始数据保留缺失。；
**重要问题**：图 5–6 展示空间分区与注意力权重差异，表 4 展示消融误差；这些材料没有给出信息保真定理、因果解释验证或与全连接注意力的无损等价证明。；
**重要问题**：四组数据属于 LargeST 加州层级子集的同一年，且表 5 未比较同样适合大规模问题的 BigST 和轻量 STID；BigST 原论文含额外长历史预计算特征。 | [评审](论文/P133/评审.md) · [评分数据](论文/P133/评审.json) · 无单独复核文件 |",
"raw_cells": [
"35",
"[P133 Efficient Large-Scale Traffic Forecasting with Transformers: A Spatial Data Management Perspective](https://doi.org/10.1145/3690624.3709177)",
"53.7",
"100%",
"50%",
"48.0–58.7",
"60",
"arXiv:2412.09972v2；本地 11 页 PDF，非已核对内容的 ACM 出版版",
"**重要问题**：S01 第 5 页称 M=C×2^(⌊log N⌋−log C)≥N。以二叉树常用的 log₂ 解释，表 2 和 §5.1.4 的 SD 参数 N=716、C=2 给出 M=512<716；这与填充后至少保留 N 个位置的要求冲突。；
**重要问题**：S01 表 5 的 GLA 约 10× 来自约 41/4 的总小时数，单轮为 1483/295≈5.03；CA 的 4× 为 32/8 批量比且未报告显存峰值。GBA 改善行的每轮 1.8× 与 STWave 的 926/262≈3.53 等明细不一致。；
**重要问题**：论文给出按时间 6:2:2 划分，但未详细说明式 5 相似度使用的数据时间范围、缺失处理、跨边界窗口、超参数选择所用数据、MAPE 零值或掩码规则和随机重复；LargeST 原始数据保留缺失。；
**重要问题**：图 5–6 展示空间分区与注意力权重差异，表 4 展示消融误差；这些材料没有给出信息保真定理、因果解释验证或与全连接注意力的无损等价证明。；
**重要问题**：四组数据属于 LargeST 加州层级子集的同一年，且表 5 未比较同样适合大规模问题的 BigST 和轻量 STID；BigST 原论文含额外长历史预计算特征。",
"[评审](论文/P133/评审.md) · [评分数据](论文/P133/评审.json) · 无单独复核文件"
],
"normalized_title": "efficientlargescaletrafficforecastingwithtransformersaspatialdatamanagementperspective",
"normalized_url": "doi.org/10.1145/3690624.3709177",
"record_type": "legacy_scored",
"legacy_rank": 35,
"legacy_score": "53.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "48.0–58.7",
"citation_verbatim": "60",
"legacy_review_version": "arXiv:2412.09972v2；本地 11 页 PDF，非已核对内容的 ACM 出版版",
"legacy_risks_verbatim": "**重要问题**：S01 第 5 页称 M=C×2^(⌊log N⌋−log C)≥N。以二叉树常用的 log₂ 解释，表 2 和 §5.1.4 的 SD 参数 N=716、C=2 给出 M=512<716；这与填充后至少保留 N 个位置的要求冲突。；
**重要问题**：S01 表 5 的 GLA 约 10× 来自约 41/4 的总小时数，单轮为 1483/295≈5.03；CA 的 4× 为 32/8 批量比且未报告显存峰值。GBA 改善行的每轮 1.8× 与 STWave 的 926/262≈3.53 等明细不一致。；
**重要问题**：论文给出按时间 6:2:2 划分，但未详细说明式 5 相似度使用的数据时间范围、缺失处理、跨边界窗口、超参数选择所用数据、MAPE 零值或掩码规则和随机重复；LargeST 原始数据保留缺失。；
**重要问题**：图 5–6 展示空间分区与注意力权重差异，表 4 展示消融误差；这些材料没有给出信息保真定理、因果解释验证或与全连接注意力的无损等价证明。；
**重要问题**：四组数据属于 LargeST 加州层级子集的同一年，且表 5 未比较同样适合大规模问题的 BigST 和轻量 STID；BigST 原论文含额外长历史预计算特征。",
"legacy_material_links_verbatim": "[评审](论文/P133/评审.md) · [评分数据](论文/P133/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-039",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 52,
"source_number": 39,
"title": "T-Rep: Representation Learning for Time Series using Time-Embeddings",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/84b946c4c4162b29464fc7aa57e480e1-Abstract-Conference.html",
"raw_line": "| 36 | [P039 T-Rep: Representation Learning for Time Series using Time-Embeddings](https://proceedings.iclr.cc/paper_files/paper/2024/hash/84b946c4c4162b29464fc7aa57e480e1-Abstract-Conference.html) | 53.4 | 95% | 37% | 47.2–59.7 | 7 | ICLR 2024 官方会议 PDF；21 页，下载 PDF 与会议页面题名、作者一致 | **重要问题**：Informer 来源叙述冲突：表 3 注称全部结果为作者复现；A.3 又称 ETTh1、ETTh2、ETTm1 采用 Informer 原文结果。T-Rep 描述预测设置无日历协变量，Informer 原文说明解码器使用目标时间戳；表内各实验的输入窗口和协变量是否完全一致仍待核实。；
**重要问题**：UEA 主表对 T-Rep 使用保留时间维后展平的 TimeDim 分类流程，对 TS2Vec 则报告实例级全局表示；附表同时报告 TS2Vec TimeDim。每时间步 128 维与实例向量 320 维的比较不足以确定最终分类器输入维数和计算成本优势。；
**重要问题**：ETT 描述了过去窗口和 12/4/4 月划分，但未清楚交代产生 z_t 时如何防止使用 t 之后的观测、归一化参数的拟合范围；Sepsis 段落未明确按患者划分预训练、分类器训练与测试。双向预训练本身与推断时的信息边界是不同事项。；
**重要问题**：式 (4) 中当全部时间嵌入相同且 G1 恒为零时，JSD 预测损失为零；同时时间嵌入直接进入编码器，基于时间索引完成任务是一种待检验的替代解释。表 4 消融同时重分配其余任务权重。；
**重要问题**：分类表 5 的 MiniRocket 汇总 0.707 与 30 个显示值均值 0.70523 不一致，与 T-Rep 的相对顺序也随使用汇总行或显示值复算而改变；预测表 6 的 TS2Vec MAE 汇总为 0.712，20 个显示值的均值为 0.7144。；
**重要问题**：缺失鲁棒性的量化曲线只展示一个分类数据集，论文未充分报告缺失掩码生成、训练与测试时掩码配置；热图仅为选取维度后的示例。 | [评审](论文/P039/评审.md) · [评分数据](论文/P039/评审.json) · [复核](论文/P039/复核.md) |",
"raw_cells": [
"36",
"[P039 T-Rep: Representation Learning for Time Series using Time-Embeddings](https://proceedings.iclr.cc/paper_files/paper/2024/hash/84b946c4c4162b29464fc7aa57e480e1-Abstract-Conference.html)",
"53.4",
"95%",
"37%",
"47.2–59.7",
"7",
"ICLR 2024 官方会议 PDF；21 页，下载 PDF 与会议页面题名、作者一致",
"**重要问题**：Informer 来源叙述冲突：表 3 注称全部结果为作者复现；A.3 又称 ETTh1、ETTh2、ETTm1 采用 Informer 原文结果。T-Rep 描述预测设置无日历协变量，Informer 原文说明解码器使用目标时间戳；表内各实验的输入窗口和协变量是否完全一致仍待核实。；
**重要问题**：UEA 主表对 T-Rep 使用保留时间维后展平的 TimeDim 分类流程，对 TS2Vec 则报告实例级全局表示；附表同时报告 TS2Vec TimeDim。每时间步 128 维与实例向量 320 维的比较不足以确定最终分类器输入维数和计算成本优势。；
**重要问题**：ETT 描述了过去窗口和 12/4/4 月划分，但未清楚交代产生 z_t 时如何防止使用 t 之后的观测、归一化参数的拟合范围；Sepsis 段落未明确按患者划分预训练、分类器训练与测试。双向预训练本身与推断时的信息边界是不同事项。；
**重要问题**：式 (4) 中当全部时间嵌入相同且 G1 恒为零时，JSD 预测损失为零；同时时间嵌入直接进入编码器，基于时间索引完成任务是一种待检验的替代解释。表 4 消融同时重分配其余任务权重。；
**重要问题**：分类表 5 的 MiniRocket 汇总 0.707 与 30 个显示值均值 0.70523 不一致，与 T-Rep 的相对顺序也随使用汇总行或显示值复算而改变；预测表 6 的 TS2Vec MAE 汇总为 0.712，20 个显示值的均值为 0.7144。；
**重要问题**：缺失鲁棒性的量化曲线只展示一个分类数据集，论文未充分报告缺失掩码生成、训练与测试时掩码配置；热图仅为选取维度后的示例。",
"[评审](论文/P039/评审.md) · [评分数据](论文/P039/评审.json) · [复核](论文/P039/复核.md)"
],
"normalized_title": "treprepresentationlearningfortimeseriesusingtimeembeddings",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/84b946c4c4162b29464fc7aa57e480e1-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 36,
"legacy_score": "53.4",
"legacy_coverage": "95%",
"legacy_tentative_share": "37%",
"legacy_range": "47.2–59.7",
"citation_verbatim": "7",
"legacy_review_version": "ICLR 2024 官方会议 PDF；21 页，下载 PDF 与会议页面题名、作者一致",
"legacy_risks_verbatim": "**重要问题**：Informer 来源叙述冲突：表 3 注称全部结果为作者复现；A.3 又称 ETTh1、ETTh2、ETTm1 采用 Informer 原文结果。T-Rep 描述预测设置无日历协变量，Informer 原文说明解码器使用目标时间戳；表内各实验的输入窗口和协变量是否完全一致仍待核实。；
**重要问题**：UEA 主表对 T-Rep 使用保留时间维后展平的 TimeDim 分类流程，对 TS2Vec 则报告实例级全局表示；附表同时报告 TS2Vec TimeDim。每时间步 128 维与实例向量 320 维的比较不足以确定最终分类器输入维数和计算成本优势。；
**重要问题**：ETT 描述了过去窗口和 12/4/4 月划分，但未清楚交代产生 z_t 时如何防止使用 t 之后的观测、归一化参数的拟合范围；Sepsis 段落未明确按患者划分预训练、分类器训练与测试。双向预训练本身与推断时的信息边界是不同事项。；
**重要问题**：式 (4) 中当全部时间嵌入相同且 G1 恒为零时，JSD 预测损失为零；同时时间嵌入直接进入编码器，基于时间索引完成任务是一种待检验的替代解释。表 4 消融同时重分配其余任务权重。；
**重要问题**：分类表 5 的 MiniRocket 汇总 0.707 与 30 个显示值均值 0.70523 不一致，与 T-Rep 的相对顺序也随使用汇总行或显示值复算而改变；预测表 6 的 TS2Vec MAE 汇总为 0.712，20 个显示值的均值为 0.7144。；
**重要问题**：缺失鲁棒性的量化曲线只展示一个分类数据集，论文未充分报告缺失掩码生成、训练与测试时掩码配置；热图仅为选取维度后的示例。",
"legacy_material_links_verbatim": "[评审](论文/P039/评审.md) · [评分数据](论文/P039/评审.json) · [复核](论文/P039/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-098",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 53,
"source_number": 98,
"title": "Hierarchical Classification Auxiliary Network for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34286",
"raw_line": "| 37 | [P098 Hierarchical Classification Auxiliary Network for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34286) | 53.2 | 100% | 40% | 49.2–57.5 | 11 | AAAI 2025 正式版；方法及实验附录另核对 arXiv:2405.18975v2 | **重要问题**：正文和 v2 附录 B 未清楚交代用于分位数边界的序列属于哪些划分，以及网格搜索的具体选择集和各骨干的完整调参预算；目前没有已确认的数据泄漏。；
**重要问题**：表 3、表 8 为整体误差消融，图 4–5 为可视化；所读材料缺少按边界距离分组的误差、直接特征熵或分类置信度校准对照。；
**重要问题**：表 1 有四组平均 MSE 恶化；v2 表 5 的 Weather/iTransformer 在 96 步 MSE 下降而 MAE 上升；表 11 的轻量骨干参数栏相对增幅较大，单位计量方式未解释。v2 附录 L 称没有特殊失败但未界定该措辞，同版表 1、5 中 Exchange/FITS 平均 MSE 为 0.360→0.426。 | [评审](论文/P098/评审.md) · [评分数据](论文/P098/评审.json) · [复核](论文/P098/复核.md) |",
"raw_cells": [
"37",
"[P098 Hierarchical Classification Auxiliary Network for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34286)",
"53.2",
"100%",
"40%",
"49.2–57.5",
"11",
"AAAI 2025 正式版；方法及实验附录另核对 arXiv:2405.18975v2",
"**重要问题**：正文和 v2 附录 B 未清楚交代用于分位数边界的序列属于哪些划分，以及网格搜索的具体选择集和各骨干的完整调参预算；目前没有已确认的数据泄漏。；
**重要问题**：表 3、表 8 为整体误差消融，图 4–5 为可视化；所读材料缺少按边界距离分组的误差、直接特征熵或分类置信度校准对照。；
**重要问题**：表 1 有四组平均 MSE 恶化；v2 表 5 的 Weather/iTransformer 在 96 步 MSE 下降而 MAE 上升；表 11 的轻量骨干参数栏相对增幅较大，单位计量方式未解释。v2 附录 L 称没有特殊失败但未界定该措辞，同版表 1、5 中 Exchange/FITS 平均 MSE 为 0.360→0.426。",
"[评审](论文/P098/评审.md) · [评分数据](论文/P098/评审.json) · [复核](论文/P098/复核.md)"
],
"normalized_title": "hierarchicalclassificationauxiliarynetworkfortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34286",
"record_type": "legacy_scored",
"legacy_rank": 37,
"legacy_score": "53.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "49.2–57.5",
"citation_verbatim": "11",
"legacy_review_version": "AAAI 2025 正式版；方法及实验附录另核对 arXiv:2405.18975v2",
"legacy_risks_verbatim": "**重要问题**：正文和 v2 附录 B 未清楚交代用于分位数边界的序列属于哪些划分，以及网格搜索的具体选择集和各骨干的完整调参预算；目前没有已确认的数据泄漏。；
**重要问题**：表 3、表 8 为整体误差消融，图 4–5 为可视化；所读材料缺少按边界距离分组的误差、直接特征熵或分类置信度校准对照。；
**重要问题**：表 1 有四组平均 MSE 恶化；v2 表 5 的 Weather/iTransformer 在 96 步 MSE 下降而 MAE 上升；表 11 的轻量骨干参数栏相对增幅较大，单位计量方式未解释。v2 附录 L 称没有特殊失败但未界定该措辞，同版表 1、5 中 Exchange/FITS 平均 MSE 为 0.360→0.426。",
"legacy_material_links_verbatim": "[评审](论文/P098/评审.md) · [评分数据](论文/P098/评审.json) · [复核](论文/P098/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-064",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 54,
"source_number": 64,
"title": "CSformer: Combining Channel Independence and Mixing for Robust Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/35406",
"raw_line": "| 38 | [P064 CSformer: Combining Channel Independence and Mixing for Robust Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35406) | 53.2 | 95% | 53% | 44.7–61.2 | 8 | AAAI-25 proceedings，39(20)，21090–21098；补充参考独立标识的 arXiv:2312.06220v2 | **重要问题**：会议摘要和结论提出先通道独立后混合；图3与方法显示每个块先通道注意力后时间注意力。表4列示C→S在ETTh1和Electricity上优于S→C、Weather打平；表7比较CI与CM，但未说明策略的具体操作、是否切换及其与块内顺序的关系。；
**重要问题**：同一2023年v1预印本表1将Weather列为21个特征，而数据文字写11个气象特征加1个目标；2024年v2附录A沿用后者。实际使用的Weather文件、变量及预测目标未得到独立核实。；
**重要问题**：已读会议正文与v2附录未见跨随机种子的波动或误差条，真实基准的逐模型调参预算、切分和预处理文件未完整列明；小差异的稳定性及比较条件仍有缺口。；
**重要问题**：v2附录C称前90%训练段加入不同标准差高斯噪声，后10%测试段无噪；合成数据依式(6)由10条正弦函数构造。图9中0.2–1.0噪声标准差各标点的CSformer MSE低于两条对照曲线；0.1处与PatchTST标记重叠，精确排序不清。 | [评审](论文/P064/评审.md) · [评分数据](论文/P064/评审.json) · [复核](论文/P064/复核.md) |",
"raw_cells": [
"38",
"[P064 CSformer: Combining Channel Independence and Mixing for Robust Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35406)",
"53.2",
"95%",
"53%",
"44.7–61.2",
"8",
"AAAI-25 proceedings，39(20)，21090–21098；补充参考独立标识的 arXiv:2312.06220v2",
"**重要问题**：会议摘要和结论提出先通道独立后混合；图3与方法显示每个块先通道注意力后时间注意力。表4列示C→S在ETTh1和Electricity上优于S→C、Weather打平；表7比较CI与CM，但未说明策略的具体操作、是否切换及其与块内顺序的关系。；
**重要问题**：同一2023年v1预印本表1将Weather列为21个特征，而数据文字写11个气象特征加1个目标；2024年v2附录A沿用后者。实际使用的Weather文件、变量及预测目标未得到独立核实。；
**重要问题**：已读会议正文与v2附录未见跨随机种子的波动或误差条，真实基准的逐模型调参预算、切分和预处理文件未完整列明；小差异的稳定性及比较条件仍有缺口。；
**重要问题**：v2附录C称前90%训练段加入不同标准差高斯噪声，后10%测试段无噪；合成数据依式(6)由10条正弦函数构造。图9中0.2–1.0噪声标准差各标点的CSformer MSE低于两条对照曲线；0.1处与PatchTST标记重叠，精确排序不清。",
"[评审](论文/P064/评审.md) · [评分数据](论文/P064/评审.json) · [复核](论文/P064/复核.md)"
],
"normalized_title": "csformercombiningchannelindependenceandmixingforrobustmultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/35406",
"record_type": "legacy_scored",
"legacy_rank": 38,
"legacy_score": "53.2",
"legacy_coverage": "95%",
"legacy_tentative_share": "53%",
"legacy_range": "44.7–61.2",
"citation_verbatim": "8",
"legacy_review_version": "AAAI-25 proceedings，39(20)，21090–21098；补充参考独立标识的 arXiv:2312.06220v2",
"legacy_risks_verbatim": "**重要问题**：会议摘要和结论提出先通道独立后混合；图3与方法显示每个块先通道注意力后时间注意力。表4列示C→S在ETTh1和Electricity上优于S→C、Weather打平；表7比较CI与CM，但未说明策略的具体操作、是否切换及其与块内顺序的关系。；
**重要问题**：同一2023年v1预印本表1将Weather列为21个特征，而数据文字写11个气象特征加1个目标；2024年v2附录A沿用后者。实际使用的Weather文件、变量及预测目标未得到独立核实。；
**重要问题**：已读会议正文与v2附录未见跨随机种子的波动或误差条，真实基准的逐模型调参预算、切分和预处理文件未完整列明；小差异的稳定性及比较条件仍有缺口。；
**重要问题**：v2附录C称前90%训练段加入不同标准差高斯噪声，后10%测试段无噪；合成数据依式(6)由10条正弦函数构造。图9中0.2–1.0噪声标准差各标点的CSformer MSE低于两条对照曲线；0.1处与PatchTST标记重叠，精确排序不清。",
"legacy_material_links_verbatim": "[评审](论文/P064/评审.md) · [评分数据](论文/P064/评审.json) · [复核](论文/P064/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-007",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 55,
"source_number": 7,
"title": "Learning from Polar Representation: An Extreme-Adaptive Model for Long-Term Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/27768",
"raw_line": "| 39 | [P007 Learning from Polar Representation: An Extreme-Adaptive Model for Long-Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/27768) | 52.9 | 95% | 32% | 47.2–58.5 | 22 | AAAI 2024 会议论文正式版；附录采用单独的 arXiv:2312.08763v2 | **重要问题**：附录表 7 的 Ross 单变量 min/max/std 为 4.23/4.35/0.02，SFC 单变量为 15.20/17.56/0.55；各列声称 6–8 次，最保守的总体标准差下界分别为 0.0275 和 0.5875，与两位小数标准差不相容。；
**重要问题**：表 2 的 UpperPen 单变量 N-BEATS RMSE 1.23 小于 DAN 1.31；附录提出四站持续胜过所有基线的表述，且未列基线多种子统计。；
**重要问题**：训练与验证随机采样于同一历史时间段，文中未明确窗口重叠隔离及标准化拟合范围；相邻 15 天输入和 3 天输出窗口可能高度相似，实际重叠或泄漏未知。；
**重要问题**：Kruskal–Wallis H 是子序列秩差异代理，图 7 的 RepMerg 加回与损失项删除同时发生，且门控没有独立定量对照。 | [评审](论文/P007/评审.md) · [评分数据](论文/P007/评审.json) · 无单独复核文件 |",
"raw_cells": [
"39",
"[P007 Learning from Polar Representation: An Extreme-Adaptive Model for Long-Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/27768)",
"52.9",
"95%",
"32%",
"47.2–58.5",
"22",
"AAAI 2024 会议论文正式版；附录采用单独的 arXiv:2312.08763v2",
"**重要问题**：附录表 7 的 Ross 单变量 min/max/std 为 4.23/4.35/0.02，SFC 单变量为 15.20/17.56/0.55；各列声称 6–8 次，最保守的总体标准差下界分别为 0.0275 和 0.5875，与两位小数标准差不相容。；
**重要问题**：表 2 的 UpperPen 单变量 N-BEATS RMSE 1.23 小于 DAN 1.31；附录提出四站持续胜过所有基线的表述，且未列基线多种子统计。；
**重要问题**：训练与验证随机采样于同一历史时间段，文中未明确窗口重叠隔离及标准化拟合范围；相邻 15 天输入和 3 天输出窗口可能高度相似，实际重叠或泄漏未知。；
**重要问题**：Kruskal–Wallis H 是子序列秩差异代理，图 7 的 RepMerg 加回与损失项删除同时发生，且门控没有独立定量对照。",
"[评审](论文/P007/评审.md) · [评分数据](论文/P007/评审.json) · 无单独复核文件"
],
"normalized_title": "learningfrompolarrepresentationanextremeadaptivemodelforlongtermtimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/27768",
"record_type": "legacy_scored",
"legacy_rank": 39,
"legacy_score": "52.9",
"legacy_coverage": "95%",
"legacy_tentative_share": "32%",
"legacy_range": "47.2–58.5",
"citation_verbatim": "22",
"legacy_review_version": "AAAI 2024 会议论文正式版；附录采用单独的 arXiv:2312.08763v2",
"legacy_risks_verbatim": "**重要问题**：附录表 7 的 Ross 单变量 min/max/std 为 4.23/4.35/0.02，SFC 单变量为 15.20/17.56/0.55；各列声称 6–8 次，最保守的总体标准差下界分别为 0.0275 和 0.5875，与两位小数标准差不相容。；
**重要问题**：表 2 的 UpperPen 单变量 N-BEATS RMSE 1.23 小于 DAN 1.31；附录提出四站持续胜过所有基线的表述，且未列基线多种子统计。；
**重要问题**：训练与验证随机采样于同一历史时间段，文中未明确窗口重叠隔离及标准化拟合范围；相邻 15 天输入和 3 天输出窗口可能高度相似，实际重叠或泄漏未知。；
**重要问题**：Kruskal–Wallis H 是子序列秩差异代理，图 7 的 RepMerg 加回与损失项删除同时发生，且门控没有独立定量对照。",
"legacy_material_links_verbatim": "[评审](论文/P007/评审.md) · [评分数据](论文/P007/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-020",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 56,
"source_number": 20,
"title": "FRNet: Frequency-based Rotation Network for Long-term Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3637528.3671713",
"raw_line": "| 40 | [P020 FRNet: Frequency-based Rotation Network for Long-term Time Series Forecasting](https://doi.org/10.1145/3637528.3671713) | 52.7 | 100% | 40% | 48.5–57.2 | 10 | KDD 2024会议排版的12页机构库存档PDF；与出版社版本未逐字或逐字节比对 | **重要问题**：FRNet式(17)较其引用的TDformer式(11)少了1−，在残差方差固定且大于零时，对季节分量方差的单调方向相反。两文相应的四个数值一致，但TDformer的0.993特指ETTm2，FRNet表8标为ETT；表8的取数或计算方式及实际实现未知。；
**重要问题**：§5.2将“所有实验场景”与跨场景平均数连用；若理解为逐场景最低，表1存在ModernTCN和CI-TSMixer胜出的行。正文声称相较FITS有改善，表1却没有FITS。；
**重要问题**：没有同架构、同容量的时域／无旋转对照；与DLinear、TimesNet的模型对照混合了分段、归一化、参数及网络结构等差异。线性嵌入后再做FFT也使物理频率解释需要额外论证。；
**重要问题**：论文报告五次重复及平均值，但未在已读论文正文和内置附录给出各次结果、标准差或区间；对0.005的总体平均MSE差距缺少直接波动信息。；
**重要问题**：按表1中七组数据、四个预测长度的28个三位小数等权重算，CI-TSMixer MAE约0.33014、Crossformer MAE约0.44314、MICN MSE为0.34800，分别与Avg.行的0.327、0.453、0.358相差超过舍入范围。论文未交代该行跨数据集及预测长度的完整汇总方法。 | [评审](论文/P020/评审.md) · [评分数据](论文/P020/评审.json) · [复核](论文/P020/复核.md) |",
"raw_cells": [
"40",
"[P020 FRNet: Frequency-based Rotation Network for Long-term Time Series Forecasting](https://doi.org/10.1145/3637528.3671713)",
"52.7",
"100%",
"40%",
"48.5–57.2",
"10",
"KDD 2024会议排版的12页机构库存档PDF；与出版社版本未逐字或逐字节比对",
"**重要问题**：FRNet式(17)较其引用的TDformer式(11)少了1−，在残差方差固定且大于零时，对季节分量方差的单调方向相反。两文相应的四个数值一致，但TDformer的0.993特指ETTm2，FRNet表8标为ETT；表8的取数或计算方式及实际实现未知。；
**重要问题**：§5.2将“所有实验场景”与跨场景平均数连用；若理解为逐场景最低，表1存在ModernTCN和CI-TSMixer胜出的行。正文声称相较FITS有改善，表1却没有FITS。；
**重要问题**：没有同架构、同容量的时域／无旋转对照；与DLinear、TimesNet的模型对照混合了分段、归一化、参数及网络结构等差异。线性嵌入后再做FFT也使物理频率解释需要额外论证。；
**重要问题**：论文报告五次重复及平均值，但未在已读论文正文和内置附录给出各次结果、标准差或区间；对0.005的总体平均MSE差距缺少直接波动信息。；
**重要问题**：按表1中七组数据、四个预测长度的28个三位小数等权重算，CI-TSMixer MAE约0.33014、Crossformer MAE约0.44314、MICN MSE为0.34800，分别与Avg.行的0.327、0.453、0.358相差超过舍入范围。论文未交代该行跨数据集及预测长度的完整汇总方法。",
"[评审](论文/P020/评审.md) · [评分数据](论文/P020/评审.json) · [复核](论文/P020/复核.md)"
],
"normalized_title": "frnetfrequencybasedrotationnetworkforlongtermtimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3637528.3671713",
"record_type": "legacy_scored",
"legacy_rank": 40,
"legacy_score": "52.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "48.5–57.2",
"citation_verbatim": "10",
"legacy_review_version": "KDD 2024会议排版的12页机构库存档PDF；与出版社版本未逐字或逐字节比对",
"legacy_risks_verbatim": "**重要问题**：FRNet式(17)较其引用的TDformer式(11)少了1−，在残差方差固定且大于零时，对季节分量方差的单调方向相反。两文相应的四个数值一致，但TDformer的0.993特指ETTm2，FRNet表8标为ETT；表8的取数或计算方式及实际实现未知。；
**重要问题**：§5.2将“所有实验场景”与跨场景平均数连用；若理解为逐场景最低，表1存在ModernTCN和CI-TSMixer胜出的行。正文声称相较FITS有改善，表1却没有FITS。；
**重要问题**：没有同架构、同容量的时域／无旋转对照；与DLinear、TimesNet的模型对照混合了分段、归一化、参数及网络结构等差异。线性嵌入后再做FFT也使物理频率解释需要额外论证。；
**重要问题**：论文报告五次重复及平均值，但未在已读论文正文和内置附录给出各次结果、标准差或区间；对0.005的总体平均MSE差距缺少直接波动信息。；
**重要问题**：按表1中七组数据、四个预测长度的28个三位小数等权重算，CI-TSMixer MAE约0.33014、Crossformer MAE约0.44314、MICN MSE为0.34800，分别与Avg.行的0.327、0.453、0.358相差超过舍入范围。论文未交代该行跨数据集及预测长度的完整汇总方法。",
"legacy_material_links_verbatim": "[评审](论文/P020/评审.md) · [评分数据](论文/P020/评审.json) · [复核](论文/P020/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-070",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 57,
"source_number": 70,
"title": "TimeCapsule: Solving the Jigsaw Puzzle of Long-Term Time Series Forecasting with Compressed Predictive Representations",
"source_url": "https://doi.org/10.1145/3711896.3737157",
"raw_line": "| 41 | [P070 TimeCapsule: Solving the Jigsaw Puzzle of Long-Term Time Series Forecasting with Compressed Predictive Representations](https://doi.org/10.1145/3711896.3737157) | 52.7 | 100% | 55% | 47.2–58.7 | 4 | arXiv:2504.12721v3预印本；会议ACM终稿未取得 | **重要问题**：按第3页模乘定义，矩阵列数须等于所乘轴长度；式(2)中 \\(\\mathcal X_1\\) 时间轴为 \\(t_c\\)，而第4页 \\(C_TE_T\\in\\mathbb R^{t_c\\times t_x}\\)，式(13)却将其作用于 \\(\\mathcal X_1\\) 的时间轴。当 \\(t_c\\ne t_x\\) 时内维不匹配。式(8)将 \\(\\mathcal Y_3\\) 时间长度写为 \\(t_x\\)，投影文字写 \\(t_y\\to t_y\\)，图3标注则为 \\(t_y\\)。[S01，第3–5页；S10，§3–4]；
**重要问题**：附录D.3先对单层级 \\(\\mathcal Y_3\\)、又对归一化后的单层级 \\(\\mathcal X_0\\) 描述同一层级轴扩展，所写 \\(M_L\\in\\mathbb R^{1\\times l}\\) 与第3页模乘方向不相容。若仅将方向改为相容的 \\(l\\times1\\) 并按该模乘操作，固定变量的第 \\(k\\) 条序列为 \\(a_kx_t\\)，非零序列的幅度谱形状相同；求和接近原序列也可能来自系数之和接近1。[S01，第3、15–20页；S10，附录D.3]；
**重要问题**：表1中TimeCapsule对ILI固定96步、其余数据集固定512步历史；基线搜索较优历史长度。正文说明部分结果来自TFB，但未逐项列出来源和选中窗口。[S01，第6页]；
**重要问题**：表5的四种随机种子只覆盖四个数据集，未提供基线的配对波动；表1中JEPA在PEMS04、720步和Traffic、96步的结果方向不同。[S01，第6、11页]；
**重要问题**：式(16)使用补零或分块EMA构造目标，只列输入与输出长度不等的分支；正文以抽象Loss表示JEPA损失，并称Huber为默认损失，但未充分列明各项权重和表示距离的完整定义。[S01，第5–6页；S10，§4.2.3、§5.1.2] | [评审](论文/P070/评审.md) · [评分数据](论文/P070/评审.json) · [复核](论文/P070/复核.md) |",
"raw_cells": [
"41",
"[P070 TimeCapsule: Solving the Jigsaw Puzzle of Long-Term Time Series Forecasting with Compressed Predictive Representations](https://doi.org/10.1145/3711896.3737157)",
"52.7",
"100%",
"55%",
"47.2–58.7",
"4",
"arXiv:2504.12721v3预印本；会议ACM终稿未取得",
"**重要问题**：按第3页模乘定义，矩阵列数须等于所乘轴长度；式(2)中 \\(\\mathcal X_1\\) 时间轴为 \\(t_c\\)，而第4页 \\(C_TE_T\\in\\mathbb R^{t_c\\times t_x}\\)，式(13)却将其作用于 \\(\\mathcal X_1\\) 的时间轴。当 \\(t_c\\ne t_x\\) 时内维不匹配。式(8)将 \\(\\mathcal Y_3\\) 时间长度写为 \\(t_x\\)，投影文字写 \\(t_y\\to t_y\\)，图3标注则为 \\(t_y\\)。[S01，第3–5页；S10，§3–4]；
**重要问题**：附录D.3先对单层级 \\(\\mathcal Y_3\\)、又对归一化后的单层级 \\(\\mathcal X_0\\) 描述同一层级轴扩展，所写 \\(M_L\\in\\mathbb R^{1\\times l}\\) 与第3页模乘方向不相容。若仅将方向改为相容的 \\(l\\times1\\) 并按该模乘操作，固定变量的第 \\(k\\) 条序列为 \\(a_kx_t\\)，非零序列的幅度谱形状相同；求和接近原序列也可能来自系数之和接近1。[S01，第3、15–20页；S10，附录D.3]；
**重要问题**：表1中TimeCapsule对ILI固定96步、其余数据集固定512步历史；基线搜索较优历史长度。正文说明部分结果来自TFB，但未逐项列出来源和选中窗口。[S01，第6页]；
**重要问题**：表5的四种随机种子只覆盖四个数据集，未提供基线的配对波动；表1中JEPA在PEMS04、720步和Traffic、96步的结果方向不同。[S01，第6、11页]；
**重要问题**：式(16)使用补零或分块EMA构造目标，只列输入与输出长度不等的分支；正文以抽象Loss表示JEPA损失，并称Huber为默认损失，但未充分列明各项权重和表示距离的完整定义。[S01，第5–6页；S10，§4.2.3、§5.1.2]",
"[评审](论文/P070/评审.md) · [评分数据](论文/P070/评审.json) · [复核](论文/P070/复核.md)"
],
"normalized_title": "timecapsulesolvingthejigsawpuzzleoflongtermtimeseriesforecastingwithcompressedpredictiverepresentations",
"normalized_url": "doi.org/10.1145/3711896.3737157",
"record_type": "legacy_scored",
"legacy_rank": 41,
"legacy_score": "52.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "47.2–58.7",
"citation_verbatim": "4",
"legacy_review_version": "arXiv:2504.12721v3预印本；会议ACM终稿未取得",
"legacy_risks_verbatim": "**重要问题**：按第3页模乘定义，矩阵列数须等于所乘轴长度；式(2)中 \\(\\mathcal X_1\\) 时间轴为 \\(t_c\\)，而第4页 \\(C_TE_T\\in\\mathbb R^{t_c\\times t_x}\\)，式(13)却将其作用于 \\(\\mathcal X_1\\) 的时间轴。当 \\(t_c\\ne t_x\\) 时内维不匹配。式(8)将 \\(\\mathcal Y_3\\) 时间长度写为 \\(t_x\\)，投影文字写 \\(t_y\\to t_y\\)，图3标注则为 \\(t_y\\)。[S01，第3–5页；S10，§3–4]；
**重要问题**：附录D.3先对单层级 \\(\\mathcal Y_3\\)、又对归一化后的单层级 \\(\\mathcal X_0\\) 描述同一层级轴扩展，所写 \\(M_L\\in\\mathbb R^{1\\times l}\\) 与第3页模乘方向不相容。若仅将方向改为相容的 \\(l\\times1\\) 并按该模乘操作，固定变量的第 \\(k\\) 条序列为 \\(a_kx_t\\)，非零序列的幅度谱形状相同；求和接近原序列也可能来自系数之和接近1。[S01，第3、15–20页；S10，附录D.3]；
**重要问题**：表1中TimeCapsule对ILI固定96步、其余数据集固定512步历史；基线搜索较优历史长度。正文说明部分结果来自TFB，但未逐项列出来源和选中窗口。[S01，第6页]；
**重要问题**：表5的四种随机种子只覆盖四个数据集，未提供基线的配对波动；表1中JEPA在PEMS04、720步和Traffic、96步的结果方向不同。[S01，第6、11页]；
**重要问题**：式(16)使用补零或分块EMA构造目标，只列输入与输出长度不等的分支；正文以抽象Loss表示JEPA损失，并称Huber为默认损失，但未充分列明各项权重和表示距离的完整定义。[S01，第5–6页；S10，§4.2.3、§5.1.2]",
"legacy_material_links_verbatim": "[评审](论文/P070/评审.md) · [评分数据](论文/P070/评审.json) · [复核](论文/P070/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-102",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 58,
"source_number": 102,
"title": "FilterTS: Comprehensive Frequency Filtering for Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/35438",
"raw_line": "| 42 | [P102 FilterTS: Comprehensive Frequency Filtering for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35438) | 52.7 | 100% | 45% | 47.7–57.2 | 27 | AAAI-25 会议出版版，39(20):21375–21383；补充使用 arXiv:2505.04158v1 附录 A–D | **重要问题**：S01表1 ETTm2/336 FilterTS 的MAE为0.398，与表列四项平均0.321不符；FreTS的Electricity四个MSE算术平均0.198，表列0.255；Weather/192 FreTS MAE列0.217低于被标为最优的0.251，其四项MAE平均0.2865又与表列0.300不符。S04仍保留后两组问题。；
**重要问题**：作者README称数据加载中误留 drop_last=True，后以False重跑并公布结果；说明未列受影响切分与样本数。；
**重要问题**：S02图A1 ETTm1中PatchTST 14ms/迭代快于FilterTS 21ms，iTransformer 0.51GB低于FilterTS 0.54GB；Traffic中FreTS 48.19GB、TimeMixer 30.76GB，Electricity中Crossformer 38.36GB，均超过附录C所述单张4090的24GB，文中未交代这些值的测量对象或获得方式。；
**重要问题**：正文和附录未报告随机种子重复次数、误差条或置信区间，而部分最优差距仅为千分之一量级。 | [评审](论文/P102/评审.md) · [评分数据](论文/P102/评审.json) · 无单独复核文件 |",
"raw_cells": [
"42",
"[P102 FilterTS: Comprehensive Frequency Filtering for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35438)",
"52.7",
"100%",
"45%",
"47.7–57.2",
"27",
"AAAI-25 会议出版版，39(20):21375–21383；补充使用 arXiv:2505.04158v1 附录 A–D",
"**重要问题**：S01表1 ETTm2/336 FilterTS 的MAE为0.398，与表列四项平均0.321不符；FreTS的Electricity四个MSE算术平均0.198，表列0.255；Weather/192 FreTS MAE列0.217低于被标为最优的0.251，其四项MAE平均0.2865又与表列0.300不符。S04仍保留后两组问题。；
**重要问题**：作者README称数据加载中误留 drop_last=True，后以False重跑并公布结果；说明未列受影响切分与样本数。；
**重要问题**：S02图A1 ETTm1中PatchTST 14ms/迭代快于FilterTS 21ms，iTransformer 0.51GB低于FilterTS 0.54GB；Traffic中FreTS 48.19GB、TimeMixer 30.76GB，Electricity中Crossformer 38.36GB，均超过附录C所述单张4090的24GB，文中未交代这些值的测量对象或获得方式。；
**重要问题**：正文和附录未报告随机种子重复次数、误差条或置信区间，而部分最优差距仅为千分之一量级。",
"[评审](论文/P102/评审.md) · [评分数据](论文/P102/评审.json) · 无单独复核文件"
],
"normalized_title": "filtertscomprehensivefrequencyfilteringformultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/35438",
"record_type": "legacy_scored",
"legacy_rank": 42,
"legacy_score": "52.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "47.7–57.2",
"citation_verbatim": "27",
"legacy_review_version": "AAAI-25 会议出版版，39(20):21375–21383；补充使用 arXiv:2505.04158v1 附录 A–D",
"legacy_risks_verbatim": "**重要问题**：S01表1 ETTm2/336 FilterTS 的MAE为0.398，与表列四项平均0.321不符；FreTS的Electricity四个MSE算术平均0.198，表列0.255；Weather/192 FreTS MAE列0.217低于被标为最优的0.251，其四项MAE平均0.2865又与表列0.300不符。S04仍保留后两组问题。；
**重要问题**：作者README称数据加载中误留 drop_last=True，后以False重跑并公布结果；说明未列受影响切分与样本数。；
**重要问题**：S02图A1 ETTm1中PatchTST 14ms/迭代快于FilterTS 21ms，iTransformer 0.51GB低于FilterTS 0.54GB；Traffic中FreTS 48.19GB、TimeMixer 30.76GB，Electricity中Crossformer 38.36GB，均超过附录C所述单张4090的24GB，文中未交代这些值的测量对象或获得方式。；
**重要问题**：正文和附录未报告随机种子重复次数、误差条或置信区间，而部分最优差距仅为千分之一量级。",
"legacy_material_links_verbatim": "[评审](论文/P102/评审.md) · [评分数据](论文/P102/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-075",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 59,
"source_number": 75,
"title": "LangTime: A Language-Guided Unified Model for Time Series Forecasting with Proximal Policy Optimization",
"source_url": "https://proceedings.mlr.press/v267/niu25e.html",
"raw_line": "| 43 | [P075 LangTime: A Language-Guided Unified Model for Time Series Forecasting with Proximal Policy Optimization](https://proceedings.mlr.press/v267/niu25e.html) | 52.2 | 100% | 40% | 47.5–57.2 | 6 | PMLR 267 官方会议论文 PDF；本地文件元数据创建日期 2025-06-06，未与 arXiv v1/v2 逐项比较。 | **重要问题**：式（11）–（12）由预测序列均值和方差构造一元高斯密度，再计算比值；旧序列 (0, 1) 与新序列 (1, 0) 的均值和方差相同，公式比值为 1，说明它对时间排列不敏感。与常规 PPO 同一状态和动作的策略比的对应关系未展开。；
**重要问题**：式（7）由 δ_t、(γλ)δ_{t+1} 等项展开至 δ_{T−1}，最后一项印作 (γλ)^(T−t+1)；若省略号表示各项依次递进且 t≤T−2，尾项的幂次应为 T−1−t。表 10 的倒数奖励和式（11）在分母为零时存在数学奇点，公式未写稳定项；实际程序、数据是否涉及这些边界未知。；
**重要问题**：表 1、2、8、18 给出点估计，未附对应的多种随机种子重复结果、标准差或置信区间；全文提取文本也未找到这些结果的重复运行统计。若干差异仅在千分位，稳定性未知。；
**重要问题**：表 1 混用统一重训与文献结果，论文未完整列出各基线的调参搜索与训练资源；附录 B.1 声称按时间划分并指向 TimesNet 协议，但未逐数据集列出时间边界、目标窗口处理和归一化统计量拟合范围，本次未核验所引协议实现。；
**重要问题**：提示消融始终保留语言指导；线性对照改变容量和架构，辅助重建改变训练目标；PPO 目标包含监督预测项，末段误差混合直接预测误差与滚动误差。 | [评审](论文/P075/评审.md) · [评分数据](论文/P075/评审.json) · [复核](论文/P075/复核.md) |",
"raw_cells": [
"43",
"[P075 LangTime: A Language-Guided Unified Model for Time Series Forecasting with Proximal Policy Optimization](https://proceedings.mlr.press/v267/niu25e.html)",
"52.2",
"100%",
"40%",
"47.5–57.2",
"6",
"PMLR 267 官方会议论文 PDF；本地文件元数据创建日期 2025-06-06，未与 arXiv v1/v2 逐项比较。",
"**重要问题**：式（11）–（12）由预测序列均值和方差构造一元高斯密度，再计算比值；旧序列 (0, 1) 与新序列 (1, 0) 的均值和方差相同，公式比值为 1，说明它对时间排列不敏感。与常规 PPO 同一状态和动作的策略比的对应关系未展开。；
**重要问题**：式（7）由 δ_t、(γλ)δ_{t+1} 等项展开至 δ_{T−1}，最后一项印作 (γλ)^(T−t+1)；若省略号表示各项依次递进且 t≤T−2，尾项的幂次应为 T−1−t。表 10 的倒数奖励和式（11）在分母为零时存在数学奇点，公式未写稳定项；实际程序、数据是否涉及这些边界未知。；
**重要问题**：表 1、2、8、18 给出点估计，未附对应的多种随机种子重复结果、标准差或置信区间；全文提取文本也未找到这些结果的重复运行统计。若干差异仅在千分位，稳定性未知。；
**重要问题**：表 1 混用统一重训与文献结果，论文未完整列出各基线的调参搜索与训练资源；附录 B.1 声称按时间划分并指向 TimesNet 协议，但未逐数据集列出时间边界、目标窗口处理和归一化统计量拟合范围，本次未核验所引协议实现。；
**重要问题**：提示消融始终保留语言指导；线性对照改变容量和架构，辅助重建改变训练目标；PPO 目标包含监督预测项，末段误差混合直接预测误差与滚动误差。",
"[评审](论文/P075/评审.md) · [评分数据](论文/P075/评审.json) · [复核](论文/P075/复核.md)"
],
"normalized_title": "langtimealanguageguidedunifiedmodelfortimeseriesforecastingwithproximalpolicyoptimization",
"normalized_url": "proceedings.mlr.press/v267/niu25e.html",
"record_type": "legacy_scored",
"legacy_rank": 43,
"legacy_score": "52.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "47.5–57.2",
"citation_verbatim": "6",
"legacy_review_version": "PMLR 267 官方会议论文 PDF；本地文件元数据创建日期 2025-06-06，未与 arXiv v1/v2 逐项比较。",
"legacy_risks_verbatim": "**重要问题**：式（11）–（12）由预测序列均值和方差构造一元高斯密度，再计算比值；旧序列 (0, 1) 与新序列 (1, 0) 的均值和方差相同，公式比值为 1，说明它对时间排列不敏感。与常规 PPO 同一状态和动作的策略比的对应关系未展开。；
**重要问题**：式（7）由 δ_t、(γλ)δ_{t+1} 等项展开至 δ_{T−1}，最后一项印作 (γλ)^(T−t+1)；若省略号表示各项依次递进且 t≤T−2，尾项的幂次应为 T−1−t。表 10 的倒数奖励和式（11）在分母为零时存在数学奇点，公式未写稳定项；实际程序、数据是否涉及这些边界未知。；
**重要问题**：表 1、2、8、18 给出点估计，未附对应的多种随机种子重复结果、标准差或置信区间；全文提取文本也未找到这些结果的重复运行统计。若干差异仅在千分位，稳定性未知。；
**重要问题**：表 1 混用统一重训与文献结果，论文未完整列出各基线的调参搜索与训练资源；附录 B.1 声称按时间划分并指向 TimesNet 协议，但未逐数据集列出时间边界、目标窗口处理和归一化统计量拟合范围，本次未核验所引协议实现。；
**重要问题**：提示消融始终保留语言指导；线性对照改变容量和架构，辅助重建改变训练目标；PPO 目标包含监督预测项，末段误差混合直接预测误差与滚动误差。",
"legacy_material_links_verbatim": "[评审](论文/P075/评审.md) · [评分数据](论文/P075/评审.json) · [复核](论文/P075/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-108",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 60,
"source_number": 108,
"title": "Revisiting Attention for Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/35455",
"raw_line": "| 44 | [P108 Revisiting Attention for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35455) | 52.0 | 100% | 70% | 44.2–61.5 | 15 | AAAI 2025 会议出版版，卷 39、期 20，页 21528–21535；同时单独参阅 arXiv:2407.13806v1 | **重要问题**：会议版表 2 的 Traffic 变量版 SOatten、表 1 的 Traffic FSatten 均差于 iTransformer；ETTm1/ETTm2 与 Exchange 的其他基线也更强，宽泛优势叙述超出表格支持的范围。；
**重要问题**：SOatten 仅正交初始化；频谱幅值丢失相位；HCC 及条件数分析未隔离邻近顺序、参数化、训练和物理依赖等因素。；
**重要问题**：会议正文与已读的 v1 全文附录未报告新方法的重复种子标准差或置信区间；部分表格差异小。会议版所引附录未取得。；
**重要问题**：作者公开仓库根目录含 Logs、README、LICENSE，README 仍称代码将发布；当前入口未显示新方法实现。 | [评审](论文/P108/评审.md) · [评分数据](论文/P108/评审.json) · 无单独复核文件 |",
"raw_cells": [
"44",
"[P108 Revisiting Attention for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/35455)",
"52.0",
"100%",
"70%",
"44.2–61.5",
"15",
"AAAI 2025 会议出版版，卷 39、期 20，页 21528–21535；同时单独参阅 arXiv:2407.13806v1",
"**重要问题**：会议版表 2 的 Traffic 变量版 SOatten、表 1 的 Traffic FSatten 均差于 iTransformer；ETTm1/ETTm2 与 Exchange 的其他基线也更强，宽泛优势叙述超出表格支持的范围。；
**重要问题**：SOatten 仅正交初始化；频谱幅值丢失相位；HCC 及条件数分析未隔离邻近顺序、参数化、训练和物理依赖等因素。；
**重要问题**：会议正文与已读的 v1 全文附录未报告新方法的重复种子标准差或置信区间；部分表格差异小。会议版所引附录未取得。；
**重要问题**：作者公开仓库根目录含 Logs、README、LICENSE，README 仍称代码将发布；当前入口未显示新方法实现。",
"[评审](论文/P108/评审.md) · [评分数据](论文/P108/评审.json) · 无单独复核文件"
],
"normalized_title": "revisitingattentionformultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/35455",
"record_type": "legacy_scored",
"legacy_rank": 44,
"legacy_score": "52.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "70%",
"legacy_range": "44.2–61.5",
"citation_verbatim": "15",
"legacy_review_version": "AAAI 2025 会议出版版，卷 39、期 20，页 21528–21535；同时单独参阅 arXiv:2407.13806v1",
"legacy_risks_verbatim": "**重要问题**：会议版表 2 的 Traffic 变量版 SOatten、表 1 的 Traffic FSatten 均差于 iTransformer；ETTm1/ETTm2 与 Exchange 的其他基线也更强，宽泛优势叙述超出表格支持的范围。；
**重要问题**：SOatten 仅正交初始化；频谱幅值丢失相位；HCC 及条件数分析未隔离邻近顺序、参数化、训练和物理依赖等因素。；
**重要问题**：会议正文与已读的 v1 全文附录未报告新方法的重复种子标准差或置信区间；部分表格差异小。会议版所引附录未取得。；
**重要问题**：作者公开仓库根目录含 Logs、README、LICENSE，README 仍称代码将发布；当前入口未显示新方法实现。",
"legacy_material_links_verbatim": "[评审](论文/P108/评审.md) · [评分数据](论文/P108/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-005",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 61,
"source_number": 5,
"title": "Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/298c3e32d7d402189444be2ff5d19979-Abstract-Conference.html",
"raw_line": "| 45 | [P005 Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling](https://proceedings.neurips.cc/paper_files/paper/2024/hash/298c3e32d7d402189444be2ff5d19979-Abstract-Conference.html) | 51.7 | 100% | 50% | 46.5–58.0 | 9 | NeurIPS 2024 官方会议论文 PDF；文件内部创建时间 2025-01-13 06:22:54 UTC；与 arXiv v5 未作逐字或逐字节同一性确认 | **重要问题**：§4.1 称 1980–2015 训练、2017 为验证和测试；附录 C 表 6 称 1980–2014 训练、2015 验证、2017–2018 测试。；
**重要问题**：§3.6 与表 5 的直接监督输出为 1、3、6 小时，表 3 称 60 和 120 分钟为已训练时效；120 分钟的称谓与第 8 个块是否直接监督及评测推理路径的关系未说明。；
**重要问题**：表 3 列出三类预报器各配两种插帧方法；未列持续性、线性插值、直接采用连续时间模型的半小时输出或专业雷达临近预报对照，插帧设置、预算及中期 IFS 配置说明亦不足。；
**重要问题**：压力坐标的物质导数式 (12) 为 w∂/∂p，湿度演化式 (23) 的位势和湿度垂直项使用 w∂/∂z；正文以 z 表示位势。附录 B 将 d_z 标为压力层方向微分，但未说明与式 (23) 的对应。；
**重要问题**：路由 r 为可学习通道权重，融合还经过 MLP；图 10 右侧报告分支范数相近，图 9 和图 10 左侧给出偏差与能量现象，表 4 依次加入 PDE 和多时效监督，尚未充分隔离分支、容量与训练目标的影响。；
**重要问题**：表 3、图 4 和图 8 未给重复训练波动或置信区间；NASA 产品版本、预测量累计窗口、时空匹配及阈值单位在已读 PDF 中未充分说明。 | [评审](论文/P005/评审.md) · [评分数据](论文/P005/评审.json) · [复核](论文/P005/复核.md) |",
"raw_cells": [
"45",
"[P005 Generalizing Weather Forecast to Fine-grained Temporal Scales via Physics-AI Hybrid Modeling](https://proceedings.neurips.cc/paper_files/paper/2024/hash/298c3e32d7d402189444be2ff5d19979-Abstract-Conference.html)",
"51.7",
"100%",
"50%",
"46.5–58.0",
"9",
"NeurIPS 2024 官方会议论文 PDF；文件内部创建时间 2025-01-13 06:22:54 UTC；与 arXiv v5 未作逐字或逐字节同一性确认",
"**重要问题**：§4.1 称 1980–2015 训练、2017 为验证和测试；附录 C 表 6 称 1980–2014 训练、2015 验证、2017–2018 测试。；
**重要问题**：§3.6 与表 5 的直接监督输出为 1、3、6 小时，表 3 称 60 和 120 分钟为已训练时效；120 分钟的称谓与第 8 个块是否直接监督及评测推理路径的关系未说明。；
**重要问题**：表 3 列出三类预报器各配两种插帧方法；未列持续性、线性插值、直接采用连续时间模型的半小时输出或专业雷达临近预报对照，插帧设置、预算及中期 IFS 配置说明亦不足。；
**重要问题**：压力坐标的物质导数式 (12) 为 w∂/∂p，湿度演化式 (23) 的位势和湿度垂直项使用 w∂/∂z；正文以 z 表示位势。附录 B 将 d_z 标为压力层方向微分，但未说明与式 (23) 的对应。；
**重要问题**：路由 r 为可学习通道权重，融合还经过 MLP；图 10 右侧报告分支范数相近，图 9 和图 10 左侧给出偏差与能量现象，表 4 依次加入 PDE 和多时效监督，尚未充分隔离分支、容量与训练目标的影响。；
**重要问题**：表 3、图 4 和图 8 未给重复训练波动或置信区间；NASA 产品版本、预测量累计窗口、时空匹配及阈值单位在已读 PDF 中未充分说明。",
"[评审](论文/P005/评审.md) · [评分数据](论文/P005/评审.json) · [复核](论文/P005/复核.md)"
],
"normalized_title": "generalizingweatherforecasttofinegrainedtemporalscalesviaphysicsaihybridmodeling",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/298c3e32d7d402189444be2ff5d19979-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 45,
"legacy_score": "51.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "46.5–58.0",
"citation_verbatim": "9",
"legacy_review_version": "NeurIPS 2024 官方会议论文 PDF；文件内部创建时间 2025-01-13 06:22:54 UTC；与 arXiv v5 未作逐字或逐字节同一性确认",
"legacy_risks_verbatim": "**重要问题**：§4.1 称 1980–2015 训练、2017 为验证和测试；附录 C 表 6 称 1980–2014 训练、2015 验证、2017–2018 测试。；
**重要问题**：§3.6 与表 5 的直接监督输出为 1、3、6 小时，表 3 称 60 和 120 分钟为已训练时效；120 分钟的称谓与第 8 个块是否直接监督及评测推理路径的关系未说明。；
**重要问题**：表 3 列出三类预报器各配两种插帧方法；未列持续性、线性插值、直接采用连续时间模型的半小时输出或专业雷达临近预报对照，插帧设置、预算及中期 IFS 配置说明亦不足。；
**重要问题**：压力坐标的物质导数式 (12) 为 w∂/∂p，湿度演化式 (23) 的位势和湿度垂直项使用 w∂/∂z；正文以 z 表示位势。附录 B 将 d_z 标为压力层方向微分，但未说明与式 (23) 的对应。；
**重要问题**：路由 r 为可学习通道权重，融合还经过 MLP；图 10 右侧报告分支范数相近，图 9 和图 10 左侧给出偏差与能量现象，表 4 依次加入 PDE 和多时效监督，尚未充分隔离分支、容量与训练目标的影响。；
**重要问题**：表 3、图 4 和图 8 未给重复训练波动或置信区间；NASA 产品版本、预测量累计窗口、时空匹配及阈值单位在已读 PDF 中未充分说明。",
"legacy_material_links_verbatim": "[评审](论文/P005/评审.md) · [评分数据](论文/P005/评审.json) · [复核](论文/P005/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-012",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 62,
"source_number": 12,
"title": "Efficient and Effective Time-Series Forecasting with Spiking Neural Networks",
"source_url": "https://proceedings.mlr.press/v235/lv24d.html",
"raw_line": "| 46 | [P012 Efficient and Effective Time-Series Forecasting with Spiking Neural Networks](https://proceedings.mlr.press/v235/lv24d.html) | 51.7 | 100% | 45% | 47.0–57.0 | 20 | ICML 2024 PMLR 235 会议版，本地 PDF SHA-256 261cf58485cb998fcbebfc02609681e3b86e3b14e1cfdbff49e97f4570246cd1 | **重要问题**：附录 A.4 式 (11) 写成逐样本误差平方与逐样本中心化平方之比的平均，和常用先聚合平方和的 R² 公式不同；分母可能为零。表格实际使用何种实现未知。；
**重要问题**：表 3 的 Spike-TCN 0.15 G SOP 按附录 B 的 0.9 pJ/次相乘得约 0.135 mJ，表中为 0.23 mJ；Spike-GRU 为约 1.467 mJ 与 1.51 mJ。三个降幅算术均值为 68.32%，与正文 70.33% 不同。；
**重要问题**：主表转引多项 Fang et al. 的结果，但未包含同一来源的 SRD 图模型，也未列简单线性或持续值参照；跨论文预处理和调参统一程度未明确。 | [评审](论文/P012/评审.md) · [评分数据](论文/P012/评审.json) · 无单独复核文件 |",
"raw_cells": [
"46",
"[P012 Efficient and Effective Time-Series Forecasting with Spiking Neural Networks](https://proceedings.mlr.press/v235/lv24d.html)",
"51.7",
"100%",
"45%",
"47.0–57.0",
"20",
"ICML 2024 PMLR 235 会议版，本地 PDF SHA-256 261cf58485cb998fcbebfc02609681e3b86e3b14e1cfdbff49e97f4570246cd1",
"**重要问题**：附录 A.4 式 (11) 写成逐样本误差平方与逐样本中心化平方之比的平均，和常用先聚合平方和的 R² 公式不同；分母可能为零。表格实际使用何种实现未知。；
**重要问题**：表 3 的 Spike-TCN 0.15 G SOP 按附录 B 的 0.9 pJ/次相乘得约 0.135 mJ，表中为 0.23 mJ；Spike-GRU 为约 1.467 mJ 与 1.51 mJ。三个降幅算术均值为 68.32%，与正文 70.33% 不同。；
**重要问题**：主表转引多项 Fang et al. 的结果，但未包含同一来源的 SRD 图模型，也未列简单线性或持续值参照；跨论文预处理和调参统一程度未明确。",
"[评审](论文/P012/评审.md) · [评分数据](论文/P012/评审.json) · 无单独复核文件"
],
"normalized_title": "efficientandeffectivetimeseriesforecastingwithspikingneuralnetworks",
"normalized_url": "proceedings.mlr.press/v235/lv24d.html",
"record_type": "legacy_scored",
"legacy_rank": 46,
"legacy_score": "51.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "47.0–57.0",
"citation_verbatim": "20",
"legacy_review_version": "ICML 2024 PMLR 235 会议版，本地 PDF SHA-256 261cf58485cb998fcbebfc02609681e3b86e3b14e1cfdbff49e97f4570246cd1",
"legacy_risks_verbatim": "**重要问题**：附录 A.4 式 (11) 写成逐样本误差平方与逐样本中心化平方之比的平均，和常用先聚合平方和的 R² 公式不同；分母可能为零。表格实际使用何种实现未知。；
**重要问题**：表 3 的 Spike-TCN 0.15 G SOP 按附录 B 的 0.9 pJ/次相乘得约 0.135 mJ，表中为 0.23 mJ；Spike-GRU 为约 1.467 mJ 与 1.51 mJ。三个降幅算术均值为 68.32%，与正文 70.33% 不同。；
**重要问题**：主表转引多项 Fang et al. 的结果，但未包含同一来源的 SRD 图模型，也未列简单线性或持续值参照；跨论文预处理和调参统一程度未明确。",
"legacy_material_links_verbatim": "[评审](论文/P012/评审.md) · [评分数据](论文/P012/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-023",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 63,
"source_number": 23,
"title": "GinAR: An End-To-End Multivariate Time Series Forecasting Model Suitable for Variable Missing",
"source_url": "https://doi.org/10.1145/3637528.3672055",
"raw_line": "| 47 | [P023 GinAR: An End-To-End Multivariate Time Series Forecasting Model Suitable for Variable Missing](https://doi.org/10.1145/3637528.3672055) | 51.7 | 100% | 45% | 46.7–58.0 | 52 | arXiv:2405.11333v1（本地预印本；KDD 2024 会议正式版全文未读） | **重要问题**：论文给出五个随机种子的均值，却未给逐种子值、标准差或区间，也未明确各模型是否共享配对掩码。部分 90% 缺失下 MAE 优势约 1%。；
**重要问题**：论文未交代 Pearson 预定义图所用的数据时间分区、掩码是否按窗口重抽、各模型掩码是否配对以及两阶段最优配置依据的分区。；
**重要问题**：对有限 logits，式 (1) 的 softmax 每个元素均大于零，结合正文的 >0 选邻居规则会包含所有正常变量；式 (2) 的系数除邻居集合外未显式依赖目标 i。；
**重要问题**：90% 缺失的结论仅有整体点预测误差与地图示例，缺少按输入是否缺失分组的指标及逐节点误差。 | [评审](论文/P023/评审.md) · [评分数据](论文/P023/评审.json) · 无单独复核文件 |",
"raw_cells": [
"47",
"[P023 GinAR: An End-To-End Multivariate Time Series Forecasting Model Suitable for Variable Missing](https://doi.org/10.1145/3637528.3672055)",
"51.7",
"100%",
"45%",
"46.7–58.0",
"52",
"arXiv:2405.11333v1（本地预印本；KDD 2024 会议正式版全文未读）",
"**重要问题**：论文给出五个随机种子的均值，却未给逐种子值、标准差或区间，也未明确各模型是否共享配对掩码。部分 90% 缺失下 MAE 优势约 1%。；
**重要问题**：论文未交代 Pearson 预定义图所用的数据时间分区、掩码是否按窗口重抽、各模型掩码是否配对以及两阶段最优配置依据的分区。；
**重要问题**：对有限 logits，式 (1) 的 softmax 每个元素均大于零，结合正文的 >0 选邻居规则会包含所有正常变量；式 (2) 的系数除邻居集合外未显式依赖目标 i。；
**重要问题**：90% 缺失的结论仅有整体点预测误差与地图示例，缺少按输入是否缺失分组的指标及逐节点误差。",
"[评审](论文/P023/评审.md) · [评分数据](论文/P023/评审.json) · 无单独复核文件"
],
"normalized_title": "ginaranendtoendmultivariatetimeseriesforecastingmodelsuitableforvariablemissing",
"normalized_url": "doi.org/10.1145/3637528.3672055",
"record_type": "legacy_scored",
"legacy_rank": 47,
"legacy_score": "51.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "46.7–58.0",
"citation_verbatim": "52",
"legacy_review_version": "arXiv:2405.11333v1（本地预印本；KDD 2024 会议正式版全文未读）",
"legacy_risks_verbatim": "**重要问题**：论文给出五个随机种子的均值，却未给逐种子值、标准差或区间，也未明确各模型是否共享配对掩码。部分 90% 缺失下 MAE 优势约 1%。；
**重要问题**：论文未交代 Pearson 预定义图所用的数据时间分区、掩码是否按窗口重抽、各模型掩码是否配对以及两阶段最优配置依据的分区。；
**重要问题**：对有限 logits，式 (1) 的 softmax 每个元素均大于零，结合正文的 >0 选邻居规则会包含所有正常变量；式 (2) 的系数除邻居集合外未显式依赖目标 i。；
**重要问题**：90% 缺失的结论仅有整体点预测误差与地图示例，缺少按输入是否缺失分组的指标及逐节点误差。",
"legacy_material_links_verbatim": "[评审](论文/P023/评审.md) · [评分数据](论文/P023/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-056",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 64,
"source_number": 56,
"title": "Addressing Spatial-Temporal Heterogeneity: General Mixed Time Series Analysis via Latent Continuity Recovery and Alignment",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/1feb87871436031bdc0f2beaa62a049b-Abstract-Conference.html",
"raw_line": "| 48 | [P056 Addressing Spatial-Temporal Heterogeneity: General Mixed Time Series Analysis via Latent Continuity Recovery and Alignment](https://proceedings.neurips.cc/paper_files/paper/2024/hash/1feb87871436031bdc0f2beaa62a049b-Abstract-Conference.html) | 51.5 | 100% | 80% | 42.5–61.7 | 7 | NeurIPS 2024 会议论文集官方 PDF，37 页；文件创建元数据为 2024-12-23，非已确认的公开日期。 | **重要问题**：式 (2) 对 log D(zC)+log(1−D(zLC)) 写判别器最小化、恢复器最大化；按该标签对数似然，准确判别的方向相反。图 4 将 GRL 置于判别器之前，而算法 1 第 32 行写在判别器之后。；
**重要问题**：图 8 与表 15 的 SMD、SWaT、PSM 数值不同；表 4 的 SMAP 为 96.83，表 15 的 SMAP 为 82.69、PSM 为 96.83，图表未交代报数是否来自相同运行及统计口径。表 15 的 Crossformer SWaT P=94.47、R=94.77、F1=96.11；若按同一统计单位对所列 P、R 计算调和均值，结果约 94.62。论文称实验重复三次，各指标汇总方式待澄清。；
**重要问题**：表 5 将 ETTh1/ETTh2 的 7 个变量标为 6 个离散变量，与同表 n=⌊p/2⌋ 及图 15 的 ETTh1 3 DVs/4 CVs 冲突；表 5 的 ILI 输入长度为 24，表 16、17 的表注为 36。第 20 页文字及表 16 表注称 RMSE，表 2、3、14、16 对应列标题为 MSE。若对同一批数据以相同权重计算，RMSE 应不小于 MAE；Weather 96 步的第二列 0.147 小于 MAE 0.231，具体计算和统计口径待核实。；
**重要问题**：论文将数据集中约一半变量经 MinMax 和阈值离散化；尚未在论文中找到原生混合变量独立评测。自然类别、离散比例与规则改变时的性能未知。 | [评审](论文/P056/评审.md) · [评分数据](论文/P056/评审.json) · [复核](论文/P056/复核.md) |",
"raw_cells": [
"48",
"[P056 Addressing Spatial-Temporal Heterogeneity: General Mixed Time Series Analysis via Latent Continuity Recovery and Alignment](https://proceedings.neurips.cc/paper_files/paper/2024/hash/1feb87871436031bdc0f2beaa62a049b-Abstract-Conference.html)",
"51.5",
"100%",
"80%",
"42.5–61.7",
"7",
"NeurIPS 2024 会议论文集官方 PDF，37 页；文件创建元数据为 2024-12-23，非已确认的公开日期。",
"**重要问题**：式 (2) 对 log D(zC)+log(1−D(zLC)) 写判别器最小化、恢复器最大化；按该标签对数似然，准确判别的方向相反。图 4 将 GRL 置于判别器之前，而算法 1 第 32 行写在判别器之后。；
**重要问题**：图 8 与表 15 的 SMD、SWaT、PSM 数值不同；表 4 的 SMAP 为 96.83，表 15 的 SMAP 为 82.69、PSM 为 96.83，图表未交代报数是否来自相同运行及统计口径。表 15 的 Crossformer SWaT P=94.47、R=94.77、F1=96.11；若按同一统计单位对所列 P、R 计算调和均值，结果约 94.62。论文称实验重复三次，各指标汇总方式待澄清。；
**重要问题**：表 5 将 ETTh1/ETTh2 的 7 个变量标为 6 个离散变量，与同表 n=⌊p/2⌋ 及图 15 的 ETTh1 3 DVs/4 CVs 冲突；表 5 的 ILI 输入长度为 24，表 16、17 的表注为 36。第 20 页文字及表 16 表注称 RMSE，表 2、3、14、16 对应列标题为 MSE。若对同一批数据以相同权重计算，RMSE 应不小于 MAE；Weather 96 步的第二列 0.147 小于 MAE 0.231，具体计算和统计口径待核实。；
**重要问题**：论文将数据集中约一半变量经 MinMax 和阈值离散化；尚未在论文中找到原生混合变量独立评测。自然类别、离散比例与规则改变时的性能未知。",
"[评审](论文/P056/评审.md) · [评分数据](论文/P056/评审.json) · [复核](论文/P056/复核.md)"
],
"normalized_title": "addressingspatialtemporalheterogeneitygeneralmixedtimeseriesanalysisvialatentcontinuityrecoveryandalignment",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/1feb87871436031bdc0f2beaa62a049b-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 48,
"legacy_score": "51.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "80%",
"legacy_range": "42.5–61.7",
"citation_verbatim": "7",
"legacy_review_version": "NeurIPS 2024 会议论文集官方 PDF，37 页；文件创建元数据为 2024-12-23，非已确认的公开日期。",
"legacy_risks_verbatim": "**重要问题**：式 (2) 对 log D(zC)+log(1−D(zLC)) 写判别器最小化、恢复器最大化；按该标签对数似然，准确判别的方向相反。图 4 将 GRL 置于判别器之前，而算法 1 第 32 行写在判别器之后。；
**重要问题**：图 8 与表 15 的 SMD、SWaT、PSM 数值不同；表 4 的 SMAP 为 96.83，表 15 的 SMAP 为 82.69、PSM 为 96.83，图表未交代报数是否来自相同运行及统计口径。表 15 的 Crossformer SWaT P=94.47、R=94.77、F1=96.11；若按同一统计单位对所列 P、R 计算调和均值，结果约 94.62。论文称实验重复三次，各指标汇总方式待澄清。；
**重要问题**：表 5 将 ETTh1/ETTh2 的 7 个变量标为 6 个离散变量，与同表 n=⌊p/2⌋ 及图 15 的 ETTh1 3 DVs/4 CVs 冲突；表 5 的 ILI 输入长度为 24，表 16、17 的表注为 36。第 20 页文字及表 16 表注称 RMSE，表 2、3、14、16 对应列标题为 MSE。若对同一批数据以相同权重计算，RMSE 应不小于 MAE；Weather 96 步的第二列 0.147 小于 MAE 0.231，具体计算和统计口径待核实。；
**重要问题**：论文将数据集中约一半变量经 MinMax 和阈值离散化；尚未在论文中找到原生混合变量独立评测。自然类别、离散比例与规则改变时的性能未知。",
"legacy_material_links_verbatim": "[评审](论文/P056/评审.md) · [评分数据](论文/P056/评审.json) · [复核](论文/P056/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-149",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 65,
"source_number": 149,
"title": "Augmented Contrastive Clustering with Uncertainty-Aware Prototyping for Time Series Test Time Adaptation",
"source_url": "https://doi.org/10.1145/3690624.3709239",
"raw_line": "| 49 | [P149 Augmented Contrastive Clustering with Uncertainty-Aware Prototyping for Time Series Test Time Adaptation](https://doi.org/10.1145/3690624.3709239) | 51.3 | 95% | 53% | 42.7–59.2 | 7 | arXiv:2501.01472v1 当前可获取的14页 PDF；会议终稿未获取 | **重要问题**：论文分别列出数据集专属 K、η、τ，图4展示敏感性，附录称对基线随机搜索20组；已读材料未交代参数选择使用的数据划分、目标标签权限、各方法统一选择规则或时间序列测试批次大小与顺序。现有证据不表明发生了泄漏，也未能独立核验公平性。；
**重要问题**：式3的 η 改变原型分支概率的尖锐程度，式4据熵在两分支间选取；论文未报告两分支熵的跨分支校准或被选伪标签真实准确率。低熵是否对应更高正确率需要额外证据。 | [评审](论文/P149/评审.md) · [评分数据](论文/P149/评审.json) · [复核](论文/P149/复核.md) |",
"raw_cells": [
"49",
"[P149 Augmented Contrastive Clustering with Uncertainty-Aware Prototyping for Time Series Test Time Adaptation](https://doi.org/10.1145/3690624.3709239)",
"51.3",
"95%",
"53%",
"42.7–59.2",
"7",
"arXiv:2501.01472v1 当前可获取的14页 PDF；会议终稿未获取",
"**重要问题**：论文分别列出数据集专属 K、η、τ，图4展示敏感性，附录称对基线随机搜索20组；已读材料未交代参数选择使用的数据划分、目标标签权限、各方法统一选择规则或时间序列测试批次大小与顺序。现有证据不表明发生了泄漏，也未能独立核验公平性。；
**重要问题**：式3的 η 改变原型分支概率的尖锐程度，式4据熵在两分支间选取；论文未报告两分支熵的跨分支校准或被选伪标签真实准确率。低熵是否对应更高正确率需要额外证据。",
"[评审](论文/P149/评审.md) · [评分数据](论文/P149/评审.json) · [复核](论文/P149/复核.md)"
],
"normalized_title": "augmentedcontrastiveclusteringwithuncertaintyawareprototypingfortimeseriestesttimeadaptation",
"normalized_url": "doi.org/10.1145/3690624.3709239",
"record_type": "legacy_scored",
"legacy_rank": 49,
"legacy_score": "51.3",
"legacy_coverage": "95%",
"legacy_tentative_share": "53%",
"legacy_range": "42.7–59.2",
"citation_verbatim": "7",
"legacy_review_version": "arXiv:2501.01472v1 当前可获取的14页 PDF；会议终稿未获取",
"legacy_risks_verbatim": "**重要问题**：论文分别列出数据集专属 K、η、τ，图4展示敏感性，附录称对基线随机搜索20组；已读材料未交代参数选择使用的数据划分、目标标签权限、各方法统一选择规则或时间序列测试批次大小与顺序。现有证据不表明发生了泄漏，也未能独立核验公平性。；
**重要问题**：式3的 η 改变原型分支概率的尖锐程度，式4据熵在两分支间选取；论文未报告两分支熵的跨分支校准或被选伪标签真实准确率。低熵是否对应更高正确率需要额外证据。",
"legacy_material_links_verbatim": "[评审](论文/P149/评审.md) · [评分数据](论文/P149/评审.json) · [复核](论文/P149/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-157",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 66,
"source_number": 157,
"title": "Federated Foundation Models on Heterogeneous Time Series",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33739",
"raw_line": "| 50 | [P157 Federated Foundation Models on Heterogeneous Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/33739) | 51.2 | 100% | 80% | 42.7–61.0 | 18 | AAAI 2025 会议正式版，卷39第15期，15839–15847 | **重要问题**：所读会议版和v1均将18个预训练数据集的完整清单及关键评估细节指向附录；官方补充检索未找到附件。原文说排除了下游数据，但独立核查所需材料缺失，数据是否重合仍未知。；
**重要问题**：会议版第15844页称目标集不训练，表5有ETT源→目标且将两组数并于FFTS列；预印本v1表6将对应两组列为FFTS(FT)、FFTS-Cen(FT)，图注称FT为在相应数据集微调，具体源端或目标端的操作未说明。；
**重要问题**：会议版表3 FFTS与FFTS-Cen的总体MSE分别为0.473和0.475，表7平均F1为88.10和87.69；正文相关表格未给重复运行波动，且完整预算及基线调参信息不可核查。 | [评审](论文/P157/评审.md) · [评分数据](论文/P157/评审.json) · 无单独复核文件 |",
"raw_cells": [
"50",
"[P157 Federated Foundation Models on Heterogeneous Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/33739)",
"51.2",
"100%",
"80%",
"42.7–61.0",
"18",
"AAAI 2025 会议正式版，卷39第15期，15839–15847",
"**重要问题**：所读会议版和v1均将18个预训练数据集的完整清单及关键评估细节指向附录；官方补充检索未找到附件。原文说排除了下游数据，但独立核查所需材料缺失，数据是否重合仍未知。；
**重要问题**：会议版第15844页称目标集不训练，表5有ETT源→目标且将两组数并于FFTS列；预印本v1表6将对应两组列为FFTS(FT)、FFTS-Cen(FT)，图注称FT为在相应数据集微调，具体源端或目标端的操作未说明。；
**重要问题**：会议版表3 FFTS与FFTS-Cen的总体MSE分别为0.473和0.475，表7平均F1为88.10和87.69；正文相关表格未给重复运行波动，且完整预算及基线调参信息不可核查。",
"[评审](论文/P157/评审.md) · [评分数据](论文/P157/评审.json) · 无单独复核文件"
],
"normalized_title": "federatedfoundationmodelsonheterogeneoustimeseries",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33739",
"record_type": "legacy_scored",
"legacy_rank": 50,
"legacy_score": "51.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "80%",
"legacy_range": "42.7–61.0",
"citation_verbatim": "18",
"legacy_review_version": "AAAI 2025 会议正式版，卷39第15期，15839–15847",
"legacy_risks_verbatim": "**重要问题**：所读会议版和v1均将18个预训练数据集的完整清单及关键评估细节指向附录；官方补充检索未找到附件。原文说排除了下游数据，但独立核查所需材料缺失，数据是否重合仍未知。；
**重要问题**：会议版第15844页称目标集不训练，表5有ETT源→目标且将两组数并于FFTS列；预印本v1表6将对应两组列为FFTS(FT)、FFTS-Cen(FT)，图注称FT为在相应数据集微调，具体源端或目标端的操作未说明。；
**重要问题**：会议版表3 FFTS与FFTS-Cen的总体MSE分别为0.473和0.475，表7平均F1为88.10和87.69；正文相关表格未给重复运行波动，且完整预算及基线调参信息不可核查。",
"legacy_material_links_verbatim": "[评审](论文/P157/评审.md) · [评分数据](论文/P157/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-198",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 67,
"source_number": 198,
"title": "Beyond Model Ranking: Predictability-Aligned Evaluation for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/feng26u.html",
"raw_line": "| 51 | [P198 Beyond Model Ranking: Predictability-Aligned Evaluation for Time Series Forecasting](https://proceedings.mlr.press/v306/feng26u.html) | 51.2 | 100% | 45% | 45.5–57.5 | 0 | ICML 2026 正式版，PMLR v306，20 页；与已有本地 PDF 哈希一致 | **关键问题**：算法1定义加入窗口均值差平方的逐实例线性 MSE 下界；取 x 均值1且非常数、y=2x，f(x)=2x，真实MSE=0而该值至少1。仅针对逐样本形式，平稳过程总体期望界另论。；
**关键问题**：x=y且预测为−x时，预测与历史的平方相干完全相同、LUR=1，但实际MSE为4mean(y²)>0；不能由LUR单独确认接近线性MSE最优。；
**重要问题**：公式使用未来 y，不能作为当时刻无需未来标签的实例评分；这是事后诊断而非自动证明测试污染。 | [评审](论文/P198/评审.md) · [评分数据](论文/P198/评审.json) · [复核](论文/P198/复核.md) |",
"raw_cells": [
"51",
"[P198 Beyond Model Ranking: Predictability-Aligned Evaluation for Time Series Forecasting](https://proceedings.mlr.press/v306/feng26u.html)",
"51.2",
"100%",
"45%",
"45.5–57.5",
"0",
"ICML 2026 正式版，PMLR v306，20 页；与已有本地 PDF 哈希一致",
"**关键问题**：算法1定义加入窗口均值差平方的逐实例线性 MSE 下界；取 x 均值1且非常数、y=2x，f(x)=2x，真实MSE=0而该值至少1。仅针对逐样本形式，平稳过程总体期望界另论。；
**关键问题**：x=y且预测为−x时，预测与历史的平方相干完全相同、LUR=1，但实际MSE为4mean(y²)>0；不能由LUR单独确认接近线性MSE最优。；
**重要问题**：公式使用未来 y，不能作为当时刻无需未来标签的实例评分；这是事后诊断而非自动证明测试污染。",
"[评审](论文/P198/评审.md) · [评分数据](论文/P198/评审.json) · [复核](论文/P198/复核.md)"
],
"normalized_title": "beyondmodelrankingpredictabilityalignedevaluationfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/feng26u.html",
"record_type": "legacy_scored",
"legacy_rank": 51,
"legacy_score": "51.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "45.5–57.5",
"citation_verbatim": "0",
"legacy_review_version": "ICML 2026 正式版，PMLR v306，20 页；与已有本地 PDF 哈希一致",
"legacy_risks_verbatim": "**关键问题**：算法1定义加入窗口均值差平方的逐实例线性 MSE 下界；取 x 均值1且非常数、y=2x，f(x)=2x，真实MSE=0而该值至少1。仅针对逐样本形式，平稳过程总体期望界另论。；
**关键问题**：x=y且预测为−x时，预测与历史的平方相干完全相同、LUR=1，但实际MSE为4mean(y²)>0；不能由LUR单独确认接近线性MSE最优。；
**重要问题**：公式使用未来 y，不能作为当时刻无需未来标签的实例评分；这是事后诊断而非自动证明测试污染。",
"legacy_material_links_verbatim": "[评审](论文/P198/评审.md) · [评分数据](论文/P198/评审.json) · [复核](论文/P198/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-222",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 68,
"source_number": 222,
"title": "Pre-training Epidemic Time Series Forecasters with Compartmental Prototypes",
"source_url": "https://doi.org/10.1145/3770855.3818928",
"raw_line": "| 52 | [P222 Pre-training Epidemic Time Series Forecasters with Compartmental Prototypes](https://doi.org/10.1145/3770855.3818928) | 51.2 | 100% | 45% | 46.0–57.5 | 7 | arXiv:2502.03393v6，本地PDF，非会议终稿 | **重要问题**：v6第8页图9将集合离散度与绝对误差的相关性称为校准，并示例±1σ、±2σ、±3σ带；所读版本未找到按名义水平汇总的区间覆盖率与平均宽度。；
**重要问题**：表2注明使用整段时间序列且没有训练、验证、测试划分；模型容量和训练方式是可能影响组件比较的竞争解释。；
**重要问题**：算法将感染组注意力矩阵的对角线置零，再类比传播矩阵；在存在群体内部传播的分组模型中，对角线可对应组内新感染。所读论文未对注意力权重与分组传播率的对应作独立验证。；
**重要问题**：附录B.6把模拟R_t写为R₀S(t)/N，§3.1同时允许β(t)=β₀[1+εcos(2πt/T+φ)]且β₀=R₀γ。在固定移除率的季节性SIR特例中，按瞬时再生数β(t)S(t)/(γN)定义，B.6公式缺少时变因子。 | [评审](论文/P222/评审.md) · [评分数据](论文/P222/评审.json) · [复核](论文/P222/复核.md) |",
"raw_cells": [
"52",
"[P222 Pre-training Epidemic Time Series Forecasters with Compartmental Prototypes](https://doi.org/10.1145/3770855.3818928)",
"51.2",
"100%",
"45%",
"46.0–57.5",
"7",
"arXiv:2502.03393v6，本地PDF，非会议终稿",
"**重要问题**：v6第8页图9将集合离散度与绝对误差的相关性称为校准，并示例±1σ、±2σ、±3σ带；所读版本未找到按名义水平汇总的区间覆盖率与平均宽度。；
**重要问题**：表2注明使用整段时间序列且没有训练、验证、测试划分；模型容量和训练方式是可能影响组件比较的竞争解释。；
**重要问题**：算法将感染组注意力矩阵的对角线置零，再类比传播矩阵；在存在群体内部传播的分组模型中，对角线可对应组内新感染。所读论文未对注意力权重与分组传播率的对应作独立验证。；
**重要问题**：附录B.6把模拟R_t写为R₀S(t)/N，§3.1同时允许β(t)=β₀[1+εcos(2πt/T+φ)]且β₀=R₀γ。在固定移除率的季节性SIR特例中，按瞬时再生数β(t)S(t)/(γN)定义，B.6公式缺少时变因子。",
"[评审](论文/P222/评审.md) · [评分数据](论文/P222/评审.json) · [复核](论文/P222/复核.md)"
],
"normalized_title": "pretrainingepidemictimeseriesforecasterswithcompartmentalprototypes",
"normalized_url": "doi.org/10.1145/3770855.3818928",
"record_type": "legacy_scored",
"legacy_rank": 52,
"legacy_score": "51.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "46.0–57.5",
"citation_verbatim": "7",
"legacy_review_version": "arXiv:2502.03393v6，本地PDF，非会议终稿",
"legacy_risks_verbatim": "**重要问题**：v6第8页图9将集合离散度与绝对误差的相关性称为校准，并示例±1σ、±2σ、±3σ带；所读版本未找到按名义水平汇总的区间覆盖率与平均宽度。；
**重要问题**：表2注明使用整段时间序列且没有训练、验证、测试划分；模型容量和训练方式是可能影响组件比较的竞争解释。；
**重要问题**：算法将感染组注意力矩阵的对角线置零，再类比传播矩阵；在存在群体内部传播的分组模型中，对角线可对应组内新感染。所读论文未对注意力权重与分组传播率的对应作独立验证。；
**重要问题**：附录B.6把模拟R_t写为R₀S(t)/N，§3.1同时允许β(t)=β₀[1+εcos(2πt/T+φ)]且β₀=R₀γ。在固定移除率的季节性SIR特例中，按瞬时再生数β(t)S(t)/(γN)定义，B.6公式缺少时变因子。",
"legacy_material_links_verbatim": "[评审](论文/P222/评审.md) · [评分数据](论文/P222/评审.json) · [复核](论文/P222/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-104",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 69,
"source_number": 104,
"title": "IN-Flow: Instance Normalization Flow for Non-stationary Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709260",
"raw_line": "| 53 | [P104 IN-Flow: Instance Normalization Flow for Non-stationary Time Series Forecasting](https://doi.org/10.1145/3690624.3709260) | 51.1 | 95% | 53% | 43.5–59.2 | 9 | arXiv:2401.16777v2，KDD 2025 格式的预印本；出版最终版未逐字比对 | **重要问题**：式 (1) 将涉及两个潜变量的条件密度连乘直接写成目标边缘密度；按一般连续密度解释缺少边缘化。在固定仿射参数时，式 (5) 对输入加常数不改变归一化结果，故单独映射丢失均值；式 (6) 依赖保存统计量，式 (8) 依赖非零缩放。；
**重要问题**：§5 的评价说明及附录 C.2 称原始数据不做归一化或缩放，§5.2 又称各骨干做 z-score 归一化并对预测反变换。可能涉及模型内处理与全局预处理的不同层次，文中未明确对应关系。；
**重要问题**：表 3 未在表题或表头注明指标；按展示值计算：NS-Informer 的 ETTm2 48 为 2.094→1.487，降幅 29.0%，表写 32.2%；CAISO 48 为 0.861→0.768，降幅 10.8%，表写 3.4%；NS-Autoformer 的 Weather 48 为 0.580→0.421，降幅 27.4%，表写 27.7%；若以展示数值之比计算并按常规就近舍入，三处均超出三位小数舍入可解释范围；论文未说明降幅是否按逐次运行比例的均值等其他方式汇总。§5.2.1 把表 1 的 Electricity 96 行数值称为 Weather 168。表 1 与表 2 的 CAISO 96 和 NordPool 96 裸 PatchTST 数字不同；表 2 标题称六个数据集，表体列出四个。；
**重要问题**：式 (4) 使用验证损失更新变换参数，附录 C.1 又使用验证集选块数，并报告了早停但未明确早停监视指标；该验证集参与了模型拟合及选择。论文另列测试集。；
**重要问题**：图 6 展示窗口均值的分布，表 5 报告 Wasserstein 距离；其计算口径未充分说明。原始与变换后坐标的单位和尺度不同，距离会随尺度变化。论文未将缩放效应与分布形状变化及预测收益充分分离。 | [评审](论文/P104/评审.md) · [评分数据](论文/P104/评审.json) · [复核](论文/P104/复核.md) |",
"raw_cells": [
"53",
"[P104 IN-Flow: Instance Normalization Flow for Non-stationary Time Series Forecasting](https://doi.org/10.1145/3690624.3709260)",
"51.1",
"95%",
"53%",
"43.5–59.2",
"9",
"arXiv:2401.16777v2，KDD 2025 格式的预印本；出版最终版未逐字比对",
"**重要问题**：式 (1) 将涉及两个潜变量的条件密度连乘直接写成目标边缘密度；按一般连续密度解释缺少边缘化。在固定仿射参数时，式 (5) 对输入加常数不改变归一化结果，故单独映射丢失均值；式 (6) 依赖保存统计量，式 (8) 依赖非零缩放。；
**重要问题**：§5 的评价说明及附录 C.2 称原始数据不做归一化或缩放，§5.2 又称各骨干做 z-score 归一化并对预测反变换。可能涉及模型内处理与全局预处理的不同层次，文中未明确对应关系。；
**重要问题**：表 3 未在表题或表头注明指标；按展示值计算：NS-Informer 的 ETTm2 48 为 2.094→1.487，降幅 29.0%，表写 32.2%；CAISO 48 为 0.861→0.768，降幅 10.8%，表写 3.4%；NS-Autoformer 的 Weather 48 为 0.580→0.421，降幅 27.4%，表写 27.7%；若以展示数值之比计算并按常规就近舍入，三处均超出三位小数舍入可解释范围；论文未说明降幅是否按逐次运行比例的均值等其他方式汇总。§5.2.1 把表 1 的 Electricity 96 行数值称为 Weather 168。表 1 与表 2 的 CAISO 96 和 NordPool 96 裸 PatchTST 数字不同；表 2 标题称六个数据集，表体列出四个。；
**重要问题**：式 (4) 使用验证损失更新变换参数，附录 C.1 又使用验证集选块数，并报告了早停但未明确早停监视指标；该验证集参与了模型拟合及选择。论文另列测试集。；
**重要问题**：图 6 展示窗口均值的分布，表 5 报告 Wasserstein 距离；其计算口径未充分说明。原始与变换后坐标的单位和尺度不同，距离会随尺度变化。论文未将缩放效应与分布形状变化及预测收益充分分离。",
"[评审](论文/P104/评审.md) · [评分数据](论文/P104/评审.json) · [复核](论文/P104/复核.md)"
],
"normalized_title": "inflowinstancenormalizationflowfornonstationarytimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709260",
"record_type": "legacy_scored",
"legacy_rank": 53,
"legacy_score": "51.1",
"legacy_coverage": "95%",
"legacy_tentative_share": "53%",
"legacy_range": "43.5–59.2",
"citation_verbatim": "9",
"legacy_review_version": "arXiv:2401.16777v2，KDD 2025 格式的预印本；出版最终版未逐字比对",
"legacy_risks_verbatim": "**重要问题**：式 (1) 将涉及两个潜变量的条件密度连乘直接写成目标边缘密度；按一般连续密度解释缺少边缘化。在固定仿射参数时，式 (5) 对输入加常数不改变归一化结果，故单独映射丢失均值；式 (6) 依赖保存统计量，式 (8) 依赖非零缩放。；
**重要问题**：§5 的评价说明及附录 C.2 称原始数据不做归一化或缩放，§5.2 又称各骨干做 z-score 归一化并对预测反变换。可能涉及模型内处理与全局预处理的不同层次，文中未明确对应关系。；
**重要问题**：表 3 未在表题或表头注明指标；按展示值计算：NS-Informer 的 ETTm2 48 为 2.094→1.487，降幅 29.0%，表写 32.2%；CAISO 48 为 0.861→0.768，降幅 10.8%，表写 3.4%；NS-Autoformer 的 Weather 48 为 0.580→0.421，降幅 27.4%，表写 27.7%；若以展示数值之比计算并按常规就近舍入，三处均超出三位小数舍入可解释范围；论文未说明降幅是否按逐次运行比例的均值等其他方式汇总。§5.2.1 把表 1 的 Electricity 96 行数值称为 Weather 168。表 1 与表 2 的 CAISO 96 和 NordPool 96 裸 PatchTST 数字不同；表 2 标题称六个数据集，表体列出四个。；
**重要问题**：式 (4) 使用验证损失更新变换参数，附录 C.1 又使用验证集选块数，并报告了早停但未明确早停监视指标；该验证集参与了模型拟合及选择。论文另列测试集。；
**重要问题**：图 6 展示窗口均值的分布，表 5 报告 Wasserstein 距离；其计算口径未充分说明。原始与变换后坐标的单位和尺度不同，距离会随尺度变化。论文未将缩放效应与分布形状变化及预测收益充分分离。",
"legacy_material_links_verbatim": "[评审](论文/P104/评审.md) · [评分数据](论文/P104/评审.json) · [复核](论文/P104/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-080",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 70,
"source_number": 80,
"title": "Disentangling Long-Short Term State Under Unknown Interventions for Online Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33717",
"raw_line": "| 54 | [P080 Disentangling Long-Short Term State Under Unknown Interventions for Online Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33717) | 51.0 | 100% | 70% | 42.7–60.0 | 4 | AAAI-25 proceedings最终出版PDF；补充证据来自独立的 arXiv:2502.12603v1 HTML | **重要问题**：正文第3页规定预测x[L+1:H]后，下一时刻即用全部真值更新并向前移动1步。若H-L>1，这与逐点到达的标签时序不同；OneNet附录C.8单列延迟反馈。本次没有审查或运行代码，表格实际反馈时点及其影响未知。；
**重要问题**：arXiv v1 PDF附录B将z^s的整段轨迹视作n_s维共享变量，并在B.19–B.27使用子视图g1的非奇异/可逆性；正文A2给出的是逐时点混合g的可逆性。上述过渡与维度对应尚未在已读证明中解释清楚。正文A1全变量密度因子化的求和上界写n_s，而附录B.30写n。；
**重要问题**：会议正文MAE表中ETTm1长度1的Online-T为0.085、小于加粗的0.187；Traffic长度48的ER为0.294、小于加粗的LSTD 0.332。预印本表6的WTH长度1中TFCL为0.169、小于加粗的LSTD 0.200。WTH长度24的OneNet MAE在P080表2为0.225，在OneNet原文表3及P080预印本三种子均值表为0.255，运行和选优方式是否相同未知。Exchange的三个MSE单项相对OneNet分别下降约23.5%、17.0%、30.6%；48步同一对数值若以LSTD为分母，OneNet误差高约44.2%，论文所称44%口径未知。；
**重要问题**：图3及附录图5比较去除组件后的柱值，但纵轴未标出指标；图4b有梯度和虚线，图注未说明干预标注来源。已读正文和附录未报告带真实潜状态的分离度量或干预检测精确率、召回率。 | [评审](论文/P080/评审.md) · [评分数据](论文/P080/评审.json) · [复核](论文/P080/复核.md) |",
"raw_cells": [
"54",
"[P080 Disentangling Long-Short Term State Under Unknown Interventions for Online Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33717)",
"51.0",
"100%",
"70%",
"42.7–60.0",
"4",
"AAAI-25 proceedings最终出版PDF；补充证据来自独立的 arXiv:2502.12603v1 HTML",
"**重要问题**：正文第3页规定预测x[L+1:H]后，下一时刻即用全部真值更新并向前移动1步。若H-L>1，这与逐点到达的标签时序不同；OneNet附录C.8单列延迟反馈。本次没有审查或运行代码，表格实际反馈时点及其影响未知。；
**重要问题**：arXiv v1 PDF附录B将z^s的整段轨迹视作n_s维共享变量，并在B.19–B.27使用子视图g1的非奇异/可逆性；正文A2给出的是逐时点混合g的可逆性。上述过渡与维度对应尚未在已读证明中解释清楚。正文A1全变量密度因子化的求和上界写n_s，而附录B.30写n。；
**重要问题**：会议正文MAE表中ETTm1长度1的Online-T为0.085、小于加粗的0.187；Traffic长度48的ER为0.294、小于加粗的LSTD 0.332。预印本表6的WTH长度1中TFCL为0.169、小于加粗的LSTD 0.200。WTH长度24的OneNet MAE在P080表2为0.225，在OneNet原文表3及P080预印本三种子均值表为0.255，运行和选优方式是否相同未知。Exchange的三个MSE单项相对OneNet分别下降约23.5%、17.0%、30.6%；48步同一对数值若以LSTD为分母，OneNet误差高约44.2%，论文所称44%口径未知。；
**重要问题**：图3及附录图5比较去除组件后的柱值，但纵轴未标出指标；图4b有梯度和虚线，图注未说明干预标注来源。已读正文和附录未报告带真实潜状态的分离度量或干预检测精确率、召回率。",
"[评审](论文/P080/评审.md) · [评分数据](论文/P080/评审.json) · [复核](论文/P080/复核.md)"
],
"normalized_title": "disentanglinglongshorttermstateunderunknowninterventionsforonlinetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33717",
"record_type": "legacy_scored",
"legacy_rank": 54,
"legacy_score": "51.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "70%",
"legacy_range": "42.7–60.0",
"citation_verbatim": "4",
"legacy_review_version": "AAAI-25 proceedings最终出版PDF；补充证据来自独立的 arXiv:2502.12603v1 HTML",
"legacy_risks_verbatim": "**重要问题**：正文第3页规定预测x[L+1:H]后，下一时刻即用全部真值更新并向前移动1步。若H-L>1，这与逐点到达的标签时序不同；OneNet附录C.8单列延迟反馈。本次没有审查或运行代码，表格实际反馈时点及其影响未知。；
**重要问题**：arXiv v1 PDF附录B将z^s的整段轨迹视作n_s维共享变量，并在B.19–B.27使用子视图g1的非奇异/可逆性；正文A2给出的是逐时点混合g的可逆性。上述过渡与维度对应尚未在已读证明中解释清楚。正文A1全变量密度因子化的求和上界写n_s，而附录B.30写n。；
**重要问题**：会议正文MAE表中ETTm1长度1的Online-T为0.085、小于加粗的0.187；Traffic长度48的ER为0.294、小于加粗的LSTD 0.332。预印本表6的WTH长度1中TFCL为0.169、小于加粗的LSTD 0.200。WTH长度24的OneNet MAE在P080表2为0.225，在OneNet原文表3及P080预印本三种子均值表为0.255，运行和选优方式是否相同未知。Exchange的三个MSE单项相对OneNet分别下降约23.5%、17.0%、30.6%；48步同一对数值若以LSTD为分母，OneNet误差高约44.2%，论文所称44%口径未知。；
**重要问题**：图3及附录图5比较去除组件后的柱值，但纵轴未标出指标；图4b有梯度和虚线，图注未说明干预标注来源。已读正文和附录未报告带真实潜状态的分离度量或干预检测精确率、召回率。",
"legacy_material_links_verbatim": "[评审](论文/P080/评审.md) · [评分数据](论文/P080/评审.json) · [复核](论文/P080/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-129",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 71,
"source_number": 129,
"title": "Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737081",
"raw_line": "| 55 | [P129 Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting](https://doi.org/10.1145/3711896.3737081) | 50.8 | 95% | 58% | 40.2–60.5 | 11 | arXiv:2505.14555v2（本地PDF；与ACM正式版逐字节关系未核对） | **重要问题**：摘要称最高170倍推理加速；表1列模块7.8秒/轮与独立模型7.1–22.2分钟/轮，已读PDF未列可比的模块与独立模型推理计时或完整系统推理计时。；
**重要问题**：虽声明按时间顺序8:1:1切分，已读PDF未详细界定各测试时刻fθ和Qπ的拟合信息、在线更新、预测标签来源和滑窗边界。；
**重要问题**：固定fθ时，若潜在力模型可表示补偿项，改变显式系数可由潜在力补偿而保持相同物理残差；这一步不保证完整优化目标相同。图5为分解示例，表6为形式对照；已读PDF缺少独立守恒指标和区分组件贡献的消融。；
**重要问题**：作者称降尺度基线依赖细粒度监督，本文使用粗粒度输入；逐基线时序上下文、监督来源和调参细节未充分列出。预测基线有架构及分辨率改造，逐基线微调和增强设置仍不清楚。；
**重要问题**：正文称三个随机种子训练取均值，表2–3声称t检验p<0.01；已读PDF未给各次结果、离散程度、检验单位、自由度和时间相关处理。；
**重要问题**：表2与表5的Average涵盖八种天气变量；第4页提及对深度学习基线进行逐通道归一化，但已读PDF未明确所报各变量RMSE使用的物理单位、是否及如何还原，以及跨变量平均前采用的尺度。 | [评审](论文/P129/评审.md) · [评分数据](论文/P129/评审.json) · [复核](论文/P129/复核.md) |",
"raw_cells": [
"55",
"[P129 Physics-Guided Learning of Meteorological Dynamics for Weather Downscaling and Forecasting](https://doi.org/10.1145/3711896.3737081)",
"50.8",
"95%",
"58%",
"40.2–60.5",
"11",
"arXiv:2505.14555v2（本地PDF；与ACM正式版逐字节关系未核对）",
"**重要问题**：摘要称最高170倍推理加速；表1列模块7.8秒/轮与独立模型7.1–22.2分钟/轮，已读PDF未列可比的模块与独立模型推理计时或完整系统推理计时。；
**重要问题**：虽声明按时间顺序8:1:1切分，已读PDF未详细界定各测试时刻fθ和Qπ的拟合信息、在线更新、预测标签来源和滑窗边界。；
**重要问题**：固定fθ时，若潜在力模型可表示补偿项，改变显式系数可由潜在力补偿而保持相同物理残差；这一步不保证完整优化目标相同。图5为分解示例，表6为形式对照；已读PDF缺少独立守恒指标和区分组件贡献的消融。；
**重要问题**：作者称降尺度基线依赖细粒度监督，本文使用粗粒度输入；逐基线时序上下文、监督来源和调参细节未充分列出。预测基线有架构及分辨率改造，逐基线微调和增强设置仍不清楚。；
**重要问题**：正文称三个随机种子训练取均值，表2–3声称t检验p<0.01；已读PDF未给各次结果、离散程度、检验单位、自由度和时间相关处理。；
**重要问题**：表2与表5的Average涵盖八种天气变量；第4页提及对深度学习基线进行逐通道归一化，但已读PDF未明确所报各变量RMSE使用的物理单位、是否及如何还原，以及跨变量平均前采用的尺度。",
"[评审](论文/P129/评审.md) · [评分数据](论文/P129/评审.json) · [复核](论文/P129/复核.md)"
],
"normalized_title": "physicsguidedlearningofmeteorologicaldynamicsforweatherdownscalingandforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737081",
"record_type": "legacy_scored",
"legacy_rank": 55,
"legacy_score": "50.8",
"legacy_coverage": "95%",
"legacy_tentative_share": "58%",
"legacy_range": "40.2–60.5",
"citation_verbatim": "11",
"legacy_review_version": "arXiv:2505.14555v2（本地PDF；与ACM正式版逐字节关系未核对）",
"legacy_risks_verbatim": "**重要问题**：摘要称最高170倍推理加速；表1列模块7.8秒/轮与独立模型7.1–22.2分钟/轮，已读PDF未列可比的模块与独立模型推理计时或完整系统推理计时。；
**重要问题**：虽声明按时间顺序8:1:1切分，已读PDF未详细界定各测试时刻fθ和Qπ的拟合信息、在线更新、预测标签来源和滑窗边界。；
**重要问题**：固定fθ时，若潜在力模型可表示补偿项，改变显式系数可由潜在力补偿而保持相同物理残差；这一步不保证完整优化目标相同。图5为分解示例，表6为形式对照；已读PDF缺少独立守恒指标和区分组件贡献的消融。；
**重要问题**：作者称降尺度基线依赖细粒度监督，本文使用粗粒度输入；逐基线时序上下文、监督来源和调参细节未充分列出。预测基线有架构及分辨率改造，逐基线微调和增强设置仍不清楚。；
**重要问题**：正文称三个随机种子训练取均值，表2–3声称t检验p<0.01；已读PDF未给各次结果、离散程度、检验单位、自由度和时间相关处理。；
**重要问题**：表2与表5的Average涵盖八种天气变量；第4页提及对深度学习基线进行逐通道归一化，但已读PDF未明确所报各变量RMSE使用的物理单位、是否及如何还原，以及跨变量平均前采用的尺度。",
"legacy_material_links_verbatim": "[评审](论文/P129/评审.md) · [评分数据](论文/P129/评审.json) · [复核](论文/P129/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-019",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 72,
"source_number": 19,
"title": "MARLP: Time-series Forecasting Control for Agricultural Managed Aquifer Recharge",
"source_url": "https://doi.org/10.1145/3637528.3671533",
"raw_line": "| 56 | [P019 MARLP: Time-series Forecasting Control for Agricultural Managed Aquifer Recharge](https://doi.org/10.1145/3637528.3671533) | 50.7 | 100% | 55% | 44.2–56.7 | 21 | arXiv:2407.01005v1，本地 PDF；与会议 DOI 的元数据匹配 | **重要问题**：田间 MARLP 采用 2024 年记录，参照为 2020–2023 年固定周期记录；表 2 显示不同地块面积，作者在 §6.3 脚注说没有田间 A/B。天气、土壤和年份可能影响差异。；
**重要问题**：已读正文和附录 A 未给出训练/验证/测试时间边界、窗口隔离及历史天气预报发行时间的可核查协议；README 仅解释天气文件及 720 步时间对齐。是否发生泄漏未知。；
**重要问题**：§3.3 将天气/动作、土壤水分、氧气按先验分层，用 Granger 型线性投影校正；图 12 是删模块及输入的预测消融，缺少对动作干预效应的单独识别。；
**重要问题**：安全按氧浓度阈值及其亏缺时间评价，补给目标以式(1)水量平衡定义；文章未报告与表 3 同期的独立地下含水层增量或作物产量比较。 | [评审](论文/P019/评审.md) · [评分数据](论文/P019/评审.json) · 无单独复核文件 |",
"raw_cells": [
"56",
"[P019 MARLP: Time-series Forecasting Control for Agricultural Managed Aquifer Recharge](https://doi.org/10.1145/3637528.3671533)",
"50.7",
"100%",
"55%",
"44.2–56.7",
"21",
"arXiv:2407.01005v1，本地 PDF；与会议 DOI 的元数据匹配",
"**重要问题**：田间 MARLP 采用 2024 年记录，参照为 2020–2023 年固定周期记录；表 2 显示不同地块面积，作者在 §6.3 脚注说没有田间 A/B。天气、土壤和年份可能影响差异。；
**重要问题**：已读正文和附录 A 未给出训练/验证/测试时间边界、窗口隔离及历史天气预报发行时间的可核查协议；README 仅解释天气文件及 720 步时间对齐。是否发生泄漏未知。；
**重要问题**：§3.3 将天气/动作、土壤水分、氧气按先验分层，用 Granger 型线性投影校正；图 12 是删模块及输入的预测消融，缺少对动作干预效应的单独识别。；
**重要问题**：安全按氧浓度阈值及其亏缺时间评价，补给目标以式(1)水量平衡定义；文章未报告与表 3 同期的独立地下含水层增量或作物产量比较。",
"[评审](论文/P019/评审.md) · [评分数据](论文/P019/评审.json) · 无单独复核文件"
],
"normalized_title": "marlptimeseriesforecastingcontrolforagriculturalmanagedaquiferrecharge",
"normalized_url": "doi.org/10.1145/3637528.3671533",
"record_type": "legacy_scored",
"legacy_rank": 56,
"legacy_score": "50.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "44.2–56.7",
"citation_verbatim": "21",
"legacy_review_version": "arXiv:2407.01005v1，本地 PDF；与会议 DOI 的元数据匹配",
"legacy_risks_verbatim": "**重要问题**：田间 MARLP 采用 2024 年记录，参照为 2020–2023 年固定周期记录；表 2 显示不同地块面积，作者在 §6.3 脚注说没有田间 A/B。天气、土壤和年份可能影响差异。；
**重要问题**：已读正文和附录 A 未给出训练/验证/测试时间边界、窗口隔离及历史天气预报发行时间的可核查协议；README 仅解释天气文件及 720 步时间对齐。是否发生泄漏未知。；
**重要问题**：§3.3 将天气/动作、土壤水分、氧气按先验分层，用 Granger 型线性投影校正；图 12 是删模块及输入的预测消融，缺少对动作干预效应的单独识别。；
**重要问题**：安全按氧浓度阈值及其亏缺时间评价，补给目标以式(1)水量平衡定义；文章未报告与表 3 同期的独立地下含水层增量或作物产量比较。",
"legacy_material_links_verbatim": "[评审](论文/P019/评审.md) · [评分数据](论文/P019/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-024",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 73,
"source_number": 24,
"title": "Generative Pretrained Hierarchical Transformer for Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3637528.3671855",
"raw_line": "| 57 | [P024 Generative Pretrained Hierarchical Transformer for Time Series Forecasting](https://doi.org/10.1145/3637528.3671855) | 50.7 | 100% | 20% | 48.5–53.5 | 44 | arXiv:2402.16516v2，KDD 2024 录用稿；与 ACM 终版的逐字一致性未核对 | **重要问题**：式(1)用完整输入 X 的均值和标准差归一化，而式(7)同时监督 X 内部后续块；按文面，较早位置的预测可通过整体统计量获得后续输入块的信息。最后一个输入位置对应的外部未来 Y 不在 X 中；本项未确立测试目标泄漏。；
**重要问题**：§4.4仅说明5%或10%的训练实例用于训练，未逐实验写明这项限制是否也约束混合预训练中目标数据的历史，以及各预训练基线接触的数据范围。；
**重要问题**：主表涉及不同类别及训练资源的模型；零样本表3仅含FPT、PatchTST、DLinear，明确未纳入ForecastPFN，也未纳入论文已引用的PreDcT；各留出目标对应不同的七数据集混合。 | [评审](论文/P024/评审.md) · [评分数据](论文/P024/评审.json) · 无单独复核文件 |",
"raw_cells": [
"57",
"[P024 Generative Pretrained Hierarchical Transformer for Time Series Forecasting](https://doi.org/10.1145/3637528.3671855)",
"50.7",
"100%",
"20%",
"48.5–53.5",
"44",
"arXiv:2402.16516v2，KDD 2024 录用稿；与 ACM 终版的逐字一致性未核对",
"**重要问题**：式(1)用完整输入 X 的均值和标准差归一化，而式(7)同时监督 X 内部后续块；按文面，较早位置的预测可通过整体统计量获得后续输入块的信息。最后一个输入位置对应的外部未来 Y 不在 X 中；本项未确立测试目标泄漏。；
**重要问题**：§4.4仅说明5%或10%的训练实例用于训练，未逐实验写明这项限制是否也约束混合预训练中目标数据的历史，以及各预训练基线接触的数据范围。；
**重要问题**：主表涉及不同类别及训练资源的模型；零样本表3仅含FPT、PatchTST、DLinear，明确未纳入ForecastPFN，也未纳入论文已引用的PreDcT；各留出目标对应不同的七数据集混合。",
"[评审](论文/P024/评审.md) · [评分数据](论文/P024/评审.json) · 无单独复核文件"
],
"normalized_title": "generativepretrainedhierarchicaltransformerfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3637528.3671855",
"record_type": "legacy_scored",
"legacy_rank": 57,
"legacy_score": "50.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "20%",
"legacy_range": "48.5–53.5",
"citation_verbatim": "44",
"legacy_review_version": "arXiv:2402.16516v2，KDD 2024 录用稿；与 ACM 终版的逐字一致性未核对",
"legacy_risks_verbatim": "**重要问题**：式(1)用完整输入 X 的均值和标准差归一化，而式(7)同时监督 X 内部后续块；按文面，较早位置的预测可通过整体统计量获得后续输入块的信息。最后一个输入位置对应的外部未来 Y 不在 X 中；本项未确立测试目标泄漏。；
**重要问题**：§4.4仅说明5%或10%的训练实例用于训练，未逐实验写明这项限制是否也约束混合预训练中目标数据的历史，以及各预训练基线接触的数据范围。；
**重要问题**：主表涉及不同类别及训练资源的模型；零样本表3仅含FPT、PatchTST、DLinear，明确未纳入ForecastPFN，也未纳入论文已引用的PreDcT；各留出目标对应不同的七数据集混合。",
"legacy_material_links_verbatim": "[评审](论文/P024/评审.md) · [评分数据](论文/P024/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-093",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 74,
"source_number": 93,
"title": "Sequence Complementor: Complementing Transformers for Time Series Forecasting with Learnable Sequences",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33747",
"raw_line": "| 58 | [P093 Sequence Complementor: Complementing Transformers for Time Series Forecasting with Learnable Sequences](https://ojs.aaai.org/index.php/AAAI/article/view/33747) | 50.5 | 100% | 45% | 45.0–55.2 | 4 | AAAI 2025 出版社九页PDF；附录参考独立的 arXiv:2501.02735v1 | **关键问题**：按正文多维目标的联合微分熵理解时，论文式(11)在高斯条件下有直接反例；式(10)的连续熵证明及式(12)连接两种网络性能的论证存在缺口。论文未明确是否按坐标计算熵，标量目标不在该反例范围。；
**重要问题**：arXiv v1附录B把MASE缩放项写成预测区间Y的季节差异，M4原定义使用样本内历史序列；该附录属于较早预印本，无法确定会议实验采用的实现。；
**重要问题**：表征丰富度与误差关联来自不同架构和少量数据集，熵与秩代理受尺度和中心化影响；图3固定K比较有无正则，但缺少能单独识别丰富度因果作用的对照。；
**重要问题**：v1附录C称对照数值取自iTransformer及TimesNet，且作者称曾运行对照代码并认为可复现；正文说所有结果为5次平均，各对照的逐次结果、调参和配对种子范围不够清楚。 | [评审](论文/P093/评审.md) · [评分数据](论文/P093/评审.json) · [复核](论文/P093/复核.md) |",
"raw_cells": [
"58",
"[P093 Sequence Complementor: Complementing Transformers for Time Series Forecasting with Learnable Sequences](https://ojs.aaai.org/index.php/AAAI/article/view/33747)",
"50.5",
"100%",
"45%",
"45.0–55.2",
"4",
"AAAI 2025 出版社九页PDF；附录参考独立的 arXiv:2501.02735v1",
"**关键问题**：按正文多维目标的联合微分熵理解时，论文式(11)在高斯条件下有直接反例；式(10)的连续熵证明及式(12)连接两种网络性能的论证存在缺口。论文未明确是否按坐标计算熵，标量目标不在该反例范围。；
**重要问题**：arXiv v1附录B把MASE缩放项写成预测区间Y的季节差异，M4原定义使用样本内历史序列；该附录属于较早预印本，无法确定会议实验采用的实现。；
**重要问题**：表征丰富度与误差关联来自不同架构和少量数据集，熵与秩代理受尺度和中心化影响；图3固定K比较有无正则，但缺少能单独识别丰富度因果作用的对照。；
**重要问题**：v1附录C称对照数值取自iTransformer及TimesNet，且作者称曾运行对照代码并认为可复现；正文说所有结果为5次平均，各对照的逐次结果、调参和配对种子范围不够清楚。",
"[评审](论文/P093/评审.md) · [评分数据](论文/P093/评审.json) · [复核](论文/P093/复核.md)"
],
"normalized_title": "sequencecomplementorcomplementingtransformersfortimeseriesforecastingwithlearnablesequences",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33747",
"record_type": "legacy_scored",
"legacy_rank": 58,
"legacy_score": "50.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "45.0–55.2",
"citation_verbatim": "4",
"legacy_review_version": "AAAI 2025 出版社九页PDF；附录参考独立的 arXiv:2501.02735v1",
"legacy_risks_verbatim": "**关键问题**：按正文多维目标的联合微分熵理解时，论文式(11)在高斯条件下有直接反例；式(10)的连续熵证明及式(12)连接两种网络性能的论证存在缺口。论文未明确是否按坐标计算熵，标量目标不在该反例范围。；
**重要问题**：arXiv v1附录B把MASE缩放项写成预测区间Y的季节差异，M4原定义使用样本内历史序列；该附录属于较早预印本，无法确定会议实验采用的实现。；
**重要问题**：表征丰富度与误差关联来自不同架构和少量数据集，熵与秩代理受尺度和中心化影响；图3固定K比较有无正则，但缺少能单独识别丰富度因果作用的对照。；
**重要问题**：v1附录C称对照数值取自iTransformer及TimesNet，且作者称曾运行对照代码并认为可复现；正文说所有结果为5次平均，各对照的逐次结果、调参和配对种子范围不够清楚。",
"legacy_material_links_verbatim": "[评审](论文/P093/评审.md) · [评分数据](论文/P093/评审.json) · [复核](论文/P093/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-008",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 75,
"source_number": 8,
"title": "Self-Supervised Learning of Time Series Representation via Diffusion Process and Imputation-Interpolation-Forecasting Mask",
"source_url": "https://doi.org/10.1145/3637528.3671673",
"raw_line": "| 59 | [P008 Self-Supervised Learning of Time Series Representation via Diffusion Process and Imputation-Interpolation-Forecasting Mask](https://doi.org/10.1145/3637528.3671673) | 50.2 | 100% | 30% | 47.2–53.7 | 31 | arXiv:2405.05959v2，KDD 2024 相关预印本；完整 23 页 PDF | **重要问题**：Solar 的 TSDE 和 CSDI 行使用重新划分，其他方法的表格数值来自不同来源或已有结果，跨组结果的直接排名缺少相同数据划分。作者称旧实现测试集与训练集相同，本评审未核查该实现。；
**重要问题**：附录 C.2 的算法 3 将噪声采样和 x_t 重置写在逆扩散循环内，未明确传递上一步 x_{t-1}，循环条件又写为 t=1；这与正文逐步逆扩散的描述不一致。；
**重要问题**：附录式 (29) 在论文的一基索引下从 L−L2 求和至 L，共 L2+1 项，与文字所说最后 L2 项不同；式 (43) 代入 t=1、T=50 给出约 0.0000923，与文字设定的 β1=0.0001 不同。；
**重要问题**：异常检测采用区间调整：连续异常段命中一个点后将整段计为异常；论文未另报未经调整的逐点 F1 或实时检测延迟。按真值边界实施该调整会用到异常段边界信息。；
**重要问题**：概率预测与推理计时的 CSDI 基线把 linear Transformer 换成 PyTorch TransformerEncoder；相对速度对应这一改写对照，作者原版实现的加速比未知。；
**重要问题**：确定性预测每个窗口只运行一次；更长预测窗口因显存不足未纳入；三次重复结果对应的随机种子或具体随机来源未清楚交代。；
**重要问题**：多任务通用性与机制归因缺少完整对照：不同数据集分别训练，比较中未出现此前的通用表征方法 TS2Vec，IIF 与纯随机掩码在 PhysioNet 上的消融差异很小且略不利于 IIF。 | [评审](论文/P008/评审.md) · [评分数据](论文/P008/评审.json) · 无单独复核文件 |",
"raw_cells": [
"59",
"[P008 Self-Supervised Learning of Time Series Representation via Diffusion Process and Imputation-Interpolation-Forecasting Mask](https://doi.org/10.1145/3637528.3671673)",
"50.2",
"100%",
"30%",
"47.2–53.7",
"31",
"arXiv:2405.05959v2，KDD 2024 相关预印本；完整 23 页 PDF",
"**重要问题**：Solar 的 TSDE 和 CSDI 行使用重新划分，其他方法的表格数值来自不同来源或已有结果，跨组结果的直接排名缺少相同数据划分。作者称旧实现测试集与训练集相同，本评审未核查该实现。；
**重要问题**：附录 C.2 的算法 3 将噪声采样和 x_t 重置写在逆扩散循环内，未明确传递上一步 x_{t-1}，循环条件又写为 t=1；这与正文逐步逆扩散的描述不一致。；
**重要问题**：附录式 (29) 在论文的一基索引下从 L−L2 求和至 L，共 L2+1 项，与文字所说最后 L2 项不同；式 (43) 代入 t=1、T=50 给出约 0.0000923，与文字设定的 β1=0.0001 不同。；
**重要问题**：异常检测采用区间调整：连续异常段命中一个点后将整段计为异常；论文未另报未经调整的逐点 F1 或实时检测延迟。按真值边界实施该调整会用到异常段边界信息。；
**重要问题**：概率预测与推理计时的 CSDI 基线把 linear Transformer 换成 PyTorch TransformerEncoder；相对速度对应这一改写对照，作者原版实现的加速比未知。；
**重要问题**：确定性预测每个窗口只运行一次；更长预测窗口因显存不足未纳入；三次重复结果对应的随机种子或具体随机来源未清楚交代。；
**重要问题**：多任务通用性与机制归因缺少完整对照：不同数据集分别训练，比较中未出现此前的通用表征方法 TS2Vec，IIF 与纯随机掩码在 PhysioNet 上的消融差异很小且略不利于 IIF。",
"[评审](论文/P008/评审.md) · [评分数据](论文/P008/评审.json) · 无单独复核文件"
],
"normalized_title": "selfsupervisedlearningoftimeseriesrepresentationviadiffusionprocessandimputationinterpolationforecastingmask",
"normalized_url": "doi.org/10.1145/3637528.3671673",
"record_type": "legacy_scored",
"legacy_rank": 59,
"legacy_score": "50.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "30%",
"legacy_range": "47.2–53.7",
"citation_verbatim": "31",
"legacy_review_version": "arXiv:2405.05959v2，KDD 2024 相关预印本；完整 23 页 PDF",
"legacy_risks_verbatim": "**重要问题**：Solar 的 TSDE 和 CSDI 行使用重新划分，其他方法的表格数值来自不同来源或已有结果，跨组结果的直接排名缺少相同数据划分。作者称旧实现测试集与训练集相同，本评审未核查该实现。；
**重要问题**：附录 C.2 的算法 3 将噪声采样和 x_t 重置写在逆扩散循环内，未明确传递上一步 x_{t-1}，循环条件又写为 t=1；这与正文逐步逆扩散的描述不一致。；
**重要问题**：附录式 (29) 在论文的一基索引下从 L−L2 求和至 L，共 L2+1 项，与文字所说最后 L2 项不同；式 (43) 代入 t=1、T=50 给出约 0.0000923，与文字设定的 β1=0.0001 不同。；
**重要问题**：异常检测采用区间调整：连续异常段命中一个点后将整段计为异常；论文未另报未经调整的逐点 F1 或实时检测延迟。按真值边界实施该调整会用到异常段边界信息。；
**重要问题**：概率预测与推理计时的 CSDI 基线把 linear Transformer 换成 PyTorch TransformerEncoder；相对速度对应这一改写对照，作者原版实现的加速比未知。；
**重要问题**：确定性预测每个窗口只运行一次；更长预测窗口因显存不足未纳入；三次重复结果对应的随机种子或具体随机来源未清楚交代。；
**重要问题**：多任务通用性与机制归因缺少完整对照：不同数据集分别训练，比较中未出现此前的通用表征方法 TS2Vec，IIF 与纯随机掩码在 PhysioNet 上的消融差异很小且略不利于 IIF。",
"legacy_material_links_verbatim": "[评审](论文/P008/评审.md) · [评分数据](论文/P008/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-087",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 76,
"source_number": 87,
"title": "TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33989",
"raw_line": "| 60 | [P087 TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents](https://ojs.aaai.org/index.php/AAAI/article/view/33989) | 50.2 | 100% | 50% | 43.2–55.7 | 23 | AAAI 2025 会议论文；本地PDF 9页，会议页18082–18090；附读作者仓库补充材料5页 | **重要问题**：正文和已获得的附录只说明6:2:2划分，未说明时间顺序、窗口边界重叠控制及卫生标签均值阈值的估计范围；这影响未来预测设定的判断，不能据此认定发生泄漏。；
**重要问题**：F1类别平均方式、多类AUROC构造及LLM离散回答对应的AUROC打分方式未明确；五次运行没有逐项波动，单尾t检验的单位和配对方式未交代。；
**重要问题**：摘要及图1给出平均28.75%提升，图示为Predict对比，但未明确足以重算的具体基线集合与聚合公式；表2的每个数据集最优其他方法口径下，本评审计算的均值约3.88%。；
**重要问题**：所用历史天气与金融数据的时段早于或跨越所用GPT-4版本出现时间，论文及附录未报告预训练数据与测试样本重合排查；模型可能接触过相关公开信息。 | [评审](论文/P087/评审.md) · [评分数据](论文/P087/评审.json) · 无单独复核文件 |",
"raw_cells": [
"60",
"[P087 TimeCAP: Learning to Contextualize, Augment, and Predict Time Series Events with Large Language Model Agents](https://ojs.aaai.org/index.php/AAAI/article/view/33989)",
"50.2",
"100%",
"50%",
"43.2–55.7",
"23",
"AAAI 2025 会议论文；本地PDF 9页，会议页18082–18090；附读作者仓库补充材料5页",
"**重要问题**：正文和已获得的附录只说明6:2:2划分，未说明时间顺序、窗口边界重叠控制及卫生标签均值阈值的估计范围；这影响未来预测设定的判断，不能据此认定发生泄漏。；
**重要问题**：F1类别平均方式、多类AUROC构造及LLM离散回答对应的AUROC打分方式未明确；五次运行没有逐项波动，单尾t检验的单位和配对方式未交代。；
**重要问题**：摘要及图1给出平均28.75%提升，图示为Predict对比，但未明确足以重算的具体基线集合与聚合公式；表2的每个数据集最优其他方法口径下，本评审计算的均值约3.88%。；
**重要问题**：所用历史天气与金融数据的时段早于或跨越所用GPT-4版本出现时间，论文及附录未报告预训练数据与测试样本重合排查；模型可能接触过相关公开信息。",
"[评审](论文/P087/评审.md) · [评分数据](论文/P087/评审.json) · 无单独复核文件"
],
"normalized_title": "timecaplearningtocontextualizeaugmentandpredicttimeserieseventswithlargelanguagemodelagents",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33989",
"record_type": "legacy_scored",
"legacy_rank": 60,
"legacy_score": "50.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "43.2–55.7",
"citation_verbatim": "23",
"legacy_review_version": "AAAI 2025 会议论文；本地PDF 9页，会议页18082–18090；附读作者仓库补充材料5页",
"legacy_risks_verbatim": "**重要问题**：正文和已获得的附录只说明6:2:2划分，未说明时间顺序、窗口边界重叠控制及卫生标签均值阈值的估计范围；这影响未来预测设定的判断，不能据此认定发生泄漏。；
**重要问题**：F1类别平均方式、多类AUROC构造及LLM离散回答对应的AUROC打分方式未明确；五次运行没有逐项波动，单尾t检验的单位和配对方式未交代。；
**重要问题**：摘要及图1给出平均28.75%提升，图示为Predict对比，但未明确足以重算的具体基线集合与聚合公式；表2的每个数据集最优其他方法口径下，本评审计算的均值约3.88%。；
**重要问题**：所用历史天气与金融数据的时段早于或跨越所用GPT-4版本出现时间，论文及附录未报告预训练数据与测试样本重合排查；模型可能接触过相关公开信息。",
"legacy_material_links_verbatim": "[评审](论文/P087/评审.md) · [评分数据](论文/P087/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-101",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 77,
"source_number": 101,
"title": "SDE: A Simplified and Disentangled Dependency Encoding Framework for State Space Models in Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737119",
"raw_line": "| 61 | [P101 SDE: A Simplified and Disentangled Dependency Encoding Framework for State Space Models in Time Series Forecasting](https://doi.org/10.1145/3711896.3737119) | 50.2 | 100% | 40% | 46.2–54.5 | 10 | arXiv:2408.12068v3；PDF首页标注KDD 2025，未取得ACM出版页正文核对版本差异 | **关键问题**：针对v3定理1书面条件构造N=2、J=2、D=1且ε=0的反例：U在{0,1}均匀、V在{0,1,2}均匀且相互独立，E¹=(U,0)、E²=(V,0)、目标Y=1[V=2]；O为保留顺序的恒等映射、S和Z恒为0，Φ逐元素为ReLU(x−1)，Ψ逐位置相加，FFN对输入特征求和。两项定理假设均满足，但Fd¹=(U+V,0)、Fttv¹=(Y,0)，H(Y|Fd¹)=1/3 bit>H(Y|Fttv¹)=0。论文未写出排除这些映射的限制。附录式(9)的条件互信息替换与式(10)–(11)的FFN拆分也缺少一般成立的条件。；
**重要问题**：§6.1称所有实验使用单张A40，表7却给Crossformer在T=720的GPU显存69771 MiB；NVIDIA给单卡48 GiB=49152 MiB。论文未解释此行与硬件条件的对应关系。；
**重要问题**：2024年公开的S-Mamba及MambaTS针对同类预测任务分别采用Mamba变量编码和随机变量排序训练，v3表4未列这两项直接前作；因此同类模型相对优势缺口明确。；
**重要问题**：v3主表无多随机种子区间；v2附录K表12报告五种子，Weather、预测192步MSE为0.271±0.002，而v2附录L.2表14及v3表8分别列0.214；v2表12的±统计定义和与主表配置的对应关系未交代充分。 | [评审](论文/P101/评审.md) · [评分数据](论文/P101/评审.json) · [复核](论文/P101/复核.md) |",
"raw_cells": [
"61",
"[P101 SDE: A Simplified and Disentangled Dependency Encoding Framework for State Space Models in Time Series Forecasting](https://doi.org/10.1145/3711896.3737119)",
"50.2",
"100%",
"40%",
"46.2–54.5",
"10",
"arXiv:2408.12068v3；PDF首页标注KDD 2025，未取得ACM出版页正文核对版本差异",
"**关键问题**：针对v3定理1书面条件构造N=2、J=2、D=1且ε=0的反例：U在{0,1}均匀、V在{0,1,2}均匀且相互独立，E¹=(U,0)、E²=(V,0)、目标Y=1[V=2]；O为保留顺序的恒等映射、S和Z恒为0，Φ逐元素为ReLU(x−1)，Ψ逐位置相加，FFN对输入特征求和。两项定理假设均满足，但Fd¹=(U+V,0)、Fttv¹=(Y,0)，H(Y|Fd¹)=1/3 bit>H(Y|Fttv¹)=0。论文未写出排除这些映射的限制。附录式(9)的条件互信息替换与式(10)–(11)的FFN拆分也缺少一般成立的条件。；
**重要问题**：§6.1称所有实验使用单张A40，表7却给Crossformer在T=720的GPU显存69771 MiB；NVIDIA给单卡48 GiB=49152 MiB。论文未解释此行与硬件条件的对应关系。；
**重要问题**：2024年公开的S-Mamba及MambaTS针对同类预测任务分别采用Mamba变量编码和随机变量排序训练，v3表4未列这两项直接前作；因此同类模型相对优势缺口明确。；
**重要问题**：v3主表无多随机种子区间；v2附录K表12报告五种子，Weather、预测192步MSE为0.271±0.002，而v2附录L.2表14及v3表8分别列0.214；v2表12的±统计定义和与主表配置的对应关系未交代充分。",
"[评审](论文/P101/评审.md) · [评分数据](论文/P101/评审.json) · [复核](论文/P101/复核.md)"
],
"normalized_title": "sdeasimplifiedanddisentangleddependencyencodingframeworkforstatespacemodelsintimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737119",
"record_type": "legacy_scored",
"legacy_rank": 61,
"legacy_score": "50.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "46.2–54.5",
"citation_verbatim": "10",
"legacy_review_version": "arXiv:2408.12068v3；PDF首页标注KDD 2025，未取得ACM出版页正文核对版本差异",
"legacy_risks_verbatim": "**关键问题**：针对v3定理1书面条件构造N=2、J=2、D=1且ε=0的反例：U在{0,1}均匀、V在{0,1,2}均匀且相互独立，E¹=(U,0)、E²=(V,0)、目标Y=1[V=2]；O为保留顺序的恒等映射、S和Z恒为0，Φ逐元素为ReLU(x−1)，Ψ逐位置相加，FFN对输入特征求和。两项定理假设均满足，但Fd¹=(U+V,0)、Fttv¹=(Y,0)，H(Y|Fd¹)=1/3 bit>H(Y|Fttv¹)=0。论文未写出排除这些映射的限制。附录式(9)的条件互信息替换与式(10)–(11)的FFN拆分也缺少一般成立的条件。；
**重要问题**：§6.1称所有实验使用单张A40，表7却给Crossformer在T=720的GPU显存69771 MiB；NVIDIA给单卡48 GiB=49152 MiB。论文未解释此行与硬件条件的对应关系。；
**重要问题**：2024年公开的S-Mamba及MambaTS针对同类预测任务分别采用Mamba变量编码和随机变量排序训练，v3表4未列这两项直接前作；因此同类模型相对优势缺口明确。；
**重要问题**：v3主表无多随机种子区间；v2附录K表12报告五种子，Weather、预测192步MSE为0.271±0.002，而v2附录L.2表14及v3表8分别列0.214；v2表12的±统计定义和与主表配置的对应关系未交代充分。",
"legacy_material_links_verbatim": "[评审](论文/P101/评审.md) · [评分数据](论文/P101/评审.json) · [复核](论文/P101/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-123",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 78,
"source_number": 123,
"title": "Less but More: Linear Adaptive Graph Learning Empowering Spatiotemporal Forecasting",
"source_url": "https://papers.neurips.cc/paper_files/paper/2025/hash/54c9bfb0885ae07f23607f617ab64c2b-Abstract-Conference.html",
"raw_line": "| 62 | [P123 Less but More: Linear Adaptive Graph Learning Empowering Spatiotemporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/54c9bfb0885ae07f23607f617ab64c2b-Abstract-Conference.html) | 50.2 | 100% | 50% | 44.7–55.2 | 5 | NeurIPS 2025 官方会议论文 PDF，具体版本日期未知 | **关键问题**：附录 A.1 引理2式36称负相似条件下 E[ρ] 有随 N 增大趋于1的下界；按排除 j 的定义，在独立高斯假设、dG=1、σ=0.5、N=716下，该条件期望至多约0.96384，低于式36的下界约0.98130。式47遗漏共享 E1 行产生的非零协方差，式40、52等近似缺少可支持严格不等式的误差界。；
**重要问题**：摘要声称单步MAGE与传统图多步卷积数学等价；已查会议论文中附录A.2讨论因子化、A.3讨论秩上界，D.6图6比较MAGE内部步数，未找到该跨模型等价证明。；
**重要问题**：式10为共同标量权重、简单因子化专家的秩上界；式11、18引入差分图及节点级稀疏权重。每节点激活数 K 不能直接套入式10，秩上界变大本身也不能确保实际秩增加或更优预测。；
**重要问题**：第25页 UrbanEV 的247节点、5分钟、8640步与同页1682区域及第26页表4的1682节点、1小时不一致；北京微博年份及米兰节点数文字与表4亦冲突。；
**重要问题**：§5.5称GBA上 D²STGNN训练慢960倍，表2对应每轮 5527/57≈96.96 倍；表只列训练每轮时长而文中另称推理加速，不同模型的最大批大小不同，BigST和GSNet未列入效率表。；
**重要问题**：表1、5、6仅给五次独立运行均值，未给这些主表的逐次结果、标准差、置信区间或显著性检验；KnowAir MAE 15.36 对 15.39 等差异较小。 | [评审](论文/P123/评审.md) · [评分数据](论文/P123/评审.json) · [复核](论文/P123/复核.md) |",
"raw_cells": [
"62",
"[P123 Less but More: Linear Adaptive Graph Learning Empowering Spatiotemporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/54c9bfb0885ae07f23607f617ab64c2b-Abstract-Conference.html)",
"50.2",
"100%",
"50%",
"44.7–55.2",
"5",
"NeurIPS 2025 官方会议论文 PDF，具体版本日期未知",
"**关键问题**：附录 A.1 引理2式36称负相似条件下 E[ρ] 有随 N 增大趋于1的下界；按排除 j 的定义，在独立高斯假设、dG=1、σ=0.5、N=716下，该条件期望至多约0.96384，低于式36的下界约0.98130。式47遗漏共享 E1 行产生的非零协方差，式40、52等近似缺少可支持严格不等式的误差界。；
**重要问题**：摘要声称单步MAGE与传统图多步卷积数学等价；已查会议论文中附录A.2讨论因子化、A.3讨论秩上界，D.6图6比较MAGE内部步数，未找到该跨模型等价证明。；
**重要问题**：式10为共同标量权重、简单因子化专家的秩上界；式11、18引入差分图及节点级稀疏权重。每节点激活数 K 不能直接套入式10，秩上界变大本身也不能确保实际秩增加或更优预测。；
**重要问题**：第25页 UrbanEV 的247节点、5分钟、8640步与同页1682区域及第26页表4的1682节点、1小时不一致；北京微博年份及米兰节点数文字与表4亦冲突。；
**重要问题**：§5.5称GBA上 D²STGNN训练慢960倍，表2对应每轮 5527/57≈96.96 倍；表只列训练每轮时长而文中另称推理加速，不同模型的最大批大小不同，BigST和GSNet未列入效率表。；
**重要问题**：表1、5、6仅给五次独立运行均值，未给这些主表的逐次结果、标准差、置信区间或显著性检验；KnowAir MAE 15.36 对 15.39 等差异较小。",
"[评审](论文/P123/评审.md) · [评分数据](论文/P123/评审.json) · [复核](论文/P123/复核.md)"
],
"normalized_title": "lessbutmorelinearadaptivegraphlearningempoweringspatiotemporalforecasting",
"normalized_url": "papers.neurips.cc/paper_files/paper/2025/hash/54c9bfb0885ae07f23607f617ab64c2b-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 62,
"legacy_score": "50.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "44.7–55.2",
"citation_verbatim": "5",
"legacy_review_version": "NeurIPS 2025 官方会议论文 PDF，具体版本日期未知",
"legacy_risks_verbatim": "**关键问题**：附录 A.1 引理2式36称负相似条件下 E[ρ] 有随 N 增大趋于1的下界；按排除 j 的定义，在独立高斯假设、dG=1、σ=0.5、N=716下，该条件期望至多约0.96384，低于式36的下界约0.98130。式47遗漏共享 E1 行产生的非零协方差，式40、52等近似缺少可支持严格不等式的误差界。；
**重要问题**：摘要声称单步MAGE与传统图多步卷积数学等价；已查会议论文中附录A.2讨论因子化、A.3讨论秩上界，D.6图6比较MAGE内部步数，未找到该跨模型等价证明。；
**重要问题**：式10为共同标量权重、简单因子化专家的秩上界；式11、18引入差分图及节点级稀疏权重。每节点激活数 K 不能直接套入式10，秩上界变大本身也不能确保实际秩增加或更优预测。；
**重要问题**：第25页 UrbanEV 的247节点、5分钟、8640步与同页1682区域及第26页表4的1682节点、1小时不一致；北京微博年份及米兰节点数文字与表4亦冲突。；
**重要问题**：§5.5称GBA上 D²STGNN训练慢960倍，表2对应每轮 5527/57≈96.96 倍；表只列训练每轮时长而文中另称推理加速，不同模型的最大批大小不同，BigST和GSNet未列入效率表。；
**重要问题**：表1、5、6仅给五次独立运行均值，未给这些主表的逐次结果、标准差、置信区间或显著性检验；KnowAir MAE 15.36 对 15.39 等差异较小。",
"legacy_material_links_verbatim": "[评审](论文/P123/评审.md) · [评分数据](论文/P123/评审.json) · [复核](论文/P123/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-137",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 79,
"source_number": 137,
"title": "ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33310",
"raw_line": "| 63 | [P137 ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data](https://ojs.aaai.org/index.php/AAAI/article/view/33310) | 50.2 | 100% | 55% | 43.7–57.2 | 6 | AAAI 2025 会议论文PDF；辅读 arXiv:2412.10912v2 扩展版附录 | **重要问题**：式(5)将cosine定义为余弦相似度且取正号；式(9)–(10)将其加入梯度下降目标，单看该项的梯度方向与正文最大化相似度的说明冲突，其他损失也影响总更新。g被定义为生成的时序，与潜向量做余弦的维度映射亦未说明。；
**重要问题**：式(8)写为sigmoid(exp((P−ε)/φ))，有限实数输入的结果均大于0.5；若直接用于伯努利连边概率且无额外筛边，期望连边比例超过一半，与高度稀疏的叙述不符；式(7)使用P而非明确使用变换后的P帽。；
**重要问题**：Problem 2规定训练时多数节点时序不可用并从训练终点之后开始预测；式(1)及扩展版算法1推理要求最近窗口节点特征，边界时刻目标节点输入的取得方式未写明。；
**重要问题**：论文引用但未将2023年FRIGATE列为基线；扩展版中骨干STGCN与独立STGCN基线采用不同代码入口，具体结构和调参是否一致未核实；TransGTR额外使用目标节点数据。；
**重要问题**：验证局限于交通PeMS、STGCN骨干和点预测；三个随机种子及±含义、目标节点选择波动和统一调参预算的细节不足。 | [评审](论文/P137/评审.md) · [评分数据](论文/P137/评审.json) · [复核](论文/P137/复核.md) |",
"raw_cells": [
"63",
"[P137 ST-FiT: Inductive Spatial-Temporal Forecasting with Limited Training Data](https://ojs.aaai.org/index.php/AAAI/article/view/33310)",
"50.2",
"100%",
"55%",
"43.7–57.2",
"6",
"AAAI 2025 会议论文PDF；辅读 arXiv:2412.10912v2 扩展版附录",
"**重要问题**：式(5)将cosine定义为余弦相似度且取正号；式(9)–(10)将其加入梯度下降目标，单看该项的梯度方向与正文最大化相似度的说明冲突，其他损失也影响总更新。g被定义为生成的时序，与潜向量做余弦的维度映射亦未说明。；
**重要问题**：式(8)写为sigmoid(exp((P−ε)/φ))，有限实数输入的结果均大于0.5；若直接用于伯努利连边概率且无额外筛边，期望连边比例超过一半，与高度稀疏的叙述不符；式(7)使用P而非明确使用变换后的P帽。；
**重要问题**：Problem 2规定训练时多数节点时序不可用并从训练终点之后开始预测；式(1)及扩展版算法1推理要求最近窗口节点特征，边界时刻目标节点输入的取得方式未写明。；
**重要问题**：论文引用但未将2023年FRIGATE列为基线；扩展版中骨干STGCN与独立STGCN基线采用不同代码入口，具体结构和调参是否一致未核实；TransGTR额外使用目标节点数据。；
**重要问题**：验证局限于交通PeMS、STGCN骨干和点预测；三个随机种子及±含义、目标节点选择波动和统一调参预算的细节不足。",
"[评审](论文/P137/评审.md) · [评分数据](论文/P137/评审.json) · [复核](论文/P137/复核.md)"
],
"normalized_title": "stfitinductivespatialtemporalforecastingwithlimitedtrainingdata",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33310",
"record_type": "legacy_scored",
"legacy_rank": 63,
"legacy_score": "50.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "43.7–57.2",
"citation_verbatim": "6",
"legacy_review_version": "AAAI 2025 会议论文PDF；辅读 arXiv:2412.10912v2 扩展版附录",
"legacy_risks_verbatim": "**重要问题**：式(5)将cosine定义为余弦相似度且取正号；式(9)–(10)将其加入梯度下降目标，单看该项的梯度方向与正文最大化相似度的说明冲突，其他损失也影响总更新。g被定义为生成的时序，与潜向量做余弦的维度映射亦未说明。；
**重要问题**：式(8)写为sigmoid(exp((P−ε)/φ))，有限实数输入的结果均大于0.5；若直接用于伯努利连边概率且无额外筛边，期望连边比例超过一半，与高度稀疏的叙述不符；式(7)使用P而非明确使用变换后的P帽。；
**重要问题**：Problem 2规定训练时多数节点时序不可用并从训练终点之后开始预测；式(1)及扩展版算法1推理要求最近窗口节点特征，边界时刻目标节点输入的取得方式未写明。；
**重要问题**：论文引用但未将2023年FRIGATE列为基线；扩展版中骨干STGCN与独立STGCN基线采用不同代码入口，具体结构和调参是否一致未核实；TransGTR额外使用目标节点数据。；
**重要问题**：验证局限于交通PeMS、STGCN骨干和点预测；三个随机种子及±含义、目标节点选择波动和统一调参预算的细节不足。",
"legacy_material_links_verbatim": "[评审](论文/P137/评审.md) · [评分数据](论文/P137/评审.json) · [复核](论文/P137/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-158",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 80,
"source_number": 158,
"title": "Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33828",
"raw_line": "| 64 | [P158 Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning](https://ojs.aaai.org/index.php/AAAI/article/view/33828) | 50.0 | 95% | 53% | 42.2–58.2 | 6 | AAAI 2025 会刊正文；作者仓库附录单独标注，PDF 元数据日期为 2024-12-17，公开时间及与会刊对应关系未知 | **重要问题**：正文第 6 页以掩码推断解释去掉动量编码器后未出现显著塌缩，并称常量编码无法最小化该分支损失。对式 (2)–(5)，在 k>0、可训练 W_emb 和编码器、投影器及预测器均允许取零的条件下，令它们及动量副本输出零，损失为零且 EMA 保持该配置；正态初始化与实际训练轨迹另当考虑。；
**重要问题**：正文表 3 FEI 五个显示行的等权均值为 85.188%，表中写 85.82%；InfoTS 为 82.212%，表中写 82.32%；表 2 TS2Vec 六行等权均值 74.57%，表中写 72.66%。论文未明确 Avg. 的聚合权重。正文第 6 页 TimesURL 与 FEI 的百分比差也与表 4 对应数值不符。；
**重要问题**：会刊正文表 3 的 Gesture TimeDRL 微调准确率为 73.33%，作者仓库附录表 7 为 66.67%；正文表 6 EMG w/o emb. infer. 精确率为 57.69%，作者附录表 9 为 36.93%。arXiv v2 同一文档正文表 3、6 和附录表 14、16 也有对应差异。会刊表 4 的 C-MAPSS TimesURL MSE 为 0.1044，作者附录表 6 四项等权均值约 0.0900，平均口径未知。；
**重要问题**：作者仓库附录明确固定种子 2024；所读正文和附录未见跨种子波动、置信区间或检验；附录与会刊的具体实验对应关系未知。；
**重要问题**：正文称 TimesURL 因内存问题使用不同编码器；作者仓库附录列出的 FEI 与基线预训练轮数配置、批大小不同。该附录与会刊具体实验对应关系及实际训练成本未知；统一协议会改变部分基线使用条件。；
**重要问题**：作者附录称 EMG 来自 3 位志愿者，每位对应一个类别；正文与附录未明确所有数据的个体或原始记录分组隔离。个体身份与类别混杂影响跨人群解释，是否存在具体数据重叠未知。 | [评审](论文/P158/评审.md) · [评分数据](论文/P158/评审.json) · [复核](论文/P158/复核.md) |",
"raw_cells": [
"64",
"[P158 Frequency-Masked Embedding Inference: A Non-Contrastive Approach for Time Series Representation Learning](https://ojs.aaai.org/index.php/AAAI/article/view/33828)",
"50.0",
"95%",
"53%",
"42.2–58.2",
"6",
"AAAI 2025 会刊正文；作者仓库附录单独标注，PDF 元数据日期为 2024-12-17，公开时间及与会刊对应关系未知",
"**重要问题**：正文第 6 页以掩码推断解释去掉动量编码器后未出现显著塌缩，并称常量编码无法最小化该分支损失。对式 (2)–(5)，在 k>0、可训练 W_emb 和编码器、投影器及预测器均允许取零的条件下，令它们及动量副本输出零，损失为零且 EMA 保持该配置；正态初始化与实际训练轨迹另当考虑。；
**重要问题**：正文表 3 FEI 五个显示行的等权均值为 85.188%，表中写 85.82%；InfoTS 为 82.212%，表中写 82.32%；表 2 TS2Vec 六行等权均值 74.57%，表中写 72.66%。论文未明确 Avg. 的聚合权重。正文第 6 页 TimesURL 与 FEI 的百分比差也与表 4 对应数值不符。；
**重要问题**：会刊正文表 3 的 Gesture TimeDRL 微调准确率为 73.33%，作者仓库附录表 7 为 66.67%；正文表 6 EMG w/o emb. infer. 精确率为 57.69%，作者附录表 9 为 36.93%。arXiv v2 同一文档正文表 3、6 和附录表 14、16 也有对应差异。会刊表 4 的 C-MAPSS TimesURL MSE 为 0.1044，作者附录表 6 四项等权均值约 0.0900，平均口径未知。；
**重要问题**：作者仓库附录明确固定种子 2024；所读正文和附录未见跨种子波动、置信区间或检验；附录与会刊的具体实验对应关系未知。；
**重要问题**：正文称 TimesURL 因内存问题使用不同编码器；作者仓库附录列出的 FEI 与基线预训练轮数配置、批大小不同。该附录与会刊具体实验对应关系及实际训练成本未知；统一协议会改变部分基线使用条件。；
**重要问题**：作者附录称 EMG 来自 3 位志愿者，每位对应一个类别；正文与附录未明确所有数据的个体或原始记录分组隔离。个体身份与类别混杂影响跨人群解释，是否存在具体数据重叠未知。",
"[评审](论文/P158/评审.md) · [评分数据](论文/P158/评审.json) · [复核](论文/P158/复核.md)"
],
"normalized_title": "frequencymaskedembeddinginferenceanoncontrastiveapproachfortimeseriesrepresentationlearning",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33828",
"record_type": "legacy_scored",
"legacy_rank": 64,
"legacy_score": "50.0",
"legacy_coverage": "95%",
"legacy_tentative_share": "53%",
"legacy_range": "42.2–58.2",
"citation_verbatim": "6",
"legacy_review_version": "AAAI 2025 会刊正文；作者仓库附录单独标注，PDF 元数据日期为 2024-12-17，公开时间及与会刊对应关系未知",
"legacy_risks_verbatim": "**重要问题**：正文第 6 页以掩码推断解释去掉动量编码器后未出现显著塌缩，并称常量编码无法最小化该分支损失。对式 (2)–(5)，在 k>0、可训练 W_emb 和编码器、投影器及预测器均允许取零的条件下，令它们及动量副本输出零，损失为零且 EMA 保持该配置；正态初始化与实际训练轨迹另当考虑。；
**重要问题**：正文表 3 FEI 五个显示行的等权均值为 85.188%，表中写 85.82%；InfoTS 为 82.212%，表中写 82.32%；表 2 TS2Vec 六行等权均值 74.57%，表中写 72.66%。论文未明确 Avg. 的聚合权重。正文第 6 页 TimesURL 与 FEI 的百分比差也与表 4 对应数值不符。；
**重要问题**：会刊正文表 3 的 Gesture TimeDRL 微调准确率为 73.33%，作者仓库附录表 7 为 66.67%；正文表 6 EMG w/o emb. infer. 精确率为 57.69%，作者附录表 9 为 36.93%。arXiv v2 同一文档正文表 3、6 和附录表 14、16 也有对应差异。会刊表 4 的 C-MAPSS TimesURL MSE 为 0.1044，作者附录表 6 四项等权均值约 0.0900，平均口径未知。；
**重要问题**：作者仓库附录明确固定种子 2024；所读正文和附录未见跨种子波动、置信区间或检验；附录与会刊的具体实验对应关系未知。；
**重要问题**：正文称 TimesURL 因内存问题使用不同编码器；作者仓库附录列出的 FEI 与基线预训练轮数配置、批大小不同。该附录与会刊具体实验对应关系及实际训练成本未知；统一协议会改变部分基线使用条件。；
**重要问题**：作者附录称 EMG 来自 3 位志愿者，每位对应一个类别；正文与附录未明确所有数据的个体或原始记录分组隔离。个体身份与类别混杂影响跨人群解释，是否存在具体数据重叠未知。",
"legacy_material_links_verbatim": "[评审](论文/P158/评审.md) · [评分数据](论文/P158/评审.json) · [复核](论文/P158/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-134",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 81,
"source_number": 134,
"title": "Decomposed Spatio-Temporal Mamba for Long-Term Traffic Prediction",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33281",
"raw_line": "| 65 | [P134 Decomposed Spatio-Temporal Mamba for Long-Term Traffic Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33281) | 49.7 | 95% | 21% | 44.7–54.7 | 26 | AAAI-25 正式会议论文，Proceedings of the AAAI Conference on Artificial Intelligence 39(11), 11772–11780 | **重要问题**：全文贡献段声称五数据集持续 SOTA；表2中 PEMS04 平均 MSE/MAE 为0.100/0.209，SOR-Mamba 为0.099/0.203；Traffic 平均 MSE 均为0.402（按展示精度）。；
**重要问题**：正文未列出 DST-Mamba 与基线的训练或推理耗时、峰值显存和 FLOPs；式5图聚合与多尺度预测的复杂度条件也未展开。；
**重要问题**：正文采用时间顺序切分与 Z-score，但未明确逐数据集切分边界、标准化拟合范围和 DTW 构图范围，基线调参或重跑方式也缺少逐模型说明。；
**重要问题**：表2–3无多次运行的均值波动或误差条，部分模型在表2中仅相差0.001；图3是超参数敏感度而非随机种子波动。 | [评审](论文/P134/评审.md) · [评分数据](论文/P134/评审.json) · 无单独复核文件 |",
"raw_cells": [
"65",
"[P134 Decomposed Spatio-Temporal Mamba for Long-Term Traffic Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33281)",
"49.7",
"95%",
"21%",
"44.7–54.7",
"26",
"AAAI-25 正式会议论文，Proceedings of the AAAI Conference on Artificial Intelligence 39(11), 11772–11780",
"**重要问题**：全文贡献段声称五数据集持续 SOTA；表2中 PEMS04 平均 MSE/MAE 为0.100/0.209，SOR-Mamba 为0.099/0.203；Traffic 平均 MSE 均为0.402（按展示精度）。；
**重要问题**：正文未列出 DST-Mamba 与基线的训练或推理耗时、峰值显存和 FLOPs；式5图聚合与多尺度预测的复杂度条件也未展开。；
**重要问题**：正文采用时间顺序切分与 Z-score，但未明确逐数据集切分边界、标准化拟合范围和 DTW 构图范围，基线调参或重跑方式也缺少逐模型说明。；
**重要问题**：表2–3无多次运行的均值波动或误差条，部分模型在表2中仅相差0.001；图3是超参数敏感度而非随机种子波动。",
"[评审](论文/P134/评审.md) · [评分数据](论文/P134/评审.json) · 无单独复核文件"
],
"normalized_title": "decomposedspatiotemporalmambaforlongtermtrafficprediction",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33281",
"record_type": "legacy_scored",
"legacy_rank": 65,
"legacy_score": "49.7",
"legacy_coverage": "95%",
"legacy_tentative_share": "21%",
"legacy_range": "44.7–54.7",
"citation_verbatim": "26",
"legacy_review_version": "AAAI-25 正式会议论文，Proceedings of the AAAI Conference on Artificial Intelligence 39(11), 11772–11780",
"legacy_risks_verbatim": "**重要问题**：全文贡献段声称五数据集持续 SOTA；表2中 PEMS04 平均 MSE/MAE 为0.100/0.209，SOR-Mamba 为0.099/0.203；Traffic 平均 MSE 均为0.402（按展示精度）。；
**重要问题**：正文未列出 DST-Mamba 与基线的训练或推理耗时、峰值显存和 FLOPs；式5图聚合与多尺度预测的复杂度条件也未展开。；
**重要问题**：正文采用时间顺序切分与 Z-score，但未明确逐数据集切分边界、标准化拟合范围和 DTW 构图范围，基线调参或重跑方式也缺少逐模型说明。；
**重要问题**：表2–3无多次运行的均值波动或误差条，部分模型在表2中仅相差0.001；图3是超参数敏感度而非随机种子波动。",
"legacy_material_links_verbatim": "[评审](论文/P134/评审.md) · [评分数据](论文/P134/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-006",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 82,
"source_number": 6,
"title": "FusionSF: Fuse Heterogeneous Modalities in a Vector Quantized Framework for Robust Solar Power Forecasting",
"source_url": "https://doi.org/10.1145/3637528.3671509",
"raw_line": "| 66 | [P006 FusionSF: Fuse Heterogeneous Modalities in a Vector Quantized Framework for Robust Solar Power Forecasting](https://doi.org/10.1145/3637528.3671509) | 49.5 | 100% | 50% | 44.0–55.2 | 10 | arXiv:2402.05823v1（2024-02-08）；会议论文 KDD 2024 的出版文本未读取 | **重要问题**：表 2 的 All、Easy、Hard 括号分别为 25210、18014、7196，原表未明确数字单位；PV-LLM 表 1 将其相同数字称为样本数，但两篇是否采用相同统计口径未核实。若括号是同一全集与互斥子集的样本数，且每个样本的 24 个输出点等权，FusionSF 分组 MAE 加权为 0.042018，而 All 为 0.04020；Crossformer 对应 0.068401 与 0.06599，均超出五位小数舍入误差。；
**重要问题**：表 3 的三模态未见站点测试仍列有目标站历史 TS；作者据此讨论新站历史不足和冷启动，表内实际检验训练阶段未见站点，未检验推理时零历史功率。；
**重要问题**：论文说明 NWP 用于预报及 6:2:2 划分，但未给出是否按时间顺序切分、各预测样本使用的 NWP 签发批次、精确切分日期及 NWP 特征归一化的拟合区间；信息时点与隔离条件待核实。；
**重要问题**：VQ 与无 VQ 的 MAE 差异及作者报告的潜变量分布 KL 差异同时出现，KL 具体估计方法未交代；缺少控制网络容量、随机波动和实际输入噪声的证据，量化压缩或正则化也可能起作用。；
**重要问题**：论文将 SolarTCN+ERA5 的 MAE 0.0422 称为随天气预报改善可达到的理论上界；表 5 No.2–3 同用 SolarTCN，分别输入 NWP 和 ERA5，而 No.1 的 FusionSF 与之在骨干及输入上均不同。ERA5 为网格再分析，存在空间分辨率和观测误差；论文未建立该值的数学上界。；
**重要问题**：作者自报在中国三省的 300 多座电站部署，并称相对旧系统准确率提升 1.5%；论文未定义该准确率或提供部署评估期、逐站结果与完整比较条件。它与 CL01 所据表 5 的 MAE 相对降幅采用不同口径。 | [评审](论文/P006/评审.md) · [评分数据](论文/P006/评审.json) · [复核](论文/P006/复核.md) |",
"raw_cells": [
"66",
"[P006 FusionSF: Fuse Heterogeneous Modalities in a Vector Quantized Framework for Robust Solar Power Forecasting](https://doi.org/10.1145/3637528.3671509)",
"49.5",
"100%",
"50%",
"44.0–55.2",
"10",
"arXiv:2402.05823v1（2024-02-08）；会议论文 KDD 2024 的出版文本未读取",
"**重要问题**：表 2 的 All、Easy、Hard 括号分别为 25210、18014、7196，原表未明确数字单位；PV-LLM 表 1 将其相同数字称为样本数，但两篇是否采用相同统计口径未核实。若括号是同一全集与互斥子集的样本数，且每个样本的 24 个输出点等权，FusionSF 分组 MAE 加权为 0.042018，而 All 为 0.04020；Crossformer 对应 0.068401 与 0.06599，均超出五位小数舍入误差。；
**重要问题**：表 3 的三模态未见站点测试仍列有目标站历史 TS；作者据此讨论新站历史不足和冷启动，表内实际检验训练阶段未见站点，未检验推理时零历史功率。；
**重要问题**：论文说明 NWP 用于预报及 6:2:2 划分，但未给出是否按时间顺序切分、各预测样本使用的 NWP 签发批次、精确切分日期及 NWP 特征归一化的拟合区间；信息时点与隔离条件待核实。；
**重要问题**：VQ 与无 VQ 的 MAE 差异及作者报告的潜变量分布 KL 差异同时出现，KL 具体估计方法未交代；缺少控制网络容量、随机波动和实际输入噪声的证据，量化压缩或正则化也可能起作用。；
**重要问题**：论文将 SolarTCN+ERA5 的 MAE 0.0422 称为随天气预报改善可达到的理论上界；表 5 No.2–3 同用 SolarTCN，分别输入 NWP 和 ERA5，而 No.1 的 FusionSF 与之在骨干及输入上均不同。ERA5 为网格再分析，存在空间分辨率和观测误差；论文未建立该值的数学上界。；
**重要问题**：作者自报在中国三省的 300 多座电站部署，并称相对旧系统准确率提升 1.5%；论文未定义该准确率或提供部署评估期、逐站结果与完整比较条件。它与 CL01 所据表 5 的 MAE 相对降幅采用不同口径。",
"[评审](论文/P006/评审.md) · [评分数据](论文/P006/评审.json) · [复核](论文/P006/复核.md)"
],
"normalized_title": "fusionsffuseheterogeneousmodalitiesinavectorquantizedframeworkforrobustsolarpowerforecasting",
"normalized_url": "doi.org/10.1145/3637528.3671509",
"record_type": "legacy_scored",
"legacy_rank": 66,
"legacy_score": "49.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "44.0–55.2",
"citation_verbatim": "10",
"legacy_review_version": "arXiv:2402.05823v1（2024-02-08）；会议论文 KDD 2024 的出版文本未读取",
"legacy_risks_verbatim": "**重要问题**：表 2 的 All、Easy、Hard 括号分别为 25210、18014、7196，原表未明确数字单位；PV-LLM 表 1 将其相同数字称为样本数，但两篇是否采用相同统计口径未核实。若括号是同一全集与互斥子集的样本数，且每个样本的 24 个输出点等权，FusionSF 分组 MAE 加权为 0.042018，而 All 为 0.04020；Crossformer 对应 0.068401 与 0.06599，均超出五位小数舍入误差。；
**重要问题**：表 3 的三模态未见站点测试仍列有目标站历史 TS；作者据此讨论新站历史不足和冷启动，表内实际检验训练阶段未见站点，未检验推理时零历史功率。；
**重要问题**：论文说明 NWP 用于预报及 6:2:2 划分，但未给出是否按时间顺序切分、各预测样本使用的 NWP 签发批次、精确切分日期及 NWP 特征归一化的拟合区间；信息时点与隔离条件待核实。；
**重要问题**：VQ 与无 VQ 的 MAE 差异及作者报告的潜变量分布 KL 差异同时出现，KL 具体估计方法未交代；缺少控制网络容量、随机波动和实际输入噪声的证据，量化压缩或正则化也可能起作用。；
**重要问题**：论文将 SolarTCN+ERA5 的 MAE 0.0422 称为随天气预报改善可达到的理论上界；表 5 No.2–3 同用 SolarTCN，分别输入 NWP 和 ERA5，而 No.1 的 FusionSF 与之在骨干及输入上均不同。ERA5 为网格再分析，存在空间分辨率和观测误差；论文未建立该值的数学上界。；
**重要问题**：作者自报在中国三省的 300 多座电站部署，并称相对旧系统准确率提升 1.5%；论文未定义该准确率或提供部署评估期、逐站结果与完整比较条件。它与 CL01 所据表 5 的 MAE 相对降幅采用不同口径。",
"legacy_material_links_verbatim": "[评审](论文/P006/评审.md) · [评分数据](论文/P006/评审.json) · [复核](论文/P006/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-083",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 83,
"source_number": 83,
"title": "VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33951",
"raw_line": "| 67 | [P083 VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33951) | 49.5 | 100% | 20% | 47.2–51.5 | 6 | AAAI-25 Proceedings 39(17), pp. 17742–17749，官方 PDF；补充分析另注明 arXiv v1 | **关键问题**：相同有序频率哈希既不记录幅值也不记录相位；上述周期序列在相同哈希下非主频能量均为零、相关系数为零。v1 附录 C 的重建引入原始幅值和相位，未从哈希推导出来。；
**重要问题**：论文已读结果表没有等保留量随机变量抽样对照；iTransformer 2024 已报告每批随机选变量训练。；
**重要问题**：正文 §4.5 称 k 增大时令牌删减率增加；已实际查看图 4(c)，固定 gs 时图示曲线随 k 增大而降低。；
**重要问题**：表 1 标 FlashAttention，表 5 标 Flashformer；前作表 2 将 Flashformer 定义为使用 FlashAttention 的 Transformer，且相应结果与本文部分基线数值相同。精确注意力内核与完整模型比较应区分，性能差异原因未知。 | [评审](论文/P083/评审.md) · [评分数据](论文/P083/评审.json) · [复核](论文/P083/复核.md) |",
"raw_cells": [
"67",
"[P083 VarDrop: Enhancing Training Efficiency by Reducing Variate Redundancy in Periodic Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33951)",
"49.5",
"100%",
"20%",
"47.2–51.5",
"6",
"AAAI-25 Proceedings 39(17), pp. 17742–17749，官方 PDF；补充分析另注明 arXiv v1",
"**关键问题**：相同有序频率哈希既不记录幅值也不记录相位；上述周期序列在相同哈希下非主频能量均为零、相关系数为零。v1 附录 C 的重建引入原始幅值和相位，未从哈希推导出来。；
**重要问题**：论文已读结果表没有等保留量随机变量抽样对照；iTransformer 2024 已报告每批随机选变量训练。；
**重要问题**：正文 §4.5 称 k 增大时令牌删减率增加；已实际查看图 4(c)，固定 gs 时图示曲线随 k 增大而降低。；
**重要问题**：表 1 标 FlashAttention，表 5 标 Flashformer；前作表 2 将 Flashformer 定义为使用 FlashAttention 的 Transformer，且相应结果与本文部分基线数值相同。精确注意力内核与完整模型比较应区分，性能差异原因未知。",
"[评审](论文/P083/评审.md) · [评分数据](论文/P083/评审.json) · [复核](论文/P083/复核.md)"
],
"normalized_title": "vardropenhancingtrainingefficiencybyreducingvariateredundancyinperiodictimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33951",
"record_type": "legacy_scored",
"legacy_rank": 67,
"legacy_score": "49.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "20%",
"legacy_range": "47.2–51.5",
"citation_verbatim": "6",
"legacy_review_version": "AAAI-25 Proceedings 39(17), pp. 17742–17749，官方 PDF；补充分析另注明 arXiv v1",
"legacy_risks_verbatim": "**关键问题**：相同有序频率哈希既不记录幅值也不记录相位；上述周期序列在相同哈希下非主频能量均为零、相关系数为零。v1 附录 C 的重建引入原始幅值和相位，未从哈希推导出来。；
**重要问题**：论文已读结果表没有等保留量随机变量抽样对照；iTransformer 2024 已报告每批随机选变量训练。；
**重要问题**：正文 §4.5 称 k 增大时令牌删减率增加；已实际查看图 4(c)，固定 gs 时图示曲线随 k 增大而降低。；
**重要问题**：表 1 标 FlashAttention，表 5 标 Flashformer；前作表 2 将 Flashformer 定义为使用 FlashAttention 的 Transformer，且相应结果与本文部分基线数值相同。精确注意力内核与完整模型比较应区分，性能差异原因未知。",
"legacy_material_links_verbatim": "[评审](论文/P083/评审.md) · [评分数据](论文/P083/评审.json) · [复核](论文/P083/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-047",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 84,
"source_number": 47,
"title": "Task-oriented Time Series Imputation Evaluation via Generalized Representers",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/f88264fcc54775ee1706116e90fe351a-Abstract-Conference.html",
"raw_line": "| 68 | [P047 Task-oriented Time Series Imputation Evaluation via Generalized Representers](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f88264fcc54775ee1706116e90fe351a-Abstract-Conference.html) | 49.2 | 100% | 40% | 45.0–54.0 | 4 | NeurIPS 2024 官方会议 PDF，29 页；SHA-256 5991b960620e09f32d09e1a21e52ee69066a3bd071fdef14eabfd3e731446d21 | **关键问题**：Remark 1 按论文应用所需的、对所列光滑函数统一成立的导数近似控制被解析反例击破；式 (6) 到式 (7) 用的三角不等式对论文采用的 MSE 也不成立。若误差阈值可依赖具体函数对，原文未明确该量词关系与其应用。；
**重要问题**：印刷式 (8) 与式 (16) 中没有式 (2) 的候选标签差因子；例如候选值相同时真实替换增益为零，而印刷式其他项一般可非零。实现是否遗漏未知。；
**重要问题**：附录 D.5 所印影响函数扰动目标的加减项完全相同，式 (17) 同样相减为零，与后续非零式 (18) 冲突。；
**重要问题**：附录 B 宣称方法满足效率等式；对两个训练标签的常数模型平方损失反例，真实单点增益之和为 −0.5、联合替换增益为 −1，而基准测试梯度为零使印刷式 (8) 的各近似项为零。；
**重要问题**：图 1 的逐点估计通过样本级联合替换验证；本地 PDF 未找到同预算随机替换对照，影响函数方案删除数据。表格括号统计量未明确，GEF 诊断与表 5 年份口径及数据处理细节待核。独立补充 ZIP 未读。 | [评审](论文/P047/评审.md) · [评分数据](论文/P047/评审.json) · [复核](论文/P047/复核.md) |",
"raw_cells": [
"68",
"[P047 Task-oriented Time Series Imputation Evaluation via Generalized Representers](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f88264fcc54775ee1706116e90fe351a-Abstract-Conference.html)",
"49.2",
"100%",
"40%",
"45.0–54.0",
"4",
"NeurIPS 2024 官方会议 PDF，29 页；SHA-256 5991b960620e09f32d09e1a21e52ee69066a3bd071fdef14eabfd3e731446d21",
"**关键问题**：Remark 1 按论文应用所需的、对所列光滑函数统一成立的导数近似控制被解析反例击破；式 (6) 到式 (7) 用的三角不等式对论文采用的 MSE 也不成立。若误差阈值可依赖具体函数对，原文未明确该量词关系与其应用。；
**重要问题**：印刷式 (8) 与式 (16) 中没有式 (2) 的候选标签差因子；例如候选值相同时真实替换增益为零，而印刷式其他项一般可非零。实现是否遗漏未知。；
**重要问题**：附录 D.5 所印影响函数扰动目标的加减项完全相同，式 (17) 同样相减为零，与后续非零式 (18) 冲突。；
**重要问题**：附录 B 宣称方法满足效率等式；对两个训练标签的常数模型平方损失反例，真实单点增益之和为 −0.5、联合替换增益为 −1，而基准测试梯度为零使印刷式 (8) 的各近似项为零。；
**重要问题**：图 1 的逐点估计通过样本级联合替换验证；本地 PDF 未找到同预算随机替换对照，影响函数方案删除数据。表格括号统计量未明确，GEF 诊断与表 5 年份口径及数据处理细节待核。独立补充 ZIP 未读。",
"[评审](论文/P047/评审.md) · [评分数据](论文/P047/评审.json) · [复核](论文/P047/复核.md)"
],
"normalized_title": "taskorientedtimeseriesimputationevaluationviageneralizedrepresenters",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/f88264fcc54775ee1706116e90fe351a-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 68,
"legacy_score": "49.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "45.0–54.0",
"citation_verbatim": "4",
"legacy_review_version": "NeurIPS 2024 官方会议 PDF，29 页；SHA-256 5991b960620e09f32d09e1a21e52ee69066a3bd071fdef14eabfd3e731446d21",
"legacy_risks_verbatim": "**关键问题**：Remark 1 按论文应用所需的、对所列光滑函数统一成立的导数近似控制被解析反例击破；式 (6) 到式 (7) 用的三角不等式对论文采用的 MSE 也不成立。若误差阈值可依赖具体函数对，原文未明确该量词关系与其应用。；
**重要问题**：印刷式 (8) 与式 (16) 中没有式 (2) 的候选标签差因子；例如候选值相同时真实替换增益为零，而印刷式其他项一般可非零。实现是否遗漏未知。；
**重要问题**：附录 D.5 所印影响函数扰动目标的加减项完全相同，式 (17) 同样相减为零，与后续非零式 (18) 冲突。；
**重要问题**：附录 B 宣称方法满足效率等式；对两个训练标签的常数模型平方损失反例，真实单点增益之和为 −0.5、联合替换增益为 −1，而基准测试梯度为零使印刷式 (8) 的各近似项为零。；
**重要问题**：图 1 的逐点估计通过样本级联合替换验证；本地 PDF 未找到同预算随机替换对照，影响函数方案删除数据。表格括号统计量未明确，GEF 诊断与表 5 年份口径及数据处理细节待核。独立补充 ZIP 未读。",
"legacy_material_links_verbatim": "[评审](论文/P047/评审.md) · [评分数据](论文/P047/评审.json) · [复核](论文/P047/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-103",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 85,
"source_number": 103,
"title": "WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34434",
"raw_line": "| 69 | [P103 WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34434) | 49.2 | 100% | 80% | 38.7–60.0 | 8 | AAAI-25 官方会议论文 PDF，22741–22749 | **重要问题**：按式 11 对最深层 ℓ=l_w 的细节预测器计算局部损失时分母 l_w−ℓ 为零；式 10 在 l_w=1 时分母为零。l_w=3、ℓ=2 时式 10 求和仅覆盖 i=0,1,2，图 2 示例却使用 i=3 的预测。式 7 将逆变换细节系数末项写为 D^(ℓ−1)，而由式 6 及图 2 可知，3 层回到 2 层的一次逆变换使用 A^3、D^3，其索引与该示例不一致。；
**重要问题**：正文把完整结果、训练伪代码、敏感性和效率表指向补充材料；官方页下载器未找到补充 PDF/ZIP。正文亦未见种子波动、数据划分和选模细节。；
**重要问题**：WaveForM 在 2023 年已使用多分辨率小波系数预测多变量长期序列；P103 正文列出的 14 个基线及表 1 未包含该方法。 | [评审](论文/P103/评审.md) · [评分数据](论文/P103/评审.json) · [复核](论文/P103/复核.md) |",
"raw_cells": [
"69",
"[P103 WaveletMixer: A Multi-Resolution Wavelets Based MLP-Mixer for Multivariate Long-Term Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34434)",
"49.2",
"100%",
"80%",
"38.7–60.0",
"8",
"AAAI-25 官方会议论文 PDF，22741–22749",
"**重要问题**：按式 11 对最深层 ℓ=l_w 的细节预测器计算局部损失时分母 l_w−ℓ 为零；式 10 在 l_w=1 时分母为零。l_w=3、ℓ=2 时式 10 求和仅覆盖 i=0,1,2，图 2 示例却使用 i=3 的预测。式 7 将逆变换细节系数末项写为 D^(ℓ−1)，而由式 6 及图 2 可知，3 层回到 2 层的一次逆变换使用 A^3、D^3，其索引与该示例不一致。；
**重要问题**：正文把完整结果、训练伪代码、敏感性和效率表指向补充材料；官方页下载器未找到补充 PDF/ZIP。正文亦未见种子波动、数据划分和选模细节。；
**重要问题**：WaveForM 在 2023 年已使用多分辨率小波系数预测多变量长期序列；P103 正文列出的 14 个基线及表 1 未包含该方法。",
"[评审](论文/P103/评审.md) · [评分数据](论文/P103/评审.json) · [复核](论文/P103/复核.md)"
],
"normalized_title": "waveletmixeramultiresolutionwaveletsbasedmlpmixerformultivariatelongtermtimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34434",
"record_type": "legacy_scored",
"legacy_rank": 69,
"legacy_score": "49.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "80%",
"legacy_range": "38.7–60.0",
"citation_verbatim": "8",
"legacy_review_version": "AAAI-25 官方会议论文 PDF，22741–22749",
"legacy_risks_verbatim": "**重要问题**：按式 11 对最深层 ℓ=l_w 的细节预测器计算局部损失时分母 l_w−ℓ 为零；式 10 在 l_w=1 时分母为零。l_w=3、ℓ=2 时式 10 求和仅覆盖 i=0,1,2，图 2 示例却使用 i=3 的预测。式 7 将逆变换细节系数末项写为 D^(ℓ−1)，而由式 6 及图 2 可知，3 层回到 2 层的一次逆变换使用 A^3、D^3，其索引与该示例不一致。；
**重要问题**：正文把完整结果、训练伪代码、敏感性和效率表指向补充材料；官方页下载器未找到补充 PDF/ZIP。正文亦未见种子波动、数据划分和选模细节。；
**重要问题**：WaveForM 在 2023 年已使用多分辨率小波系数预测多变量长期序列；P103 正文列出的 14 个基线及表 1 未包含该方法。",
"legacy_material_links_verbatim": "[评审](论文/P103/评审.md) · [评分数据](论文/P103/评审.json) · [复核](论文/P103/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-085",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 86,
"source_number": 85,
"title": "Unsupervised Time Series Anomaly Prediction with Importance-based Generative Contrastive Learning",
"source_url": "https://doi.org/10.1145/3711896.3737174",
"raw_line": "| 70 | [P085 Unsupervised Time Series Anomaly Prediction with Importance-based Generative Contrastive Learning](https://doi.org/10.1145/3711896.3737174) | 48.6 | 90% | 44% | 37.7–58.7 | 6 | 本地 arXiv:2410.16888v3 PDF；首页列七位作者，带 KDD 2025 引文；与出版版未逐项比对 | **重要问题**：§2.2 以能够识别前兆为定理条件；附录A用存在一个使条件期望不同的函数支撑可预测性，但未论证如何仅凭无标签数据识别真实前兆分布或学到该函数。；
**重要问题**：附录 D.2 称所有方法选择最佳阈值，但没有写清采用训练、验证还是测试标签，也未给选择规则；无标签验证设置使该问题尤其重要。；
**重要问题**：式(9)、(13)、(14)均写求和下限 i=t、上限 t−h；h>0 时按通常递增求和约定为空，正文没有定义倒序记法。；
**重要问题**：论文报告三次运行均值，并称结果显著优于次优；未在所评版本的表格和实验说明中报告离散度、显著性检验和统计单位。；
**重要问题**：FCM 的相关预测方法于 2024-10-16 首次公开，早于本文 v1 的 2024-10-22；本文比较中未列 FCM，且 FCM 的窗口与步幅设置不同。；
**重要问题**：按式(3)的标量写法，KL 项直接约束每步预测方差，均值在该项中抵消；论文缺少合成片段接近真实前兆的独立验证。；
**重要问题**：第6页 §3.4 提出以重要性分数替代先进先出淘汰；第8页表7在记忆库方面仅比较有无整个记忆库，没有单独比较重要性淘汰与先进先出等保留策略。 | [评审](论文/P085/评审.md) · [评分数据](论文/P085/评审.json) · [复核](论文/P085/复核.md) |",
"raw_cells": [
"70",
"[P085 Unsupervised Time Series Anomaly Prediction with Importance-based Generative Contrastive Learning](https://doi.org/10.1145/3711896.3737174)",
"48.6",
"90%",
"44%",
"37.7–58.7",
"6",
"本地 arXiv:2410.16888v3 PDF；首页列七位作者，带 KDD 2025 引文；与出版版未逐项比对",
"**重要问题**：§2.2 以能够识别前兆为定理条件；附录A用存在一个使条件期望不同的函数支撑可预测性，但未论证如何仅凭无标签数据识别真实前兆分布或学到该函数。；
**重要问题**：附录 D.2 称所有方法选择最佳阈值，但没有写清采用训练、验证还是测试标签，也未给选择规则；无标签验证设置使该问题尤其重要。；
**重要问题**：式(9)、(13)、(14)均写求和下限 i=t、上限 t−h；h>0 时按通常递增求和约定为空，正文没有定义倒序记法。；
**重要问题**：论文报告三次运行均值，并称结果显著优于次优；未在所评版本的表格和实验说明中报告离散度、显著性检验和统计单位。；
**重要问题**：FCM 的相关预测方法于 2024-10-16 首次公开，早于本文 v1 的 2024-10-22；本文比较中未列 FCM，且 FCM 的窗口与步幅设置不同。；
**重要问题**：按式(3)的标量写法，KL 项直接约束每步预测方差，均值在该项中抵消；论文缺少合成片段接近真实前兆的独立验证。；
**重要问题**：第6页 §3.4 提出以重要性分数替代先进先出淘汰；第8页表7在记忆库方面仅比较有无整个记忆库，没有单独比较重要性淘汰与先进先出等保留策略。",
"[评审](论文/P085/评审.md) · [评分数据](论文/P085/评审.json) · [复核](论文/P085/复核.md)"
],
"normalized_title": "unsupervisedtimeseriesanomalypredictionwithimportancebasedgenerativecontrastivelearning",
"normalized_url": "doi.org/10.1145/3711896.3737174",
"record_type": "legacy_scored",
"legacy_rank": 70,
"legacy_score": "48.6",
"legacy_coverage": "90%",
"legacy_tentative_share": "44%",
"legacy_range": "37.7–58.7",
"citation_verbatim": "6",
"legacy_review_version": "本地 arXiv:2410.16888v3 PDF；首页列七位作者，带 KDD 2025 引文；与出版版未逐项比对",
"legacy_risks_verbatim": "**重要问题**：§2.2 以能够识别前兆为定理条件；附录A用存在一个使条件期望不同的函数支撑可预测性，但未论证如何仅凭无标签数据识别真实前兆分布或学到该函数。；
**重要问题**：附录 D.2 称所有方法选择最佳阈值，但没有写清采用训练、验证还是测试标签，也未给选择规则；无标签验证设置使该问题尤其重要。；
**重要问题**：式(9)、(13)、(14)均写求和下限 i=t、上限 t−h；h>0 时按通常递增求和约定为空，正文没有定义倒序记法。；
**重要问题**：论文报告三次运行均值，并称结果显著优于次优；未在所评版本的表格和实验说明中报告离散度、显著性检验和统计单位。；
**重要问题**：FCM 的相关预测方法于 2024-10-16 首次公开，早于本文 v1 的 2024-10-22；本文比较中未列 FCM，且 FCM 的窗口与步幅设置不同。；
**重要问题**：按式(3)的标量写法，KL 项直接约束每步预测方差，均值在该项中抵消；论文缺少合成片段接近真实前兆的独立验证。；
**重要问题**：第6页 §3.4 提出以重要性分数替代先进先出淘汰；第8页表7在记忆库方面仅比较有无整个记忆库，没有单独比较重要性淘汰与先进先出等保留策略。",
"legacy_material_links_verbatim": "[评审](论文/P085/评审.md) · [评分数据](论文/P085/评审.json) · [复核](论文/P085/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-059",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 87,
"source_number": 59,
"title": "Boosting Transferability and Discriminability for Time Series Domain Adaptation",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/b61da4f02b271cb7b5e3d538e2b78fb9-Abstract-Conference.html",
"raw_line": "| 71 | [P059 Boosting Transferability and Discriminability for Time Series Domain Adaptation](https://proceedings.neurips.cc/paper_files/paper/2024/hash/b61da4f02b271cb7b5e3d538e2b78fb9-Abstract-Conference.html) | 47.6 | 95% | 58% | 37.7–58.0 | 4 | NeurIPS 2024 官方会议录 PDF；本地 SHA-256 已与下载元数据核对 | **重要问题**：P059 表 14 的 WISDM RAINCOAT 十个逐域准确率等权平均为 85.315%，Avg 为 76.60%；ACON 同列逐域平均为 84.807%，Avg 为 84.80%。只有在逐域值及等权口径正确时，重算排名才与主表相反。表 3 的 HHAR-P RAINCOAT 逐域等权平均为 75.143%，Avg 为 74.21%，影响提升幅度的核对。前作 RAINCOAT 表 10 有相同基线数值及均值冲突。原始结果、实际测评条件及汇总口径未知。；
**重要问题**：第 6 页称时频外积空间差异降低会同时降低两个原始特征边缘差异。标量反例 f_s=1、z_s=2、f_t=2、z_t=1 中两域外积均为 2，但 f 和 z 的边缘取值都不同；平均池化取单维时仍成立。；
**重要问题**：主结果与附录 F 给出点估计，未报告这些结果的误差条、标准差或置信区间；第 23 页 checklist 对统计信息回答 Yes 并指向附录 F。当前仓库 README 称每实验使用五个随机种子，但未说明其如何聚合为论文表格及不确定性。 | [评审](论文/P059/评审.md) · [评分数据](论文/P059/评审.json) · [复核](论文/P059/复核.md) |",
"raw_cells": [
"71",
"[P059 Boosting Transferability and Discriminability for Time Series Domain Adaptation](https://proceedings.neurips.cc/paper_files/paper/2024/hash/b61da4f02b271cb7b5e3d538e2b78fb9-Abstract-Conference.html)",
"47.6",
"95%",
"58%",
"37.7–58.0",
"4",
"NeurIPS 2024 官方会议录 PDF；本地 SHA-256 已与下载元数据核对",
"**重要问题**：P059 表 14 的 WISDM RAINCOAT 十个逐域准确率等权平均为 85.315%，Avg 为 76.60%；ACON 同列逐域平均为 84.807%，Avg 为 84.80%。只有在逐域值及等权口径正确时，重算排名才与主表相反。表 3 的 HHAR-P RAINCOAT 逐域等权平均为 75.143%，Avg 为 74.21%，影响提升幅度的核对。前作 RAINCOAT 表 10 有相同基线数值及均值冲突。原始结果、实际测评条件及汇总口径未知。；
**重要问题**：第 6 页称时频外积空间差异降低会同时降低两个原始特征边缘差异。标量反例 f_s=1、z_s=2、f_t=2、z_t=1 中两域外积均为 2，但 f 和 z 的边缘取值都不同；平均池化取单维时仍成立。；
**重要问题**：主结果与附录 F 给出点估计，未报告这些结果的误差条、标准差或置信区间；第 23 页 checklist 对统计信息回答 Yes 并指向附录 F。当前仓库 README 称每实验使用五个随机种子，但未说明其如何聚合为论文表格及不确定性。",
"[评审](论文/P059/评审.md) · [评分数据](论文/P059/评审.json) · [复核](论文/P059/复核.md)"
],
"normalized_title": "boostingtransferabilityanddiscriminabilityfortimeseriesdomainadaptation",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/b61da4f02b271cb7b5e3d538e2b78fb9-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 71,
"legacy_score": "47.6",
"legacy_coverage": "95%",
"legacy_tentative_share": "58%",
"legacy_range": "37.7–58.0",
"citation_verbatim": "4",
"legacy_review_version": "NeurIPS 2024 官方会议录 PDF；本地 SHA-256 已与下载元数据核对",
"legacy_risks_verbatim": "**重要问题**：P059 表 14 的 WISDM RAINCOAT 十个逐域准确率等权平均为 85.315%，Avg 为 76.60%；ACON 同列逐域平均为 84.807%，Avg 为 84.80%。只有在逐域值及等权口径正确时，重算排名才与主表相反。表 3 的 HHAR-P RAINCOAT 逐域等权平均为 75.143%，Avg 为 74.21%，影响提升幅度的核对。前作 RAINCOAT 表 10 有相同基线数值及均值冲突。原始结果、实际测评条件及汇总口径未知。；
**重要问题**：第 6 页称时频外积空间差异降低会同时降低两个原始特征边缘差异。标量反例 f_s=1、z_s=2、f_t=2、z_t=1 中两域外积均为 2，但 f 和 z 的边缘取值都不同；平均池化取单维时仍成立。；
**重要问题**：主结果与附录 F 给出点估计，未报告这些结果的误差条、标准差或置信区间；第 23 页 checklist 对统计信息回答 Yes 并指向附录 F。当前仓库 README 称每实验使用五个随机种子，但未说明其如何聚合为论文表格及不确定性。",
"legacy_material_links_verbatim": "[评审](论文/P059/评审.md) · [评分数据](论文/P059/评审.json) · [复核](论文/P059/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-136",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 88,
"source_number": 136,
"title": "EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/31997",
"raw_line": "| 72 | [P136 EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts](https://ojs.aaai.org/index.php/AAAI/article/view/31997) | 47.2 | 100% | 50% | 41.2–54.0 | 9 | AAAI-25 会议正式 PDF，刊页 210–218；不同于 arXiv v1/v2 版本记录 | **重要问题**：论文摘要声称各时效优于 ClimaX，图 3 的 U10 在 192 小时显示 ClimaX 的 RMSE 曲线低于 EWMoE；图文反例限于该图示点。；
**重要问题**：跨模型图表未充分交代各基线的版本、完整训练配置、逐项匹配的初始化/样本、气候态和插值处理。W-MAE §5.2 的多变量 FourCastNet 对照使用第一轮训练，其小样本划分的年份描述与两年说法有歧义；EWMoE 图 2–3 的 FourCastNet 训练阶段未知。；
**重要问题**：通道在每个空间位置聚合为单个 token 后，高度位置嵌入如何逐变量/层施加未交代清楚；式 (5) 的 argmax 只统计首选专家，未计第二派发。E=20、k=2 时，存在辅助损失低于均衡值但派发高度不均的合法路由构造；f(v) 的约束及辅助损失系数也未明确。；
**重要问题**：训练资源来自不同硬件、输出变量及配置。论文没有给出等硬件的消融 FLOPs、吞吐量或峰值显存；按上限算 EWMoE 少于 432 个 3090 卡时，FourCastNet 报告为 1024 个 A100 卡时，卡时不可直接当成等量算力。；
**重要问题**：论文 ACC 仅说从训练数据历史样本计算长期均值，未说是否按年内日期、时刻构造，也未交代各模型是否使用同一气候态；WeatherBench 2 明确采用年内日期与六小时时刻对应的气候态。 | [评审](论文/P136/评审.md) · [评分数据](论文/P136/评审.json) · [复核](论文/P136/复核.md) |",
"raw_cells": [
"72",
"[P136 EWMoE: An Effective Model for Global Weather Forecasting with Mixture-of-Experts](https://ojs.aaai.org/index.php/AAAI/article/view/31997)",
"47.2",
"100%",
"50%",
"41.2–54.0",
"9",
"AAAI-25 会议正式 PDF，刊页 210–218；不同于 arXiv v1/v2 版本记录",
"**重要问题**：论文摘要声称各时效优于 ClimaX，图 3 的 U10 在 192 小时显示 ClimaX 的 RMSE 曲线低于 EWMoE；图文反例限于该图示点。；
**重要问题**：跨模型图表未充分交代各基线的版本、完整训练配置、逐项匹配的初始化/样本、气候态和插值处理。W-MAE §5.2 的多变量 FourCastNet 对照使用第一轮训练，其小样本划分的年份描述与两年说法有歧义；EWMoE 图 2–3 的 FourCastNet 训练阶段未知。；
**重要问题**：通道在每个空间位置聚合为单个 token 后，高度位置嵌入如何逐变量/层施加未交代清楚；式 (5) 的 argmax 只统计首选专家，未计第二派发。E=20、k=2 时，存在辅助损失低于均衡值但派发高度不均的合法路由构造；f(v) 的约束及辅助损失系数也未明确。；
**重要问题**：训练资源来自不同硬件、输出变量及配置。论文没有给出等硬件的消融 FLOPs、吞吐量或峰值显存；按上限算 EWMoE 少于 432 个 3090 卡时，FourCastNet 报告为 1024 个 A100 卡时，卡时不可直接当成等量算力。；
**重要问题**：论文 ACC 仅说从训练数据历史样本计算长期均值，未说是否按年内日期、时刻构造，也未交代各模型是否使用同一气候态；WeatherBench 2 明确采用年内日期与六小时时刻对应的气候态。",
"[评审](论文/P136/评审.md) · [评分数据](论文/P136/评审.json) · [复核](论文/P136/复核.md)"
],
"normalized_title": "ewmoeaneffectivemodelforglobalweatherforecastingwithmixtureofexperts",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/31997",
"record_type": "legacy_scored",
"legacy_rank": 72,
"legacy_score": "47.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "41.2–54.0",
"citation_verbatim": "9",
"legacy_review_version": "AAAI-25 会议正式 PDF，刊页 210–218；不同于 arXiv v1/v2 版本记录",
"legacy_risks_verbatim": "**重要问题**：论文摘要声称各时效优于 ClimaX，图 3 的 U10 在 192 小时显示 ClimaX 的 RMSE 曲线低于 EWMoE；图文反例限于该图示点。；
**重要问题**：跨模型图表未充分交代各基线的版本、完整训练配置、逐项匹配的初始化/样本、气候态和插值处理。W-MAE §5.2 的多变量 FourCastNet 对照使用第一轮训练，其小样本划分的年份描述与两年说法有歧义；EWMoE 图 2–3 的 FourCastNet 训练阶段未知。；
**重要问题**：通道在每个空间位置聚合为单个 token 后，高度位置嵌入如何逐变量/层施加未交代清楚；式 (5) 的 argmax 只统计首选专家，未计第二派发。E=20、k=2 时，存在辅助损失低于均衡值但派发高度不均的合法路由构造；f(v) 的约束及辅助损失系数也未明确。；
**重要问题**：训练资源来自不同硬件、输出变量及配置。论文没有给出等硬件的消融 FLOPs、吞吐量或峰值显存；按上限算 EWMoE 少于 432 个 3090 卡时，FourCastNet 报告为 1024 个 A100 卡时，卡时不可直接当成等量算力。；
**重要问题**：论文 ACC 仅说从训练数据历史样本计算长期均值，未说是否按年内日期、时刻构造，也未交代各模型是否使用同一气候态；WeatherBench 2 明确采用年内日期与六小时时刻对应的气候态。",
"legacy_material_links_verbatim": "[评审](论文/P136/评审.md) · [评分数据](论文/P136/评审.json) · [复核](论文/P136/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-042",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 89,
"source_number": 42,
"title": "Segment, Shuffle, and Stitch: A Simple Layer for Improving Time-Series Representations",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/0898f05f6c1d247be3eab8da93d33da1-Abstract-Conference.html",
"raw_line": "| 73 | [P042 Segment, Shuffle, and Stitch: A Simple Layer for Improving Time-Series Representations](https://proceedings.neurips.cc/paper_files/paper/2024/hash/0898f05f6c1d247be3eab8da93d33da1-Abstract-Conference.html) | 47.0 | 100% | 50% | 41.2–52.5 | 5 | NeurIPS 2024 主会正式论文集 PDF；28 页；具体上线日期未知 | **关键问题**：第 4 页令非零 Ω=p 并以 1/Ω 缩放为 1；固定输入片段和其他参数，且排序参数非零、互不相等时，按所写乘积和通常链式法则，p·(1/p)=1 的导数为零，固定排序区间内硬排列对 P 局部恒定；零点和排序边界另有未定义或不光滑问题。该结论针对 P 的真实梯度，不涉及输入或其他参数梯度。图 7 的参数变化未给出实际反向规则；具体实现未知。；
**重要问题**：表 1、2、3、8、A3 多处所列差值与所列原值按第 5 页公式计算不符；表 3 有负 MAE。示例：表 3 LaST 的 ETTh2、H=720 MSE 1.1952→1.7197，按原值是约 -43.88%，表内写 +11.92%；Weather、H=168 的 LaST 原始 MAE 写成 -0.2431。错误可能在原值、差值或两者，现有材料无法判定。；
**重要问题**：第 6 页称 SoftCLT 在全部 UCR 数据集上的训练损失标准差平均下降 36.66%，第 15 页也将表 A4 归于 SoftCLT；第 18 页表 A4 的标题和列名写 TS2Vec 和前 75 个数据集。按提取文本中的 75 个百分比作简单平均约 34.69%，其中 5 行为负；图像已核对标题、列名和 5 个负值，未逐格视觉核对其余数字。两处模型标注及统计范围不同，75 行均值与 36.66% 的差异不构成同条件算术反证，表 A4 的真实模型归属及完整统计明细待核。；
**重要问题**：第 10 页表 10 给出少量实例的新增参数数目，并未给出同条件端到端耗时、峰值显存或能耗；第 4 页还描述重复构造与分段数有关的中间矩阵。参数少本身无法量化推理或训练开销。；
**重要问题**：作者称两组沿用相同骨干协议，并提供 S3 超参数范围；论文未逐项列出每个数据集的时间切分、归一化、验证选择和两组搜索预算。附录 A5 也未注明预测窗口。测试集信息是否进入选择流程及不同方法的比较条件待核。 | [评审](论文/P042/评审.md) · [评分数据](论文/P042/评审.json) · [复核](论文/P042/复核.md) |",
"raw_cells": [
"73",
"[P042 Segment, Shuffle, and Stitch: A Simple Layer for Improving Time-Series Representations](https://proceedings.neurips.cc/paper_files/paper/2024/hash/0898f05f6c1d247be3eab8da93d33da1-Abstract-Conference.html)",
"47.0",
"100%",
"50%",
"41.2–52.5",
"5",
"NeurIPS 2024 主会正式论文集 PDF；28 页；具体上线日期未知",
"**关键问题**：第 4 页令非零 Ω=p 并以 1/Ω 缩放为 1；固定输入片段和其他参数，且排序参数非零、互不相等时，按所写乘积和通常链式法则，p·(1/p)=1 的导数为零，固定排序区间内硬排列对 P 局部恒定；零点和排序边界另有未定义或不光滑问题。该结论针对 P 的真实梯度，不涉及输入或其他参数梯度。图 7 的参数变化未给出实际反向规则；具体实现未知。；
**重要问题**：表 1、2、3、8、A3 多处所列差值与所列原值按第 5 页公式计算不符；表 3 有负 MAE。示例：表 3 LaST 的 ETTh2、H=720 MSE 1.1952→1.7197，按原值是约 -43.88%，表内写 +11.92%；Weather、H=168 的 LaST 原始 MAE 写成 -0.2431。错误可能在原值、差值或两者，现有材料无法判定。；
**重要问题**：第 6 页称 SoftCLT 在全部 UCR 数据集上的训练损失标准差平均下降 36.66%，第 15 页也将表 A4 归于 SoftCLT；第 18 页表 A4 的标题和列名写 TS2Vec 和前 75 个数据集。按提取文本中的 75 个百分比作简单平均约 34.69%，其中 5 行为负；图像已核对标题、列名和 5 个负值，未逐格视觉核对其余数字。两处模型标注及统计范围不同，75 行均值与 36.66% 的差异不构成同条件算术反证，表 A4 的真实模型归属及完整统计明细待核。；
**重要问题**：第 10 页表 10 给出少量实例的新增参数数目，并未给出同条件端到端耗时、峰值显存或能耗；第 4 页还描述重复构造与分段数有关的中间矩阵。参数少本身无法量化推理或训练开销。；
**重要问题**：作者称两组沿用相同骨干协议，并提供 S3 超参数范围；论文未逐项列出每个数据集的时间切分、归一化、验证选择和两组搜索预算。附录 A5 也未注明预测窗口。测试集信息是否进入选择流程及不同方法的比较条件待核。",
"[评审](论文/P042/评审.md) · [评分数据](论文/P042/评审.json) · [复核](论文/P042/复核.md)"
],
"normalized_title": "segmentshuffleandstitchasimplelayerforimprovingtimeseriesrepresentations",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/0898f05f6c1d247be3eab8da93d33da1-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 73,
"legacy_score": "47.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "41.2–52.5",
"citation_verbatim": "5",
"legacy_review_version": "NeurIPS 2024 主会正式论文集 PDF；28 页；具体上线日期未知",
"legacy_risks_verbatim": "**关键问题**：第 4 页令非零 Ω=p 并以 1/Ω 缩放为 1；固定输入片段和其他参数，且排序参数非零、互不相等时，按所写乘积和通常链式法则，p·(1/p)=1 的导数为零，固定排序区间内硬排列对 P 局部恒定；零点和排序边界另有未定义或不光滑问题。该结论针对 P 的真实梯度，不涉及输入或其他参数梯度。图 7 的参数变化未给出实际反向规则；具体实现未知。；
**重要问题**：表 1、2、3、8、A3 多处所列差值与所列原值按第 5 页公式计算不符；表 3 有负 MAE。示例：表 3 LaST 的 ETTh2、H=720 MSE 1.1952→1.7197，按原值是约 -43.88%，表内写 +11.92%；Weather、H=168 的 LaST 原始 MAE 写成 -0.2431。错误可能在原值、差值或两者，现有材料无法判定。；
**重要问题**：第 6 页称 SoftCLT 在全部 UCR 数据集上的训练损失标准差平均下降 36.66%，第 15 页也将表 A4 归于 SoftCLT；第 18 页表 A4 的标题和列名写 TS2Vec 和前 75 个数据集。按提取文本中的 75 个百分比作简单平均约 34.69%，其中 5 行为负；图像已核对标题、列名和 5 个负值，未逐格视觉核对其余数字。两处模型标注及统计范围不同，75 行均值与 36.66% 的差异不构成同条件算术反证，表 A4 的真实模型归属及完整统计明细待核。；
**重要问题**：第 10 页表 10 给出少量实例的新增参数数目，并未给出同条件端到端耗时、峰值显存或能耗；第 4 页还描述重复构造与分段数有关的中间矩阵。参数少本身无法量化推理或训练开销。；
**重要问题**：作者称两组沿用相同骨干协议，并提供 S3 超参数范围；论文未逐项列出每个数据集的时间切分、归一化、验证选择和两组搜索预算。附录 A5 也未注明预测窗口。测试集信息是否进入选择流程及不同方法的比较条件待核。",
"legacy_material_links_verbatim": "[评审](论文/P042/评审.md) · [评分数据](论文/P042/评审.json) · [复核](论文/P042/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-045",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 90,
"source_number": 45,
"title": "TEST: Text Prototype Aligned Embedding to Activate LLM's Ability for Time Series",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/a4352e2c9d93582898a2a20e1f514e8f-Abstract-Conference.html",
"raw_line": "| 74 | [P045 TEST: Text Prototype Aligned Embedding to Activate LLM's Ability for Time Series](https://proceedings.iclr.cc/paper_files/paper/2024/hash/a4352e2c9d93582898a2a20e1f514e8f-Abstract-Conference.html) | 47.0 | 100% | 35% | 43.5–50.7 | 92 | ICLR 2024 官方会议论文及 PDF 内附录；PDF 创建时间 2024-03-07，未视为发表日期 | **关键问题**：正式版式 (5) 中 max p 与 max log p 的数值被写成等号，且对数概率之和被近似为两个对数概率和的乘积，未说明成立条件和误差界；现有论证不能建立所称软提示与监督微调等价。；
**重要问题**：正式版表 S8 的 ETTh1、ETTm1、ETTm2 中 TEST 的多个均值与四个预测长度的纸面数值算术不一致；表 S7 的 Weather/TimesNet MSE 均值亦不一致。；
**重要问题**：论文没有给出针对 TEST 的随机初始化骨干、去语言模型或非文本原型的匹配对照，S13–S16 的现有消融混合多个因素。；
**重要问题**：表 S9 的 M3 与 ELECTR 数值存在胜过 TEST 的基线，且 M4 中 N-BEATS 的来源数据与其他模型不同；部分比较参数和预处理细节未完整报告。 | [评审](论文/P045/评审.md) · [评分数据](论文/P045/评审.json) · [复核](论文/P045/复核.md) |",
"raw_cells": [
"74",
"[P045 TEST: Text Prototype Aligned Embedding to Activate LLM's Ability for Time Series](https://proceedings.iclr.cc/paper_files/paper/2024/hash/a4352e2c9d93582898a2a20e1f514e8f-Abstract-Conference.html)",
"47.0",
"100%",
"35%",
"43.5–50.7",
"92",
"ICLR 2024 官方会议论文及 PDF 内附录；PDF 创建时间 2024-03-07，未视为发表日期",
"**关键问题**：正式版式 (5) 中 max p 与 max log p 的数值被写成等号，且对数概率之和被近似为两个对数概率和的乘积，未说明成立条件和误差界；现有论证不能建立所称软提示与监督微调等价。；
**重要问题**：正式版表 S8 的 ETTh1、ETTm1、ETTm2 中 TEST 的多个均值与四个预测长度的纸面数值算术不一致；表 S7 的 Weather/TimesNet MSE 均值亦不一致。；
**重要问题**：论文没有给出针对 TEST 的随机初始化骨干、去语言模型或非文本原型的匹配对照，S13–S16 的现有消融混合多个因素。；
**重要问题**：表 S9 的 M3 与 ELECTR 数值存在胜过 TEST 的基线，且 M4 中 N-BEATS 的来源数据与其他模型不同；部分比较参数和预处理细节未完整报告。",
"[评审](论文/P045/评审.md) · [评分数据](论文/P045/评审.json) · [复核](论文/P045/复核.md)"
],
"normalized_title": "testtextprototypealignedembeddingtoactivatellmsabilityfortimeseries",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/a4352e2c9d93582898a2a20e1f514e8f-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 74,
"legacy_score": "47.0",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "43.5–50.7",
"citation_verbatim": "92",
"legacy_review_version": "ICLR 2024 官方会议论文及 PDF 内附录；PDF 创建时间 2024-03-07，未视为发表日期",
"legacy_risks_verbatim": "**关键问题**：正式版式 (5) 中 max p 与 max log p 的数值被写成等号，且对数概率之和被近似为两个对数概率和的乘积，未说明成立条件和误差界；现有论证不能建立所称软提示与监督微调等价。；
**重要问题**：正式版表 S8 的 ETTh1、ETTm1、ETTm2 中 TEST 的多个均值与四个预测长度的纸面数值算术不一致；表 S7 的 Weather/TimesNet MSE 均值亦不一致。；
**重要问题**：论文没有给出针对 TEST 的随机初始化骨干、去语言模型或非文本原型的匹配对照，S13–S16 的现有消融混合多个因素。；
**重要问题**：表 S9 的 M3 与 ELECTR 数值存在胜过 TEST 的基线，且 M4 中 N-BEATS 的来源数据与其他模型不同；部分比较参数和预处理细节未完整报告。",
"legacy_material_links_verbatim": "[评审](论文/P045/评审.md) · [评分数据](论文/P045/评审.json) · [复核](论文/P045/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-112",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 91,
"source_number": 112,
"title": "Exploiting Language Power for Time Series Forecasting with Exogenous Variables",
"source_url": "https://doi.org/10.1145/3696410.3714793",
"raw_line": "| 75 | [P112 Exploiting Language Power for Time Series Forecasting with Exogenous Variables](https://doi.org/10.1145/3696410.3714793) | 46.8 | 95% | 74% | 37.7–58.5 | 7 | 作者主页的 WWW 2025 会议版式 PDF，10 页，含 DOI；与 ACM 发布文件逐字节一致性未知 | **重要问题**：会议版表 4 标注 AVG 为五数据集平均。ExoLLM MAE 的各行显示值均值为 0.2476，AVG 写 0.251；GPT4TS 的 MSE/MAE 重算 0.4068/0.3570，AVG 写 0.325/0.326；TimeLLM 重算 0.3696/0.3374，写 0.338/0.378；LLM4TS 重算 0.3964/0.3600，写 0.399/0.408。这七个 AVG 值与所显示的五行均值之差均超出三位小数舍入可解释的范围。；
**重要问题**：会议版表 3 Traffic/iTransformer 显示 MSE 0.161、MAE 0.412，表注称数值按预测长度平均。若两项来自同一误差、尺度及相同汇总权重，MSE 应不小于 MAE²；0.412²=0.169744>0.161，即使考虑三位小数舍入亦有 0.4115²>0.1615。共同口径及差异原因尚未核实。；
**重要问题**：移除 MGP、MTI 或 DT²Attention 同时改变输入和结构。匿名稿声称试过多个预训练语言模型；所读材料未提供等容量非语言骨干、随机预训练或语义错配等对照，语言预训练知识和伪相关变化未被独立分离。；
**重要问题**：匿名稿附录 A.1 的网页提取文本中，表 1 将 Weather 目标写成 CO2-Contamination，同节正文写 Wet Bulb；TimeXer v4 的表 6 与正文分别使用 CO2-Concentration 与 Wet Bulb。匿名稿表格未视觉核对，提取错误的可能性保留；会议版正文未指定该目标。；
**重要问题**：匿名稿附录 D 以 GPT-2 训练语料截止时间、数据集发布日期及模型问答作为无泄漏依据；这些信息未交代窗口提示生成、预处理拟合和跨数据集切分的具体边界。 | [评审](论文/P112/评审.md) · [评分数据](论文/P112/评审.json) · [复核](论文/P112/复核.md) |",
"raw_cells": [
"75",
"[P112 Exploiting Language Power for Time Series Forecasting with Exogenous Variables](https://doi.org/10.1145/3696410.3714793)",
"46.8",
"95%",
"74%",
"37.7–58.5",
"7",
"作者主页的 WWW 2025 会议版式 PDF，10 页，含 DOI；与 ACM 发布文件逐字节一致性未知",
"**重要问题**：会议版表 4 标注 AVG 为五数据集平均。ExoLLM MAE 的各行显示值均值为 0.2476，AVG 写 0.251；GPT4TS 的 MSE/MAE 重算 0.4068/0.3570，AVG 写 0.325/0.326；TimeLLM 重算 0.3696/0.3374，写 0.338/0.378；LLM4TS 重算 0.3964/0.3600，写 0.399/0.408。这七个 AVG 值与所显示的五行均值之差均超出三位小数舍入可解释的范围。；
**重要问题**：会议版表 3 Traffic/iTransformer 显示 MSE 0.161、MAE 0.412，表注称数值按预测长度平均。若两项来自同一误差、尺度及相同汇总权重，MSE 应不小于 MAE²；0.412²=0.169744>0.161，即使考虑三位小数舍入亦有 0.4115²>0.1615。共同口径及差异原因尚未核实。；
**重要问题**：移除 MGP、MTI 或 DT²Attention 同时改变输入和结构。匿名稿声称试过多个预训练语言模型；所读材料未提供等容量非语言骨干、随机预训练或语义错配等对照，语言预训练知识和伪相关变化未被独立分离。；
**重要问题**：匿名稿附录 A.1 的网页提取文本中，表 1 将 Weather 目标写成 CO2-Contamination，同节正文写 Wet Bulb；TimeXer v4 的表 6 与正文分别使用 CO2-Concentration 与 Wet Bulb。匿名稿表格未视觉核对，提取错误的可能性保留；会议版正文未指定该目标。；
**重要问题**：匿名稿附录 D 以 GPT-2 训练语料截止时间、数据集发布日期及模型问答作为无泄漏依据；这些信息未交代窗口提示生成、预处理拟合和跨数据集切分的具体边界。",
"[评审](论文/P112/评审.md) · [评分数据](论文/P112/评审.json) · [复核](论文/P112/复核.md)"
],
"normalized_title": "exploitinglanguagepowerfortimeseriesforecastingwithexogenousvariables",
"normalized_url": "doi.org/10.1145/3696410.3714793",
"record_type": "legacy_scored",
"legacy_rank": 75,
"legacy_score": "46.8",
"legacy_coverage": "95%",
"legacy_tentative_share": "74%",
"legacy_range": "37.7–58.5",
"citation_verbatim": "7",
"legacy_review_version": "作者主页的 WWW 2025 会议版式 PDF，10 页，含 DOI；与 ACM 发布文件逐字节一致性未知",
"legacy_risks_verbatim": "**重要问题**：会议版表 4 标注 AVG 为五数据集平均。ExoLLM MAE 的各行显示值均值为 0.2476，AVG 写 0.251；GPT4TS 的 MSE/MAE 重算 0.4068/0.3570，AVG 写 0.325/0.326；TimeLLM 重算 0.3696/0.3374，写 0.338/0.378；LLM4TS 重算 0.3964/0.3600，写 0.399/0.408。这七个 AVG 值与所显示的五行均值之差均超出三位小数舍入可解释的范围。；
**重要问题**：会议版表 3 Traffic/iTransformer 显示 MSE 0.161、MAE 0.412，表注称数值按预测长度平均。若两项来自同一误差、尺度及相同汇总权重，MSE 应不小于 MAE²；0.412²=0.169744>0.161，即使考虑三位小数舍入亦有 0.4115²>0.1615。共同口径及差异原因尚未核实。；
**重要问题**：移除 MGP、MTI 或 DT²Attention 同时改变输入和结构。匿名稿声称试过多个预训练语言模型；所读材料未提供等容量非语言骨干、随机预训练或语义错配等对照，语言预训练知识和伪相关变化未被独立分离。；
**重要问题**：匿名稿附录 A.1 的网页提取文本中，表 1 将 Weather 目标写成 CO2-Contamination，同节正文写 Wet Bulb；TimeXer v4 的表 6 与正文分别使用 CO2-Concentration 与 Wet Bulb。匿名稿表格未视觉核对，提取错误的可能性保留；会议版正文未指定该目标。；
**重要问题**：匿名稿附录 D 以 GPT-2 训练语料截止时间、数据集发布日期及模型问答作为无泄漏依据；这些信息未交代窗口提示生成、预处理拟合和跨数据集切分的具体边界。",
"legacy_material_links_verbatim": "[评审](论文/P112/评审.md) · [评分数据](论文/P112/评审.json) · [复核](论文/P112/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-113",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 92,
"source_number": 113,
"title": "Utilizing Strategic Pre-training to Reduce Overfitting: Baguan – A Pre-trained Weather Forecasting Model",
"source_url": "https://doi.org/10.1145/3711896.3737178",
"raw_line": "| 76 | [P113 Utilizing Strategic Pre-training to Reduce Overfitting: Baguan – A Pre-trained Weather Forecasting Model](https://doi.org/10.1145/3711896.3737178) | 46.7 | 100% | 50% | 40.5–52.5 | 6 | arXiv:2505.13873v1，作者预印本；不冒称会议终稿 | **关键问题**：按第18页上下文，预训练岭回归闭式解给出变换特征M*x上的系数；第19页及式(8)却将它直接与原空间真参数w*比较。复核构造了符合文中λ_k=R/√k、输入范数≤R、K=1≪d及w*=P_Kw*等假设的分布，使M*第一特征值为1/2、岭系数趋于2w*，故参数误差趋于1而非0。第18页未限定子空间的正最小特征值界与投影改写也存在问题；若最小特征值意指投影子空间，应明确该限制。 第19页还把幂律λ_k=R/√k代入为λ_K=R²/K及不同的迹表达式，符号之间存在不一致。；
**重要问题**：第11页式(11)将分母写成同一求和内预测异常平方与真实异常平方的乘积；同一格点两个等权时刻的预测与真实异常均为(1,-1)时，该式给出√2，而常规相关系数为1，且该例的真值时间均值为0。式(12)使用整段测试集真值均值作气候态；代码如何实现及各模型是否用同一计算口径未知。；
**重要问题**：本文§4.1称训练1979–2015；Guo等称S2S训练1979–2014、测试2015–2018，并使用Baguan预训练参数。若参数来自本文相同训练期的检查点，则2015年重叠；该检查点训练期未知。Guo的1.40625°模型为8层，本文表4为16层，本文A.7另有8层配置，仍无法识别是否为同一检查点。Guo等使用2002–2021气候态；其模型和ERA5参考使用ERA5气候态，ECMWF使用自身气候态。；
**重要问题**：全球主比较限2018年Pangu/IFS；作者说明未比较FuXi、FengWu、GraphCast；区域模型额外使用高分辨率区域分析输入。论文未见重复种子、跨年不确定性或等总训练成本对照。；
**重要问题**：图4按单一查询Token的注意力分数展示累计占比，图11展示选定层的注意力图，各图色标不同；所示量与协方差特征谱及降低过拟合之间的因果联系未充分建立。 | [评审](论文/P113/评审.md) · [评分数据](论文/P113/评审.json) · [复核](论文/P113/复核.md) |",
"raw_cells": [
"76",
"[P113 Utilizing Strategic Pre-training to Reduce Overfitting: Baguan – A Pre-trained Weather Forecasting Model](https://doi.org/10.1145/3711896.3737178)",
"46.7",
"100%",
"50%",
"40.5–52.5",
"6",
"arXiv:2505.13873v1，作者预印本；不冒称会议终稿",
"**关键问题**：按第18页上下文，预训练岭回归闭式解给出变换特征M*x上的系数；第19页及式(8)却将它直接与原空间真参数w*比较。复核构造了符合文中λ_k=R/√k、输入范数≤R、K=1≪d及w*=P_Kw*等假设的分布，使M*第一特征值为1/2、岭系数趋于2w*，故参数误差趋于1而非0。第18页未限定子空间的正最小特征值界与投影改写也存在问题；若最小特征值意指投影子空间，应明确该限制。 第19页还把幂律λ_k=R/√k代入为λ_K=R²/K及不同的迹表达式，符号之间存在不一致。；
**重要问题**：第11页式(11)将分母写成同一求和内预测异常平方与真实异常平方的乘积；同一格点两个等权时刻的预测与真实异常均为(1,-1)时，该式给出√2，而常规相关系数为1，且该例的真值时间均值为0。式(12)使用整段测试集真值均值作气候态；代码如何实现及各模型是否用同一计算口径未知。；
**重要问题**：本文§4.1称训练1979–2015；Guo等称S2S训练1979–2014、测试2015–2018，并使用Baguan预训练参数。若参数来自本文相同训练期的检查点，则2015年重叠；该检查点训练期未知。Guo的1.40625°模型为8层，本文表4为16层，本文A.7另有8层配置，仍无法识别是否为同一检查点。Guo等使用2002–2021气候态；其模型和ERA5参考使用ERA5气候态，ECMWF使用自身气候态。；
**重要问题**：全球主比较限2018年Pangu/IFS；作者说明未比较FuXi、FengWu、GraphCast；区域模型额外使用高分辨率区域分析输入。论文未见重复种子、跨年不确定性或等总训练成本对照。；
**重要问题**：图4按单一查询Token的注意力分数展示累计占比，图11展示选定层的注意力图，各图色标不同；所示量与协方差特征谱及降低过拟合之间的因果联系未充分建立。",
"[评审](论文/P113/评审.md) · [评分数据](论文/P113/评审.json) · [复核](论文/P113/复核.md)"
],
"normalized_title": "utilizingstrategicpretrainingtoreduceoverfittingbaguanapretrainedweatherforecastingmodel",
"normalized_url": "doi.org/10.1145/3711896.3737178",
"record_type": "legacy_scored",
"legacy_rank": 76,
"legacy_score": "46.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "40.5–52.5",
"citation_verbatim": "6",
"legacy_review_version": "arXiv:2505.13873v1，作者预印本；不冒称会议终稿",
"legacy_risks_verbatim": "**关键问题**：按第18页上下文，预训练岭回归闭式解给出变换特征M*x上的系数；第19页及式(8)却将它直接与原空间真参数w*比较。复核构造了符合文中λ_k=R/√k、输入范数≤R、K=1≪d及w*=P_Kw*等假设的分布，使M*第一特征值为1/2、岭系数趋于2w*，故参数误差趋于1而非0。第18页未限定子空间的正最小特征值界与投影改写也存在问题；若最小特征值意指投影子空间，应明确该限制。 第19页还把幂律λ_k=R/√k代入为λ_K=R²/K及不同的迹表达式，符号之间存在不一致。；
**重要问题**：第11页式(11)将分母写成同一求和内预测异常平方与真实异常平方的乘积；同一格点两个等权时刻的预测与真实异常均为(1,-1)时，该式给出√2，而常规相关系数为1，且该例的真值时间均值为0。式(12)使用整段测试集真值均值作气候态；代码如何实现及各模型是否用同一计算口径未知。；
**重要问题**：本文§4.1称训练1979–2015；Guo等称S2S训练1979–2014、测试2015–2018，并使用Baguan预训练参数。若参数来自本文相同训练期的检查点，则2015年重叠；该检查点训练期未知。Guo的1.40625°模型为8层，本文表4为16层，本文A.7另有8层配置，仍无法识别是否为同一检查点。Guo等使用2002–2021气候态；其模型和ERA5参考使用ERA5气候态，ECMWF使用自身气候态。；
**重要问题**：全球主比较限2018年Pangu/IFS；作者说明未比较FuXi、FengWu、GraphCast；区域模型额外使用高分辨率区域分析输入。论文未见重复种子、跨年不确定性或等总训练成本对照。；
**重要问题**：图4按单一查询Token的注意力分数展示累计占比，图11展示选定层的注意力图，各图色标不同；所示量与协方差特征谱及降低过拟合之间的因果联系未充分建立。",
"legacy_material_links_verbatim": "[评审](论文/P113/评审.md) · [评分数据](论文/P113/评审.json) · [复核](论文/P113/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-018",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 93,
"source_number": 18,
"title": "Considering Nonstationary within Multivariate Time Series with Variational Hierarchical Transformer for Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/29483",
"raw_line": "| 77 | [P018 Considering Nonstationary within Multivariate Time Series with Variational Hierarchical Transformer for Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/29483) | 46.6 | 95% | 47% | 38.5–55.2 | 11 | AAAI 2024 proceedings 正式版，38(14), 15563–15570 | **关键问题**：就正式版原样写出的数学主张而言，式(2)一般情况下不是式(1)的逆；式(5)以加号连接项，式(6)的对数之和与该式缺少可见的标准ELBO推导。；
**重要问题**：已获取正文以点误差、示例曲线、层数与融合权重变化支持恢复非平稳信息和抗噪解释；没有显示参数量匹配、分布恢复或噪声扰动对照。；
**重要问题**：主表只列六个Transformer基线，部分数值引用或复跑但未逐项注明；五次均值未附波动，已公开的线性模型不在表中。 | [评审](论文/P018/评审.md) · [评分数据](论文/P018/评审.json) · 无单独复核文件 |",
"raw_cells": [
"77",
"[P018 Considering Nonstationary within Multivariate Time Series with Variational Hierarchical Transformer for Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/29483)",
"46.6",
"95%",
"47%",
"38.5–55.2",
"11",
"AAAI 2024 proceedings 正式版，38(14), 15563–15570",
"**关键问题**：就正式版原样写出的数学主张而言，式(2)一般情况下不是式(1)的逆；式(5)以加号连接项，式(6)的对数之和与该式缺少可见的标准ELBO推导。；
**重要问题**：已获取正文以点误差、示例曲线、层数与融合权重变化支持恢复非平稳信息和抗噪解释；没有显示参数量匹配、分布恢复或噪声扰动对照。；
**重要问题**：主表只列六个Transformer基线，部分数值引用或复跑但未逐项注明；五次均值未附波动，已公开的线性模型不在表中。",
"[评审](论文/P018/评审.md) · [评分数据](论文/P018/评审.json) · 无单独复核文件"
],
"normalized_title": "consideringnonstationarywithinmultivariatetimeserieswithvariationalhierarchicaltransformerforforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/29483",
"record_type": "legacy_scored",
"legacy_rank": 77,
"legacy_score": "46.6",
"legacy_coverage": "95%",
"legacy_tentative_share": "47%",
"legacy_range": "38.5–55.2",
"citation_verbatim": "11",
"legacy_review_version": "AAAI 2024 proceedings 正式版，38(14), 15563–15570",
"legacy_risks_verbatim": "**关键问题**：就正式版原样写出的数学主张而言，式(2)一般情况下不是式(1)的逆；式(5)以加号连接项，式(6)的对数之和与该式缺少可见的标准ELBO推导。；
**重要问题**：已获取正文以点误差、示例曲线、层数与融合权重变化支持恢复非平稳信息和抗噪解释；没有显示参数量匹配、分布恢复或噪声扰动对照。；
**重要问题**：主表只列六个Transformer基线，部分数值引用或复跑但未逐项注明；五次均值未附波动，已公开的线性模型不在表中。",
"legacy_material_links_verbatim": "[评审](论文/P018/评审.md) · [评分数据](论文/P018/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-027",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 94,
"source_number": 27,
"title": "Addressing Prediction Delays in Time Series Forecasting: A Continuous GRU Approach with Derivative Regularization",
"source_url": "https://doi.org/10.1145/3637528.3671969",
"raw_line": "| 78 | [P027 Addressing Prediction Delays in Time Series Forecasting: A Continuous GRU Approach with Derivative Regularization](https://doi.org/10.1145/3637528.3671969) | 46.2 | 100% | 60% | 40.0–53.0 | 39 | arXiv:2407.01622v1，本地 15 页 PDF；非已核实的 ACM 出版最终文件 | **重要问题**：Weather 各长度、三指标全面领先的说法与表 3 直接冲突：P=48 的 DTW，PatchTST 1.473 < CONTIME 1.805；P=60 的 DTW，DLinear 1.771 < CONTIME 2.121。；
**重要问题**：表 3 的 AAPL、P=24、DLinear TDI 为 3.180，附表 11 为 3.810；Weather、P=36、PatchTST DTW 在表 3 为 1.128、附表 12 为 1.428。另在表 12 中，Weather P=36 的 DLinear MSE 为 0.144±0.838；若 ± 是三次非负 MSE 的样本标准差，考虑均值四舍五入后上界约 0.2503。；
**重要问题**：式 (4) 的输出隐藏态包含双向分支，而式 (12) 只说明 f1；式 (6) 和附录 A 还涉及 h(t−τ) 与其导数。论文未充分说明后向导数、延迟历史与普通 ODE 假设如何闭合，也未证明良定性即可推出训练稳定。；
**重要问题**：附录 F 的 TDI 权重为 (h−j)^2/P^2，聚合后不保留时差正负；最优 DTW 路径也随数值形状改变。较低 TDI 的方向性滞后含义有限。；
**重要问题**：依照式 (14)，若原首个预测与末次输入相差 d，则加上文中定义的 shift 后相差 2d；论文未解释此式如何实现常见的首点对齐，表 6 的效果亦随条件变化。；
**重要问题**：附录 I 的 Weather DLinear 搜索 T∈{96,104,144}、100 轮，CONTIME 搜索 T∈{60,72,96}、150 轮；缺少持久性预测基线，表 4 的 TDI 项也不是完整 DILATE 目标。 | [评审](论文/P027/评审.md) · [评分数据](论文/P027/评审.json) · 无单独复核文件 |",
"raw_cells": [
"78",
"[P027 Addressing Prediction Delays in Time Series Forecasting: A Continuous GRU Approach with Derivative Regularization](https://doi.org/10.1145/3637528.3671969)",
"46.2",
"100%",
"60%",
"40.0–53.0",
"39",
"arXiv:2407.01622v1，本地 15 页 PDF；非已核实的 ACM 出版最终文件",
"**重要问题**：Weather 各长度、三指标全面领先的说法与表 3 直接冲突：P=48 的 DTW，PatchTST 1.473 < CONTIME 1.805；P=60 的 DTW，DLinear 1.771 < CONTIME 2.121。；
**重要问题**：表 3 的 AAPL、P=24、DLinear TDI 为 3.180，附表 11 为 3.810；Weather、P=36、PatchTST DTW 在表 3 为 1.128、附表 12 为 1.428。另在表 12 中，Weather P=36 的 DLinear MSE 为 0.144±0.838；若 ± 是三次非负 MSE 的样本标准差，考虑均值四舍五入后上界约 0.2503。；
**重要问题**：式 (4) 的输出隐藏态包含双向分支，而式 (12) 只说明 f1；式 (6) 和附录 A 还涉及 h(t−τ) 与其导数。论文未充分说明后向导数、延迟历史与普通 ODE 假设如何闭合，也未证明良定性即可推出训练稳定。；
**重要问题**：附录 F 的 TDI 权重为 (h−j)^2/P^2，聚合后不保留时差正负；最优 DTW 路径也随数值形状改变。较低 TDI 的方向性滞后含义有限。；
**重要问题**：依照式 (14)，若原首个预测与末次输入相差 d，则加上文中定义的 shift 后相差 2d；论文未解释此式如何实现常见的首点对齐，表 6 的效果亦随条件变化。；
**重要问题**：附录 I 的 Weather DLinear 搜索 T∈{96,104,144}、100 轮，CONTIME 搜索 T∈{60,72,96}、150 轮；缺少持久性预测基线，表 4 的 TDI 项也不是完整 DILATE 目标。",
"[评审](论文/P027/评审.md) · [评分数据](论文/P027/评审.json) · 无单独复核文件"
],
"normalized_title": "addressingpredictiondelaysintimeseriesforecastingacontinuousgruapproachwithderivativeregularization",
"normalized_url": "doi.org/10.1145/3637528.3671969",
"record_type": "legacy_scored",
"legacy_rank": 78,
"legacy_score": "46.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "60%",
"legacy_range": "40.0–53.0",
"citation_verbatim": "39",
"legacy_review_version": "arXiv:2407.01622v1，本地 15 页 PDF；非已核实的 ACM 出版最终文件",
"legacy_risks_verbatim": "**重要问题**：Weather 各长度、三指标全面领先的说法与表 3 直接冲突：P=48 的 DTW，PatchTST 1.473 < CONTIME 1.805；P=60 的 DTW，DLinear 1.771 < CONTIME 2.121。；
**重要问题**：表 3 的 AAPL、P=24、DLinear TDI 为 3.180，附表 11 为 3.810；Weather、P=36、PatchTST DTW 在表 3 为 1.128、附表 12 为 1.428。另在表 12 中，Weather P=36 的 DLinear MSE 为 0.144±0.838；若 ± 是三次非负 MSE 的样本标准差，考虑均值四舍五入后上界约 0.2503。；
**重要问题**：式 (4) 的输出隐藏态包含双向分支，而式 (12) 只说明 f1；式 (6) 和附录 A 还涉及 h(t−τ) 与其导数。论文未充分说明后向导数、延迟历史与普通 ODE 假设如何闭合，也未证明良定性即可推出训练稳定。；
**重要问题**：附录 F 的 TDI 权重为 (h−j)^2/P^2，聚合后不保留时差正负；最优 DTW 路径也随数值形状改变。较低 TDI 的方向性滞后含义有限。；
**重要问题**：依照式 (14)，若原首个预测与末次输入相差 d，则加上文中定义的 shift 后相差 2d；论文未解释此式如何实现常见的首点对齐，表 6 的效果亦随条件变化。；
**重要问题**：附录 I 的 Weather DLinear 搜索 T∈{96,104,144}、100 轮，CONTIME 搜索 T∈{60,72,96}、150 轮；缺少持久性预测基线，表 4 的 TDI 项也不是完整 DILATE 目标。",
"legacy_material_links_verbatim": "[评审](论文/P027/评审.md) · [评分数据](论文/P027/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-084",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 95,
"source_number": 84,
"title": "Times2D: Multi-Period Decomposition and Derivative Mapping for General Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34164",
"raw_line": "| 79 | [P084 Times2D: Multi-Period Decomposition and Derivative Mapping for General Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34164) | 46.2 | 100% | 55% | 39.7–54.0 | 14 | AAAI 2025 会议出版 PDF，卷 39 第 18 期，页 19651–19658；本地 SHA-256 见 S02 | **重要问题**：ETTh1 对 GPT4TS 和 CARD 的 7% MAE 降幅在表 2 的逐跨度值及平均行中均无对应支持；表 2 多处平均行与四个跨度数据的算术均值不符，Avg 的具体聚合方式及各项数据的来源未知。；
**重要问题**：论文式 (20) 的 MASE 分母以上界 H 求和；M4 官方指南以样本内长度 n 求和。实际计算是否遵循式 (20) 未核实，不能据此断定表 1 数值错误。；
**重要问题**：正文未给出去除差分分支或控制其他模块的消融，也未提供转折点分组评价；尚未获得独立补充材料。；
**重要问题**：表 3 中 Times2D 在三个预测跨度均慢于 PatchTST，所列内存均高于 TimesNet；资源口径和全部比较条件未写清。 | [评审](论文/P084/评审.md) · [评分数据](论文/P084/评审.json) · 无单独复核文件 |",
"raw_cells": [
"79",
"[P084 Times2D: Multi-Period Decomposition and Derivative Mapping for General Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34164)",
"46.2",
"100%",
"55%",
"39.7–54.0",
"14",
"AAAI 2025 会议出版 PDF，卷 39 第 18 期，页 19651–19658；本地 SHA-256 见 S02",
"**重要问题**：ETTh1 对 GPT4TS 和 CARD 的 7% MAE 降幅在表 2 的逐跨度值及平均行中均无对应支持；表 2 多处平均行与四个跨度数据的算术均值不符，Avg 的具体聚合方式及各项数据的来源未知。；
**重要问题**：论文式 (20) 的 MASE 分母以上界 H 求和；M4 官方指南以样本内长度 n 求和。实际计算是否遵循式 (20) 未核实，不能据此断定表 1 数值错误。；
**重要问题**：正文未给出去除差分分支或控制其他模块的消融，也未提供转折点分组评价；尚未获得独立补充材料。；
**重要问题**：表 3 中 Times2D 在三个预测跨度均慢于 PatchTST，所列内存均高于 TimesNet；资源口径和全部比较条件未写清。",
"[评审](论文/P084/评审.md) · [评分数据](论文/P084/评审.json) · 无单独复核文件"
],
"normalized_title": "times2dmultiperioddecompositionandderivativemappingforgeneraltimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34164",
"record_type": "legacy_scored",
"legacy_rank": 79,
"legacy_score": "46.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "39.7–54.0",
"citation_verbatim": "14",
"legacy_review_version": "AAAI 2025 会议出版 PDF，卷 39 第 18 期，页 19651–19658；本地 SHA-256 见 S02",
"legacy_risks_verbatim": "**重要问题**：ETTh1 对 GPT4TS 和 CARD 的 7% MAE 降幅在表 2 的逐跨度值及平均行中均无对应支持；表 2 多处平均行与四个跨度数据的算术均值不符，Avg 的具体聚合方式及各项数据的来源未知。；
**重要问题**：论文式 (20) 的 MASE 分母以上界 H 求和；M4 官方指南以样本内长度 n 求和。实际计算是否遵循式 (20) 未核实，不能据此断定表 1 数值错误。；
**重要问题**：正文未给出去除差分分支或控制其他模块的消融，也未提供转折点分组评价；尚未获得独立补充材料。；
**重要问题**：表 3 中 Times2D 在三个预测跨度均慢于 PatchTST，所列内存均高于 TimesNet；资源口径和全部比较条件未写清。",
"legacy_material_links_verbatim": "[评审](论文/P084/评审.md) · [评分数据](论文/P084/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-055",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 96,
"source_number": 55,
"title": "Parametric Augmentation for Time Series Contrastive Learning",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/ccf6d8b4a1fe9d9c8192f00c713872ea-Abstract-Conference.html",
"raw_line": "| 80 | [P055 Parametric Augmentation for Time Series Contrastive Learning](https://proceedings.iclr.cc/paper_files/paper/2024/hash/ccf6d8b4a1fe9d9c8192f00c713872ea-Abstract-Conference.html) | 45.3 | 95% | 26% | 40.5–51.0 | 12 | ICLR 2024 官方会议版；PDF 文件元数据创建于 2024-02-25，会议版确切发布日期未核实 | **关键问题**：定义1仅要求一对多和噪声熵条件，独立公平比特反例同时使性质2和性质3失效；附录把 η 的输出与其输入对等同。；
**重要问题**：g 由 x 生成时，非零系数只保证给定 g 的逐点乘法可逆；仅观察 v* 时可构造 g(x)=1/x 使不同 x 映到常数。式(8)只比较编码均值，未直接验证任务语义。；
**重要问题**：表12的 DTW 仅29个可用准确率的均值约0.650，表3/12写0.629；表6的29个 Cutout MAE 均值约0.294，表4/6却写0.394。；
**重要问题**：论文内表格未报告重复次数、跨种子波动或区间；基线采用默认/改造设置，AutoTCL 每数据集做网格搜索，调参预算及完整划分未充分说明。 | [评审](论文/P055/评审.md) · [评分数据](论文/P055/评审.json) · [复核](论文/P055/复核.md) |",
"raw_cells": [
"80",
"[P055 Parametric Augmentation for Time Series Contrastive Learning](https://proceedings.iclr.cc/paper_files/paper/2024/hash/ccf6d8b4a1fe9d9c8192f00c713872ea-Abstract-Conference.html)",
"45.3",
"95%",
"26%",
"40.5–51.0",
"12",
"ICLR 2024 官方会议版；PDF 文件元数据创建于 2024-02-25，会议版确切发布日期未核实",
"**关键问题**：定义1仅要求一对多和噪声熵条件，独立公平比特反例同时使性质2和性质3失效；附录把 η 的输出与其输入对等同。；
**重要问题**：g 由 x 生成时，非零系数只保证给定 g 的逐点乘法可逆；仅观察 v* 时可构造 g(x)=1/x 使不同 x 映到常数。式(8)只比较编码均值，未直接验证任务语义。；
**重要问题**：表12的 DTW 仅29个可用准确率的均值约0.650，表3/12写0.629；表6的29个 Cutout MAE 均值约0.294，表4/6却写0.394。；
**重要问题**：论文内表格未报告重复次数、跨种子波动或区间；基线采用默认/改造设置，AutoTCL 每数据集做网格搜索，调参预算及完整划分未充分说明。",
"[评审](论文/P055/评审.md) · [评分数据](论文/P055/评审.json) · [复核](论文/P055/复核.md)"
],
"normalized_title": "parametricaugmentationfortimeseriescontrastivelearning",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/ccf6d8b4a1fe9d9c8192f00c713872ea-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 80,
"legacy_score": "45.3",
"legacy_coverage": "95%",
"legacy_tentative_share": "26%",
"legacy_range": "40.5–51.0",
"citation_verbatim": "12",
"legacy_review_version": "ICLR 2024 官方会议版；PDF 文件元数据创建于 2024-02-25，会议版确切发布日期未核实",
"legacy_risks_verbatim": "**关键问题**：定义1仅要求一对多和噪声熵条件，独立公平比特反例同时使性质2和性质3失效；附录把 η 的输出与其输入对等同。；
**重要问题**：g 由 x 生成时，非零系数只保证给定 g 的逐点乘法可逆；仅观察 v* 时可构造 g(x)=1/x 使不同 x 映到常数。式(8)只比较编码均值，未直接验证任务语义。；
**重要问题**：表12的 DTW 仅29个可用准确率的均值约0.650，表3/12写0.629；表6的29个 Cutout MAE 均值约0.294，表4/6却写0.394。；
**重要问题**：论文内表格未报告重复次数、跨种子波动或区间；基线采用默认/改造设置，AutoTCL 每数据集做网格搜索，调参预算及完整划分未充分说明。",
"legacy_material_links_verbatim": "[评审](论文/P055/评审.md) · [评分数据](论文/P055/评审.json) · [复核](论文/P055/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-138",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 97,
"source_number": 138,
"title": "Drawing Informative Gradients from Sources: A One-stage Transfer Learning Framework for Cross-city Spatiotemporal Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/32102",
"raw_line": "| 81 | [P138 Drawing Informative Gradients from Sources: A One-stage Transfer Learning Framework for Cross-city Spatiotemporal Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/32102) | 45.3 | 95% | 47% | 36.7–54.5 | 11 | AAAI 2025 会议正式版，pp. 1147–1155，DOI 10.1609/aaai.v39i1.32102 | **重要问题**：按正文所述的标准投影，可构造可微损失使单独使用正交源梯度时目标损失增大；同一例子按式 (10) 和论文给出的学习率作完整更新时源损失增大。反例针对不附加条件的精确有限步机制解释。；
**重要问题**：印刷式 (9) 乘子的 θ_T 没有梯度点号；式 (5) 未写出对 K 个模式的聚合，Z 的 D′ 与模式库 D 的适配也未说明。；
**重要问题**：本篇表 2 的 STGP 10/30/60 分钟单时距值与 STGP v1 表 1 的 15/30/60 分钟值相同；正文称两套 5 分钟数据预测 6 步，却列出 60 分钟。METR-LA STGP 的 avg. 亦分别为 3.58 和 3.52。；
**重要问题**：正文未交代三日样本具体时间位置、验证集或 Z-score 拟合范围；说明网格搜索但缺选择流程。 | [评审](论文/P138/评审.md) · [评分数据](论文/P138/评审.json) · [复核](论文/P138/复核.md) |",
"raw_cells": [
"81",
"[P138 Drawing Informative Gradients from Sources: A One-stage Transfer Learning Framework for Cross-city Spatiotemporal Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/32102)",
"45.3",
"95%",
"47%",
"36.7–54.5",
"11",
"AAAI 2025 会议正式版，pp. 1147–1155，DOI 10.1609/aaai.v39i1.32102",
"**重要问题**：按正文所述的标准投影，可构造可微损失使单独使用正交源梯度时目标损失增大；同一例子按式 (10) 和论文给出的学习率作完整更新时源损失增大。反例针对不附加条件的精确有限步机制解释。；
**重要问题**：印刷式 (9) 乘子的 θ_T 没有梯度点号；式 (5) 未写出对 K 个模式的聚合，Z 的 D′ 与模式库 D 的适配也未说明。；
**重要问题**：本篇表 2 的 STGP 10/30/60 分钟单时距值与 STGP v1 表 1 的 15/30/60 分钟值相同；正文称两套 5 分钟数据预测 6 步，却列出 60 分钟。METR-LA STGP 的 avg. 亦分别为 3.58 和 3.52。；
**重要问题**：正文未交代三日样本具体时间位置、验证集或 Z-score 拟合范围；说明网格搜索但缺选择流程。",
"[评审](论文/P138/评审.md) · [评分数据](论文/P138/评审.json) · [复核](论文/P138/复核.md)"
],
"normalized_title": "drawinginformativegradientsfromsourcesaonestagetransferlearningframeworkforcrosscityspatiotemporalforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/32102",
"record_type": "legacy_scored",
"legacy_rank": 81,
"legacy_score": "45.3",
"legacy_coverage": "95%",
"legacy_tentative_share": "47%",
"legacy_range": "36.7–54.5",
"citation_verbatim": "11",
"legacy_review_version": "AAAI 2025 会议正式版，pp. 1147–1155，DOI 10.1609/aaai.v39i1.32102",
"legacy_risks_verbatim": "**重要问题**：按正文所述的标准投影，可构造可微损失使单独使用正交源梯度时目标损失增大；同一例子按式 (10) 和论文给出的学习率作完整更新时源损失增大。反例针对不附加条件的精确有限步机制解释。；
**重要问题**：印刷式 (9) 乘子的 θ_T 没有梯度点号；式 (5) 未写出对 K 个模式的聚合，Z 的 D′ 与模式库 D 的适配也未说明。；
**重要问题**：本篇表 2 的 STGP 10/30/60 分钟单时距值与 STGP v1 表 1 的 15/30/60 分钟值相同；正文称两套 5 分钟数据预测 6 步，却列出 60 分钟。METR-LA STGP 的 avg. 亦分别为 3.58 和 3.52。；
**重要问题**：正文未交代三日样本具体时间位置、验证集或 Z-score 拟合范围；说明网格搜索但缺选择流程。",
"legacy_material_links_verbatim": "[评审](论文/P138/评审.md) · [评分数据](论文/P138/评审.json) · [复核](论文/P138/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-035",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 98,
"source_number": 35,
"title": "STEMO: Early Spatio-temporal Forecasting with Multi-Objective Reinforcement Learning",
"source_url": "https://doi.org/10.1145/3637528.3671922",
"raw_line": "| 82 | [P035 STEMO: Early Spatio-temporal Forecasting with Multi-Objective Reinforcement Learning](https://doi.org/10.1145/3637528.3671922) | 45.2 | 100% | 50% | 40.0–52.0 | 6 | arXiv:2406.04035v3，2024-06-18 修订的作者预印本；与 ACM KDD 2024 出版终稿的差异未知 | **重要问题**：S01 式 (1) 限制 t≤T−1，表 2 标题却定义平均比例为 mean(t*)/T 并列出 100%；T=12 时按该式上界是 11/12。可能存在时点与样本数记号混用，实际计算方式未知。；
**重要问题**：S01 §5.2.1 对基线优势的概括未交代表 2 EMS 中 Graph-WaveNet 四个标称档位的 MAE 中心值更低；§5.2.2 的全面优于消融与表 4 部分 RMSE 不符。实际观测时间对齐和 ± 的含义待澄清。；
**重要问题**：S01 §5.2.3、表 5 报告两组推断偏好和 100 回合；g1 目标有定性描述，但已读版本没有给出真实偏好数值、估计误差与波动，g2 的侧重根据推断权重解释。；
**重要问题**：S01 式 (4) 在 t′=t 同时满足两个不同分支；式 (7) 将最大 Q 值写成动作。式 (1) 使用预测值的条件似然表示准确度，式 (8) 采用真值绝对误差，两者关系未解释。实现的具体处理未检查。；
**重要问题**：S01 §5.1 给出训练、验证、测试占比；已读版本未说明时间切分、跨界窗口、缺失值处理、± 的统计含义和重复次数。§5.2.1 提到 NYPD 中可能有缺失值；EARLIEST 回归适配、基线调参及停止分布的细节也不足。 | [评审](论文/P035/评审.md) · [评分数据](论文/P035/评审.json) · [复核](论文/P035/复核.md) |",
"raw_cells": [
"82",
"[P035 STEMO: Early Spatio-temporal Forecasting with Multi-Objective Reinforcement Learning](https://doi.org/10.1145/3637528.3671922)",
"45.2",
"100%",
"50%",
"40.0–52.0",
"6",
"arXiv:2406.04035v3，2024-06-18 修订的作者预印本；与 ACM KDD 2024 出版终稿的差异未知",
"**重要问题**：S01 式 (1) 限制 t≤T−1，表 2 标题却定义平均比例为 mean(t*)/T 并列出 100%；T=12 时按该式上界是 11/12。可能存在时点与样本数记号混用，实际计算方式未知。；
**重要问题**：S01 §5.2.1 对基线优势的概括未交代表 2 EMS 中 Graph-WaveNet 四个标称档位的 MAE 中心值更低；§5.2.2 的全面优于消融与表 4 部分 RMSE 不符。实际观测时间对齐和 ± 的含义待澄清。；
**重要问题**：S01 §5.2.3、表 5 报告两组推断偏好和 100 回合；g1 目标有定性描述，但已读版本没有给出真实偏好数值、估计误差与波动，g2 的侧重根据推断权重解释。；
**重要问题**：S01 式 (4) 在 t′=t 同时满足两个不同分支；式 (7) 将最大 Q 值写成动作。式 (1) 使用预测值的条件似然表示准确度，式 (8) 采用真值绝对误差，两者关系未解释。实现的具体处理未检查。；
**重要问题**：S01 §5.1 给出训练、验证、测试占比；已读版本未说明时间切分、跨界窗口、缺失值处理、± 的统计含义和重复次数。§5.2.1 提到 NYPD 中可能有缺失值；EARLIEST 回归适配、基线调参及停止分布的细节也不足。",
"[评审](论文/P035/评审.md) · [评分数据](论文/P035/评审.json) · [复核](论文/P035/复核.md)"
],
"normalized_title": "stemoearlyspatiotemporalforecastingwithmultiobjectivereinforcementlearning",
"normalized_url": "doi.org/10.1145/3637528.3671922",
"record_type": "legacy_scored",
"legacy_rank": 82,
"legacy_score": "45.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "50%",
"legacy_range": "40.0–52.0",
"citation_verbatim": "6",
"legacy_review_version": "arXiv:2406.04035v3，2024-06-18 修订的作者预印本；与 ACM KDD 2024 出版终稿的差异未知",
"legacy_risks_verbatim": "**重要问题**：S01 式 (1) 限制 t≤T−1，表 2 标题却定义平均比例为 mean(t*)/T 并列出 100%；T=12 时按该式上界是 11/12。可能存在时点与样本数记号混用，实际计算方式未知。；
**重要问题**：S01 §5.2.1 对基线优势的概括未交代表 2 EMS 中 Graph-WaveNet 四个标称档位的 MAE 中心值更低；§5.2.2 的全面优于消融与表 4 部分 RMSE 不符。实际观测时间对齐和 ± 的含义待澄清。；
**重要问题**：S01 §5.2.3、表 5 报告两组推断偏好和 100 回合；g1 目标有定性描述，但已读版本没有给出真实偏好数值、估计误差与波动，g2 的侧重根据推断权重解释。；
**重要问题**：S01 式 (4) 在 t′=t 同时满足两个不同分支；式 (7) 将最大 Q 值写成动作。式 (1) 使用预测值的条件似然表示准确度，式 (8) 采用真值绝对误差，两者关系未解释。实现的具体处理未检查。；
**重要问题**：S01 §5.1 给出训练、验证、测试占比；已读版本未说明时间切分、跨界窗口、缺失值处理、± 的统计含义和重复次数。§5.2.1 提到 NYPD 中可能有缺失值；EARLIEST 回归适配、基线调参及停止分布的细节也不足。",
"legacy_material_links_verbatim": "[评审](论文/P035/评审.md) · [评分数据](论文/P035/评审.json) · [复核](论文/P035/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-140",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 99,
"source_number": 140,
"title": "DynST: Dynamic Sparse Training for Resource-Constrained Spatio-Temporal Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709391",
"raw_line": "| 83 | [P140 DynST: Dynamic Sparse Training for Resource-Constrained Spatio-Temporal Forecasting](https://doi.org/10.1145/3690624.3709391) | 45.2 | 100% | 80% | 36.2–55.7 | 10 | arXiv:2403.02914v2，本地PDF；与KDD 2025官方列表题名及作者一致，尚未逐页比对出版社最终版 | **重要问题**：表1的Avg Speedup列分别在相应数据集/变量行列出1.721、1.541、1.987；§6.2以相同数值描述STGCN、EGNN、ViT单骨干速度，同时分别写23.7%、14.5%、34.5%。若倍数表示推理速度之比，增幅依次为72.1%、54.1%、98.7%。平均对象、逐骨干原始计时及百分比口径均未给出，数值相同本身不足以确定归属错误。；
**重要问题**：图5右侧(425,7)曲线在约22秒处真值约160°C、预测约130°C，局部差距约30°C，超过图注提到的10°C；图注未定义“整体”统计量，§6.3的15%亦未给分母。图4是预期平台示意。；
**重要问题**：第4页式2中节点掩码非零数除以邻接矩阵非零数，通常不能表示节点删除比例；第6页算法1未明确列出逐轮净剪枝操作。第6页给出1/稀疏度的加速比，但未推导其成立条件；若删除比例为s且成本与保留比例成正比、没有固定开销，理想加速比为1/(1−s)，实际计时仍取决于执行方式及开销。；
**重要问题**：式8以半监督节点分类为例，称示例损失在未掩码节点上计算；预测实验的训练目标和密集、稀疏模型的测试目标位置是否一致未明确。8:1:1之外亦未描述时序切分边界、归一化拟合和窗口处理。；
**重要问题**：表1图注称10次运行并报告基线最佳表现，但未说明是否从10次中选最佳值、逐项离散度及选择细节；表2的±未定义、表内未标明指标，且没有0%稀疏行。表4的DST在六项中有三项优于IP，正文总结为DST表现最好。 | [评审](论文/P140/评审.md) · [评分数据](论文/P140/评审.json) · [复核](论文/P140/复核.md) |",
"raw_cells": [
"83",
"[P140 DynST: Dynamic Sparse Training for Resource-Constrained Spatio-Temporal Forecasting](https://doi.org/10.1145/3690624.3709391)",
"45.2",
"100%",
"80%",
"36.2–55.7",
"10",
"arXiv:2403.02914v2，本地PDF；与KDD 2025官方列表题名及作者一致，尚未逐页比对出版社最终版",
"**重要问题**：表1的Avg Speedup列分别在相应数据集/变量行列出1.721、1.541、1.987；§6.2以相同数值描述STGCN、EGNN、ViT单骨干速度，同时分别写23.7%、14.5%、34.5%。若倍数表示推理速度之比，增幅依次为72.1%、54.1%、98.7%。平均对象、逐骨干原始计时及百分比口径均未给出，数值相同本身不足以确定归属错误。；
**重要问题**：图5右侧(425,7)曲线在约22秒处真值约160°C、预测约130°C，局部差距约30°C，超过图注提到的10°C；图注未定义“整体”统计量，§6.3的15%亦未给分母。图4是预期平台示意。；
**重要问题**：第4页式2中节点掩码非零数除以邻接矩阵非零数，通常不能表示节点删除比例；第6页算法1未明确列出逐轮净剪枝操作。第6页给出1/稀疏度的加速比，但未推导其成立条件；若删除比例为s且成本与保留比例成正比、没有固定开销，理想加速比为1/(1−s)，实际计时仍取决于执行方式及开销。；
**重要问题**：式8以半监督节点分类为例，称示例损失在未掩码节点上计算；预测实验的训练目标和密集、稀疏模型的测试目标位置是否一致未明确。8:1:1之外亦未描述时序切分边界、归一化拟合和窗口处理。；
**重要问题**：表1图注称10次运行并报告基线最佳表现，但未说明是否从10次中选最佳值、逐项离散度及选择细节；表2的±未定义、表内未标明指标，且没有0%稀疏行。表4的DST在六项中有三项优于IP，正文总结为DST表现最好。",
"[评审](论文/P140/评审.md) · [评分数据](论文/P140/评审.json) · [复核](论文/P140/复核.md)"
],
"normalized_title": "dynstdynamicsparsetrainingforresourceconstrainedspatiotemporalforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709391",
"record_type": "legacy_scored",
"legacy_rank": 83,
"legacy_score": "45.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "80%",
"legacy_range": "36.2–55.7",
"citation_verbatim": "10",
"legacy_review_version": "arXiv:2403.02914v2，本地PDF；与KDD 2025官方列表题名及作者一致，尚未逐页比对出版社最终版",
"legacy_risks_verbatim": "**重要问题**：表1的Avg Speedup列分别在相应数据集/变量行列出1.721、1.541、1.987；§6.2以相同数值描述STGCN、EGNN、ViT单骨干速度，同时分别写23.7%、14.5%、34.5%。若倍数表示推理速度之比，增幅依次为72.1%、54.1%、98.7%。平均对象、逐骨干原始计时及百分比口径均未给出，数值相同本身不足以确定归属错误。；
**重要问题**：图5右侧(425,7)曲线在约22秒处真值约160°C、预测约130°C，局部差距约30°C，超过图注提到的10°C；图注未定义“整体”统计量，§6.3的15%亦未给分母。图4是预期平台示意。；
**重要问题**：第4页式2中节点掩码非零数除以邻接矩阵非零数，通常不能表示节点删除比例；第6页算法1未明确列出逐轮净剪枝操作。第6页给出1/稀疏度的加速比，但未推导其成立条件；若删除比例为s且成本与保留比例成正比、没有固定开销，理想加速比为1/(1−s)，实际计时仍取决于执行方式及开销。；
**重要问题**：式8以半监督节点分类为例，称示例损失在未掩码节点上计算；预测实验的训练目标和密集、稀疏模型的测试目标位置是否一致未明确。8:1:1之外亦未描述时序切分边界、归一化拟合和窗口处理。；
**重要问题**：表1图注称10次运行并报告基线最佳表现，但未说明是否从10次中选最佳值、逐项离散度及选择细节；表2的±未定义、表内未标明指标，且没有0%稀疏行。表4的DST在六项中有三项优于IP，正文总结为DST表现最好。",
"legacy_material_links_verbatim": "[评审](论文/P140/评审.md) · [评分数据](论文/P140/评审.json) · [复核](论文/P140/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-034",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 100,
"source_number": 34,
"title": "Unveiling Delay Effects in Traffic Forecasting: A Perspective from Spatial-Temporal Delay Differential Equations",
"source_url": "https://doi.org/10.1145/3589334.3645688",
"raw_line": "| 84 | [P034 Unveiling Delay Effects in Traffic Forecasting: A Perspective from Spatial-Temporal Delay Differential Equations](https://doi.org/10.1145/3589334.3645688) | 44.7 | 100% | 45% | 40.2–50.0 | 36 | arXiv:2402.01231v2，2024-02-26；本地 PDF 实际 10 页，含附录 A | **关键问题**：附录式(27)的导数夹逼在 g_i<h_i 时失效；K=1、c=1、单位映射、固定门 z=1/2、两节点非零常数历史给出不衰减解，直接反驳含等号的渐近稳定断言。；
**重要问题**：PDF 图 5(c) 标为 MAPE，但其三条曲线和 28–34 纵轴与图 5(a) RMSE 相同；图 5(b) 标为 MAE 且纵轴约 10.5–14.5，完整三指标无法由图可靠辨认。；
**重要问题**：P034 表 2 的 PeMS08 PDFormer MAE/RMSE/MAPE 为 15.66/24.61/9.61；2023 年 PDFormer 原文表 2 为 13.583/23.505/9.046。两文名义划分、输入输出长度相同，但 P034 未逐项标识该基线取自论文还是重跑，也未提供足以解释差异的全部条件。；
**重要问题**：时延分组、无延迟消融和案例反映预测表现，图 1 使用互相关估计；论文未报告独立物理传播时间真值或排除共同周期、流量强度及节点差异的控制。；
**重要问题**：本次读到的正文与附录报告 6:2:2 和标准化，但未交代划分是否按时间顺序、归一化统计量或互相关预处理所用数据范围，也未说明重复运行和随机种子。 | [评审](论文/P034/评审.md) · [评分数据](论文/P034/评审.json) · [复核](论文/P034/复核.md) |",
"raw_cells": [
"84",
"[P034 Unveiling Delay Effects in Traffic Forecasting: A Perspective from Spatial-Temporal Delay Differential Equations](https://doi.org/10.1145/3589334.3645688)",
"44.7",
"100%",
"45%",
"40.2–50.0",
"36",
"arXiv:2402.01231v2，2024-02-26；本地 PDF 实际 10 页，含附录 A",
"**关键问题**：附录式(27)的导数夹逼在 g_i<h_i 时失效；K=1、c=1、单位映射、固定门 z=1/2、两节点非零常数历史给出不衰减解，直接反驳含等号的渐近稳定断言。；
**重要问题**：PDF 图 5(c) 标为 MAPE，但其三条曲线和 28–34 纵轴与图 5(a) RMSE 相同；图 5(b) 标为 MAE 且纵轴约 10.5–14.5，完整三指标无法由图可靠辨认。；
**重要问题**：P034 表 2 的 PeMS08 PDFormer MAE/RMSE/MAPE 为 15.66/24.61/9.61；2023 年 PDFormer 原文表 2 为 13.583/23.505/9.046。两文名义划分、输入输出长度相同，但 P034 未逐项标识该基线取自论文还是重跑，也未提供足以解释差异的全部条件。；
**重要问题**：时延分组、无延迟消融和案例反映预测表现，图 1 使用互相关估计；论文未报告独立物理传播时间真值或排除共同周期、流量强度及节点差异的控制。；
**重要问题**：本次读到的正文与附录报告 6:2:2 和标准化，但未交代划分是否按时间顺序、归一化统计量或互相关预处理所用数据范围，也未说明重复运行和随机种子。",
"[评审](论文/P034/评审.md) · [评分数据](论文/P034/评审.json) · [复核](论文/P034/复核.md)"
],
"normalized_title": "unveilingdelayeffectsintrafficforecastingaperspectivefromspatialtemporaldelaydifferentialequations",
"normalized_url": "doi.org/10.1145/3589334.3645688",
"record_type": "legacy_scored",
"legacy_rank": 84,
"legacy_score": "44.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "40.2–50.0",
"citation_verbatim": "36",
"legacy_review_version": "arXiv:2402.01231v2，2024-02-26；本地 PDF 实际 10 页，含附录 A",
"legacy_risks_verbatim": "**关键问题**：附录式(27)的导数夹逼在 g_i<h_i 时失效；K=1、c=1、单位映射、固定门 z=1/2、两节点非零常数历史给出不衰减解，直接反驳含等号的渐近稳定断言。；
**重要问题**：PDF 图 5(c) 标为 MAPE，但其三条曲线和 28–34 纵轴与图 5(a) RMSE 相同；图 5(b) 标为 MAE 且纵轴约 10.5–14.5，完整三指标无法由图可靠辨认。；
**重要问题**：P034 表 2 的 PeMS08 PDFormer MAE/RMSE/MAPE 为 15.66/24.61/9.61；2023 年 PDFormer 原文表 2 为 13.583/23.505/9.046。两文名义划分、输入输出长度相同，但 P034 未逐项标识该基线取自论文还是重跑，也未提供足以解释差异的全部条件。；
**重要问题**：时延分组、无延迟消融和案例反映预测表现，图 1 使用互相关估计；论文未报告独立物理传播时间真值或排除共同周期、流量强度及节点差异的控制。；
**重要问题**：本次读到的正文与附录报告 6:2:2 和标准化，但未交代划分是否按时间顺序、归一化统计量或互相关预处理所用数据范围，也未说明重复运行和随机种子。",
"legacy_material_links_verbatim": "[评审](论文/P034/评审.md) · [评分数据](论文/P034/评审.json) · [复核](论文/P034/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-048",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 101,
"source_number": 48,
"title": "Large Pre-trained time series models for cross-domain Time series analysis tasks",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/662dcc5c2b9aca77b2a0ec8a98aefae9-Abstract-Conference.html",
"raw_line": "| 85 | [P048 Large Pre-trained time series models for cross-domain Time series analysis tasks](https://proceedings.neurips.cc/paper_files/paper/2024/hash/662dcc5c2b9aca77b2a0ec8a98aefae9-Abstract-Conference.html) | 44.7 | 100% | 35% | 41.2–49.0 | 20 | NeurIPS 2024 正式会议论文 PDF（含附录）；未断言与 arXiv v3 完全相同 | **重要问题**：会议 PDF 第4页算法1第11、14行对 h(j) 取最小值而非对分段得分 s(j,h(j)) 取最小值；第12–13行在删除后不覆盖时继续删除，与正文保覆盖的描述相反。；
**重要问题**：第18页表5的23个模型行中，Flu-Japan/Flu-US 标准差比值均约542.7，ETT1/Flu-US均约0.3293；跨列近乎成比例。论文未交代原始逐次结果及检验细节。；
**重要问题**：第7页表1 NY-T 的 LPTM RMSE 13.12 高于 Lag-LLAMA 12.84、TimesFM 13.11，却被标为最优；第8页表2 NY-T和Nasdaq各有三个基线的RMSE更低。；
**重要问题**：第6页说明 ETT、纽约需求和股票使用完整训练段预训练；第8页数据效率实验仅明确缩减微调训练比例，没有交代每一比例是否重新限定预训练观测。；
**重要问题**：预训练所述 Project Tycho 数据覆盖至2021年，下游美国及日本流感回测为2004–2019年，时间范围跨越；来源、目标、地域及切分不充分，是否包含同一目标或相关未来信息未知。；
**重要问题**：第4–5页先基于序列计算 GRU 嵌入和可能重叠的段，再描述掩码段 token；若未对其他 token 的信息路径加隔离，待重建值可能进入未掩码 token。 | [评审](论文/P048/评审.md) · [评分数据](论文/P048/评审.json) · 无单独复核文件 |",
"raw_cells": [
"85",
"[P048 Large Pre-trained time series models for cross-domain Time series analysis tasks](https://proceedings.neurips.cc/paper_files/paper/2024/hash/662dcc5c2b9aca77b2a0ec8a98aefae9-Abstract-Conference.html)",
"44.7",
"100%",
"35%",
"41.2–49.0",
"20",
"NeurIPS 2024 正式会议论文 PDF（含附录）；未断言与 arXiv v3 完全相同",
"**重要问题**：会议 PDF 第4页算法1第11、14行对 h(j) 取最小值而非对分段得分 s(j,h(j)) 取最小值；第12–13行在删除后不覆盖时继续删除，与正文保覆盖的描述相反。；
**重要问题**：第18页表5的23个模型行中，Flu-Japan/Flu-US 标准差比值均约542.7，ETT1/Flu-US均约0.3293；跨列近乎成比例。论文未交代原始逐次结果及检验细节。；
**重要问题**：第7页表1 NY-T 的 LPTM RMSE 13.12 高于 Lag-LLAMA 12.84、TimesFM 13.11，却被标为最优；第8页表2 NY-T和Nasdaq各有三个基线的RMSE更低。；
**重要问题**：第6页说明 ETT、纽约需求和股票使用完整训练段预训练；第8页数据效率实验仅明确缩减微调训练比例，没有交代每一比例是否重新限定预训练观测。；
**重要问题**：预训练所述 Project Tycho 数据覆盖至2021年，下游美国及日本流感回测为2004–2019年，时间范围跨越；来源、目标、地域及切分不充分，是否包含同一目标或相关未来信息未知。；
**重要问题**：第4–5页先基于序列计算 GRU 嵌入和可能重叠的段，再描述掩码段 token；若未对其他 token 的信息路径加隔离，待重建值可能进入未掩码 token。",
"[评审](论文/P048/评审.md) · [评分数据](论文/P048/评审.json) · 无单独复核文件"
],
"normalized_title": "largepretrainedtimeseriesmodelsforcrossdomaintimeseriesanalysistasks",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/662dcc5c2b9aca77b2a0ec8a98aefae9-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 85,
"legacy_score": "44.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "35%",
"legacy_range": "41.2–49.0",
"citation_verbatim": "20",
"legacy_review_version": "NeurIPS 2024 正式会议论文 PDF（含附录）；未断言与 arXiv v3 完全相同",
"legacy_risks_verbatim": "**重要问题**：会议 PDF 第4页算法1第11、14行对 h(j) 取最小值而非对分段得分 s(j,h(j)) 取最小值；第12–13行在删除后不覆盖时继续删除，与正文保覆盖的描述相反。；
**重要问题**：第18页表5的23个模型行中，Flu-Japan/Flu-US 标准差比值均约542.7，ETT1/Flu-US均约0.3293；跨列近乎成比例。论文未交代原始逐次结果及检验细节。；
**重要问题**：第7页表1 NY-T 的 LPTM RMSE 13.12 高于 Lag-LLAMA 12.84、TimesFM 13.11，却被标为最优；第8页表2 NY-T和Nasdaq各有三个基线的RMSE更低。；
**重要问题**：第6页说明 ETT、纽约需求和股票使用完整训练段预训练；第8页数据效率实验仅明确缩减微调训练比例，没有交代每一比例是否重新限定预训练观测。；
**重要问题**：预训练所述 Project Tycho 数据覆盖至2021年，下游美国及日本流感回测为2004–2019年，时间范围跨越；来源、目标、地域及切分不充分，是否包含同一目标或相关未来信息未知。；
**重要问题**：第4–5页先基于序列计算 GRU 嵌入和可能重叠的段，再描述掩码段 token；若未对其他 token 的信息路径加隔离，待重建值可能进入未掩码 token。",
"legacy_material_links_verbatim": "[评审](论文/P048/评审.md) · [评分数据](论文/P048/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-135",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 102,
"source_number": 135,
"title": "Personalized Federated Learning for Spatio-Temporal Forecasting: A Dual Semantic Alignment-Based Contrastive Approach",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33328",
"raw_line": "| 86 | [P135 Personalized Federated Learning for Spatio-Temporal Forecasting: A Dual Semantic Alignment-Based Contrastive Approach](https://ojs.aaai.org/index.php/AAAI/article/view/33328) | 44.5 | 100% | 85% | 33.0–59.7 | 11 | AAAI-25 会议论文出版版；附录参照另行标注的 arXiv:2404.03702v1 | **重要问题**：会议版中 SM=exp(sim/τ)>0，Z=ReLU(SM⊙W)；在相似度有定义、温度有限非零和实数运算条件下，Z_bi>0 当且仅当 W_bi>0。给定 W，入选状态不随相似度变化，但入选负例的损失值仍受相似度影响。式 (2) 仅通过离散集合资格使用 W，算法 1 更新未列 W。预印本 v1 式 (6) 的求和使用 Z，和会议版写法不同。；
**重要问题**：会议版式 (5) 将 ρ 乘于 L_inter，算法 1 第 18 行将 ρ 乘于 L_pred；ρ≠1 时两式不同。式 (3) 使用 d_r 维 r 与 d_p 维原型做相似度，图 2 则画为 r′；若 d_r≠d_p，按该式无法直接计算余弦相似度。；
**重要问题**：会议版定理 5.2 右侧包含 C D sqrt(log(1/κ)/2)。若 ρ≥0、0≤L≤C、C>0、D≥1 且 κ=0.05，该项大于损失差的平凡上界 C；定理 5.6 的式 (9) 要求正学习率时须有 ξ>ρ L_h N α I G。预印本 v1 附录 C.1 式 (26) 的最后一步，对两个均为 100 个样本的客户端和 d=1，左、右分别约 0.334818 和 0.250964。 | [评审](论文/P135/评审.md) · [评分数据](论文/P135/评审.json) · [复核](论文/P135/复核.md) |",
"raw_cells": [
"86",
"[P135 Personalized Federated Learning for Spatio-Temporal Forecasting: A Dual Semantic Alignment-Based Contrastive Approach](https://ojs.aaai.org/index.php/AAAI/article/view/33328)",
"44.5",
"100%",
"85%",
"33.0–59.7",
"11",
"AAAI-25 会议论文出版版；附录参照另行标注的 arXiv:2404.03702v1",
"**重要问题**：会议版中 SM=exp(sim/τ)>0，Z=ReLU(SM⊙W)；在相似度有定义、温度有限非零和实数运算条件下，Z_bi>0 当且仅当 W_bi>0。给定 W，入选状态不随相似度变化，但入选负例的损失值仍受相似度影响。式 (2) 仅通过离散集合资格使用 W，算法 1 更新未列 W。预印本 v1 式 (6) 的求和使用 Z，和会议版写法不同。；
**重要问题**：会议版式 (5) 将 ρ 乘于 L_inter，算法 1 第 18 行将 ρ 乘于 L_pred；ρ≠1 时两式不同。式 (3) 使用 d_r 维 r 与 d_p 维原型做相似度，图 2 则画为 r′；若 d_r≠d_p，按该式无法直接计算余弦相似度。；
**重要问题**：会议版定理 5.2 右侧包含 C D sqrt(log(1/κ)/2)。若 ρ≥0、0≤L≤C、C>0、D≥1 且 κ=0.05，该项大于损失差的平凡上界 C；定理 5.6 的式 (9) 要求正学习率时须有 ξ>ρ L_h N α I G。预印本 v1 附录 C.1 式 (26) 的最后一步，对两个均为 100 个样本的客户端和 d=1，左、右分别约 0.334818 和 0.250964。",
"[评审](论文/P135/评审.md) · [评分数据](论文/P135/评审.json) · [复核](论文/P135/复核.md)"
],
"normalized_title": "personalizedfederatedlearningforspatiotemporalforecastingadualsemanticalignmentbasedcontrastiveapproach",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33328",
"record_type": "legacy_scored",
"legacy_rank": 86,
"legacy_score": "44.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "85%",
"legacy_range": "33.0–59.7",
"citation_verbatim": "11",
"legacy_review_version": "AAAI-25 会议论文出版版；附录参照另行标注的 arXiv:2404.03702v1",
"legacy_risks_verbatim": "**重要问题**：会议版中 SM=exp(sim/τ)>0，Z=ReLU(SM⊙W)；在相似度有定义、温度有限非零和实数运算条件下，Z_bi>0 当且仅当 W_bi>0。给定 W，入选状态不随相似度变化，但入选负例的损失值仍受相似度影响。式 (2) 仅通过离散集合资格使用 W，算法 1 更新未列 W。预印本 v1 式 (6) 的求和使用 Z，和会议版写法不同。；
**重要问题**：会议版式 (5) 将 ρ 乘于 L_inter，算法 1 第 18 行将 ρ 乘于 L_pred；ρ≠1 时两式不同。式 (3) 使用 d_r 维 r 与 d_p 维原型做相似度，图 2 则画为 r′；若 d_r≠d_p，按该式无法直接计算余弦相似度。；
**重要问题**：会议版定理 5.2 右侧包含 C D sqrt(log(1/κ)/2)。若 ρ≥0、0≤L≤C、C>0、D≥1 且 κ=0.05，该项大于损失差的平凡上界 C；定理 5.6 的式 (9) 要求正学习率时须有 ξ>ρ L_h N α I G。预印本 v1 附录 C.1 式 (26) 的最后一步，对两个均为 100 个样本的客户端和 d=1，左、右分别约 0.334818 和 0.250964。",
"legacy_material_links_verbatim": "[评审](论文/P135/评审.md) · [评分数据](论文/P135/评审.json) · [复核](论文/P135/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-106",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 103,
"source_number": 106,
"title": "CAMEF: Causal-Augmented Multi-Modality Event-Driven Financial Forecasting by Integrating Time Series Patterns and Salient Macroeconomic Announcements",
"source_url": "https://doi.org/10.1145/3711896.3736872",
"raw_line": "| 87 | [P106 CAMEF: Causal-Augmented Multi-Modality Event-Driven Financial Forecasting by Integrating Time Series Patterns and Salient Macroeconomic Announcements](https://doi.org/10.1145/3711896.3736872) | 44.2 | 95% | 37% | 38.0–51.7 | 9 | arXiv:2502.04592v3，2025-08-08；含会议信息的作者预印本 | **重要问题**：§3、式 (16)–(17) 使用与真实公告匹配的历史序列和生成文本形成区分性监督；生成文本未配对应真实反事实市场结果，论文未提出识别公告干预效应的充分假设或验证。图 2 把 E 指向当期趋势 X，未明确 X 与公告前输入段的关系；若两者相同，箭头有时序疑问。；
**重要问题**：若对表 3 同一预测长度的五资产 MSE 取等权算术平均，去 GPT-2 的 70、140 步分别为 0.149884、0.187972，表中为 0.14593、0.03801；70 步去文本、去因果项、去融合的复算值也分别与表中平均列不同。论文未明确平均权重，差异原因及正确原始结果未知。；
**重要问题**：已读 v3 正文及附录说明 6:2:2 比例和部分预处理，未明确是否按时间划分、各段日期、重叠事件窗口隔离、数值标准化和最近日期异类负例是否受训练边界限制。；
**重要问题**：表 4 的 FOMC 70 步 MAE 0.0170094 高于就业形势 0.0154431；另外五个按长度及指标划分的 FOMC 格子最优。作者没有明确 §6.4 概括的跨长度汇总方式，也未清楚交代各事件类别的训练、测试样本及对照构造。；
**重要问题**：v3 式 (6) 的 W3 标为 1024×768，却左乘前一层 1024 维输出，偏置为 768 维而残差输出标为 1024 维。式 (17) 将余弦相似度列为距离 d 的例子；若直接代入原始相似度，激活的损失项会推动正例相似度降低、负例相似度提高，与文字目标相反。；
**重要问题**：v3 正文与 PDF 内附录未见多随机种子、误差条、置信区间、逐事件误差分布、跨时期压力检验或可比延迟和成本；图 4 仅展示 SPX/35 的有限参数点，样本数和层数的最大点分别为 10 和 4。；
**重要问题**：v3 附录 C 报告 GPT4MTS 由作者重实现且将 BERT 改为 Longformer；单模态、TEST/GPT4MTS 和 CAMEF 的批量设置分别为 32、7、10。§6.1 报告验证集用于早停，论文未完整说明统一调参预算与实际停止轮次，表 2 也未列简单末值持续预测。 | [评审](论文/P106/评审.md) · [评分数据](论文/P106/评审.json) · [复核](论文/P106/复核.md) |",
"raw_cells": [
"87",
"[P106 CAMEF: Causal-Augmented Multi-Modality Event-Driven Financial Forecasting by Integrating Time Series Patterns and Salient Macroeconomic Announcements](https://doi.org/10.1145/3711896.3736872)",
"44.2",
"95%",
"37%",
"38.0–51.7",
"9",
"arXiv:2502.04592v3，2025-08-08；含会议信息的作者预印本",
"**重要问题**：§3、式 (16)–(17) 使用与真实公告匹配的历史序列和生成文本形成区分性监督；生成文本未配对应真实反事实市场结果，论文未提出识别公告干预效应的充分假设或验证。图 2 把 E 指向当期趋势 X，未明确 X 与公告前输入段的关系；若两者相同，箭头有时序疑问。；
**重要问题**：若对表 3 同一预测长度的五资产 MSE 取等权算术平均，去 GPT-2 的 70、140 步分别为 0.149884、0.187972，表中为 0.14593、0.03801；70 步去文本、去因果项、去融合的复算值也分别与表中平均列不同。论文未明确平均权重，差异原因及正确原始结果未知。；
**重要问题**：已读 v3 正文及附录说明 6:2:2 比例和部分预处理，未明确是否按时间划分、各段日期、重叠事件窗口隔离、数值标准化和最近日期异类负例是否受训练边界限制。；
**重要问题**：表 4 的 FOMC 70 步 MAE 0.0170094 高于就业形势 0.0154431；另外五个按长度及指标划分的 FOMC 格子最优。作者没有明确 §6.4 概括的跨长度汇总方式，也未清楚交代各事件类别的训练、测试样本及对照构造。；
**重要问题**：v3 式 (6) 的 W3 标为 1024×768，却左乘前一层 1024 维输出，偏置为 768 维而残差输出标为 1024 维。式 (17) 将余弦相似度列为距离 d 的例子；若直接代入原始相似度，激活的损失项会推动正例相似度降低、负例相似度提高，与文字目标相反。；
**重要问题**：v3 正文与 PDF 内附录未见多随机种子、误差条、置信区间、逐事件误差分布、跨时期压力检验或可比延迟和成本；图 4 仅展示 SPX/35 的有限参数点，样本数和层数的最大点分别为 10 和 4。；
**重要问题**：v3 附录 C 报告 GPT4MTS 由作者重实现且将 BERT 改为 Longformer；单模态、TEST/GPT4MTS 和 CAMEF 的批量设置分别为 32、7、10。§6.1 报告验证集用于早停，论文未完整说明统一调参预算与实际停止轮次，表 2 也未列简单末值持续预测。",
"[评审](论文/P106/评审.md) · [评分数据](论文/P106/评审.json) · [复核](论文/P106/复核.md)"
],
"normalized_title": "camefcausalaugmentedmultimodalityeventdrivenfinancialforecastingbyintegratingtimeseriespatternsandsalientmacroeconomicannouncements",
"normalized_url": "doi.org/10.1145/3711896.3736872",
"record_type": "legacy_scored",
"legacy_rank": 87,
"legacy_score": "44.2",
"legacy_coverage": "95%",
"legacy_tentative_share": "37%",
"legacy_range": "38.0–51.7",
"citation_verbatim": "9",
"legacy_review_version": "arXiv:2502.04592v3，2025-08-08；含会议信息的作者预印本",
"legacy_risks_verbatim": "**重要问题**：§3、式 (16)–(17) 使用与真实公告匹配的历史序列和生成文本形成区分性监督；生成文本未配对应真实反事实市场结果，论文未提出识别公告干预效应的充分假设或验证。图 2 把 E 指向当期趋势 X，未明确 X 与公告前输入段的关系；若两者相同，箭头有时序疑问。；
**重要问题**：若对表 3 同一预测长度的五资产 MSE 取等权算术平均，去 GPT-2 的 70、140 步分别为 0.149884、0.187972，表中为 0.14593、0.03801；70 步去文本、去因果项、去融合的复算值也分别与表中平均列不同。论文未明确平均权重，差异原因及正确原始结果未知。；
**重要问题**：已读 v3 正文及附录说明 6:2:2 比例和部分预处理，未明确是否按时间划分、各段日期、重叠事件窗口隔离、数值标准化和最近日期异类负例是否受训练边界限制。；
**重要问题**：表 4 的 FOMC 70 步 MAE 0.0170094 高于就业形势 0.0154431；另外五个按长度及指标划分的 FOMC 格子最优。作者没有明确 §6.4 概括的跨长度汇总方式，也未清楚交代各事件类别的训练、测试样本及对照构造。；
**重要问题**：v3 式 (6) 的 W3 标为 1024×768，却左乘前一层 1024 维输出，偏置为 768 维而残差输出标为 1024 维。式 (17) 将余弦相似度列为距离 d 的例子；若直接代入原始相似度，激活的损失项会推动正例相似度降低、负例相似度提高，与文字目标相反。；
**重要问题**：v3 正文与 PDF 内附录未见多随机种子、误差条、置信区间、逐事件误差分布、跨时期压力检验或可比延迟和成本；图 4 仅展示 SPX/35 的有限参数点，样本数和层数的最大点分别为 10 和 4。；
**重要问题**：v3 附录 C 报告 GPT4MTS 由作者重实现且将 BERT 改为 Longformer；单模态、TEST/GPT4MTS 和 CAMEF 的批量设置分别为 32、7、10。§6.1 报告验证集用于早停，论文未完整说明统一调参预算与实际停止轮次，表 2 也未列简单末值持续预测。",
"legacy_material_links_verbatim": "[评审](论文/P106/评审.md) · [评分数据](论文/P106/评审.json) · [复核](论文/P106/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-032",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 104,
"source_number": 32,
"title": "ModWaveMLP: MLP-Based Mode Decomposition and Wavelet Denoising Model to Defeat Complex Structures in Traffic Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/28753",
"raw_line": "| 88 | [P032 ModWaveMLP: MLP-Based Mode Decomposition and Wavelet Denoising Model to Defeat Complex Structures in Traffic Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/28753) | 43.2 | 95% | 58% | 34.7–53.2 | 27 | AAAI 2024会议论文，卷38第8期，9035–9043页；另核对论文所指仓库的2024-03-29结果更新 | **重要问题**：论文所指仓库在2024-03-29更新主结果，两数据集三个报告时距的MAE均与会议表1不同；按表面数值与原表最佳基线比较，六项均更高。新旧模型、数据、训练及测试条件的对应关系和差异成因未知。；
**重要问题**：表1的DCRNN在METR-LA 15/30/60分钟MAE为2.77/3.15/3.60；DCRNN官方README列修错后2.67/3.08/3.56。此差异表明基线版本或结果来源有差别，条件是否可比尚不清楚。；
**重要问题**：正文将小波张量、基线硬件、波基选择和成本指向附录A、B、D、E，所获九页PDF及关联仓库入口未找到附录；小波处理的预测时点边界与结果更新协议在已获材料中不清。；
**重要问题**：表2移除小波分解时也移除了编码支路，图3–4仅展示单一节点和层权重；这些观察把滤噪、额外输入与容量变化混在一起。 | [评审](论文/P032/评审.md) · [评分数据](论文/P032/评审.json) · 无单独复核文件 |",
"raw_cells": [
"88",
"[P032 ModWaveMLP: MLP-Based Mode Decomposition and Wavelet Denoising Model to Defeat Complex Structures in Traffic Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/28753)",
"43.2",
"95%",
"58%",
"34.7–53.2",
"27",
"AAAI 2024会议论文，卷38第8期，9035–9043页；另核对论文所指仓库的2024-03-29结果更新",
"**重要问题**：论文所指仓库在2024-03-29更新主结果，两数据集三个报告时距的MAE均与会议表1不同；按表面数值与原表最佳基线比较，六项均更高。新旧模型、数据、训练及测试条件的对应关系和差异成因未知。；
**重要问题**：表1的DCRNN在METR-LA 15/30/60分钟MAE为2.77/3.15/3.60；DCRNN官方README列修错后2.67/3.08/3.56。此差异表明基线版本或结果来源有差别，条件是否可比尚不清楚。；
**重要问题**：正文将小波张量、基线硬件、波基选择和成本指向附录A、B、D、E，所获九页PDF及关联仓库入口未找到附录；小波处理的预测时点边界与结果更新协议在已获材料中不清。；
**重要问题**：表2移除小波分解时也移除了编码支路，图3–4仅展示单一节点和层权重；这些观察把滤噪、额外输入与容量变化混在一起。",
"[评审](论文/P032/评审.md) · [评分数据](论文/P032/评审.json) · 无单独复核文件"
],
"normalized_title": "modwavemlpmlpbasedmodedecompositionandwaveletdenoisingmodeltodefeatcomplexstructuresintrafficforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/28753",
"record_type": "legacy_scored",
"legacy_rank": 88,
"legacy_score": "43.2",
"legacy_coverage": "95%",
"legacy_tentative_share": "58%",
"legacy_range": "34.7–53.2",
"citation_verbatim": "27",
"legacy_review_version": "AAAI 2024会议论文，卷38第8期，9035–9043页；另核对论文所指仓库的2024-03-29结果更新",
"legacy_risks_verbatim": "**重要问题**：论文所指仓库在2024-03-29更新主结果，两数据集三个报告时距的MAE均与会议表1不同；按表面数值与原表最佳基线比较，六项均更高。新旧模型、数据、训练及测试条件的对应关系和差异成因未知。；
**重要问题**：表1的DCRNN在METR-LA 15/30/60分钟MAE为2.77/3.15/3.60；DCRNN官方README列修错后2.67/3.08/3.56。此差异表明基线版本或结果来源有差别，条件是否可比尚不清楚。；
**重要问题**：正文将小波张量、基线硬件、波基选择和成本指向附录A、B、D、E，所获九页PDF及关联仓库入口未找到附录；小波处理的预测时点边界与结果更新协议在已获材料中不清。；
**重要问题**：表2移除小波分解时也移除了编码支路，图3–4仅展示单一节点和层权重；这些观察把滤噪、额外输入与容量变化混在一起。",
"legacy_material_links_verbatim": "[评审](论文/P032/评审.md) · [评分数据](论文/P032/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-142",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 105,
"source_number": 142,
"title": "Efficient Traffic Prediction Through Spatio-Temporal Distillation",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/32096",
"raw_line": "| 89 | [P142 Efficient Traffic Prediction Through Spatio-Temporal Distillation](https://ojs.aaai.org/index.php/AAAI/article/view/32096) | 42.7 | 100% | 45% | 38.0–48.5 | 28 | AAAI-25 出版版，Proceedings of the AAAI Conference on Artificial Intelligence 39(1):1093–1101，DOI 10.1609/aaai.v39i1.32096 | **重要问题**：式 (6) 所写为教师 TCN 输出乘学生 log-softmax 的正号和式，缺少作为概率的教师归一化及 KL 所需项；式 (8) 是联合损失，算法 1 第 12 行却按 L(S) 求导更新。两者影响按文稿重建蒸馏过程。；
**重要问题**：表 1 未比较已在相关工作提及的 STID、ST-MLP，也没有教师行或完全无蒸馏同结构学生；训练/验证/测试划分和基线调参细节未报告。；
**重要问题**：表 3 写 PeMS-Bay 为 2017-01-01 至 2017-05-31、5 分钟间隔且有 52,116 个时间点；该时间范围含 151 天，最多 43,488 个不重复的 5 分钟时间点。；
**重要问题**：表 1–2 未给出重复运行次数、误差波动、基线和学生等量输入或计时范围；速度差与小幅精度优势的稳定性不明。；
**重要问题**：式 (9) 的路径式未包含前述学习权重和 ReLU，也未说明简化条件；论文未报告节点表示过平滑指标或受控图噪声实验。 | [评审](论文/P142/评审.md) · [评分数据](论文/P142/评审.json) · 无单独复核文件 |",
"raw_cells": [
"89",
"[P142 Efficient Traffic Prediction Through Spatio-Temporal Distillation](https://ojs.aaai.org/index.php/AAAI/article/view/32096)",
"42.7",
"100%",
"45%",
"38.0–48.5",
"28",
"AAAI-25 出版版，Proceedings of the AAAI Conference on Artificial Intelligence 39(1):1093–1101，DOI 10.1609/aaai.v39i1.32096",
"**重要问题**：式 (6) 所写为教师 TCN 输出乘学生 log-softmax 的正号和式，缺少作为概率的教师归一化及 KL 所需项；式 (8) 是联合损失，算法 1 第 12 行却按 L(S) 求导更新。两者影响按文稿重建蒸馏过程。；
**重要问题**：表 1 未比较已在相关工作提及的 STID、ST-MLP，也没有教师行或完全无蒸馏同结构学生；训练/验证/测试划分和基线调参细节未报告。；
**重要问题**：表 3 写 PeMS-Bay 为 2017-01-01 至 2017-05-31、5 分钟间隔且有 52,116 个时间点；该时间范围含 151 天，最多 43,488 个不重复的 5 分钟时间点。；
**重要问题**：表 1–2 未给出重复运行次数、误差波动、基线和学生等量输入或计时范围；速度差与小幅精度优势的稳定性不明。；
**重要问题**：式 (9) 的路径式未包含前述学习权重和 ReLU，也未说明简化条件；论文未报告节点表示过平滑指标或受控图噪声实验。",
"[评审](论文/P142/评审.md) · [评分数据](论文/P142/评审.json) · 无单独复核文件"
],
"normalized_title": "efficienttrafficpredictionthroughspatiotemporaldistillation",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/32096",
"record_type": "legacy_scored",
"legacy_rank": 89,
"legacy_score": "42.7",
"legacy_coverage": "100%",
"legacy_tentative_share": "45%",
"legacy_range": "38.0–48.5",
"citation_verbatim": "28",
"legacy_review_version": "AAAI-25 出版版，Proceedings of the AAAI Conference on Artificial Intelligence 39(1):1093–1101，DOI 10.1609/aaai.v39i1.32096",
"legacy_risks_verbatim": "**重要问题**：式 (6) 所写为教师 TCN 输出乘学生 log-softmax 的正号和式，缺少作为概率的教师归一化及 KL 所需项；式 (8) 是联合损失，算法 1 第 12 行却按 L(S) 求导更新。两者影响按文稿重建蒸馏过程。；
**重要问题**：表 1 未比较已在相关工作提及的 STID、ST-MLP，也没有教师行或完全无蒸馏同结构学生；训练/验证/测试划分和基线调参细节未报告。；
**重要问题**：表 3 写 PeMS-Bay 为 2017-01-01 至 2017-05-31、5 分钟间隔且有 52,116 个时间点；该时间范围含 151 天，最多 43,488 个不重复的 5 分钟时间点。；
**重要问题**：表 1–2 未给出重复运行次数、误差波动、基线和学生等量输入或计时范围；速度差与小幅精度优势的稳定性不明。；
**重要问题**：式 (9) 的路径式未包含前述学习权重和 ReLU，也未说明简化条件；论文未报告节点表示过平滑指标或受控图噪声实验。",
"legacy_material_links_verbatim": "[评审](论文/P142/评审.md) · [评分数据](论文/P142/评审.json) · 无单独复核文件",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-144",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 106,
"source_number": 144,
"title": "UoMo: A Universal Model of Mobile Traffic Forecasting for Wireless Network Optimization",
"source_url": "https://doi.org/10.1145/3711896.3737272",
"raw_line": "| 90 | [P144 UoMo: A Universal Model of Mobile Traffic Forecasting for Wireless Network Optimization](https://doi.org/10.1145/3711896.3737272) | 42.5 | 100% | 55% | 37.0–49.7 | 7 | arXiv:2410.15322v4（本地 PDF，13 页；与会议 DOI 题名、作者一致，未核实与会议终稿逐项相同） | **重要问题**：附录 A.3 将互信息写成条件/边缘密度比；独立变量时互信息为 0 而密度比为 1，等式错误。式（18）至（19）缺少支撑近似的条件。式（11）的负误差项在无界且其余项不抵消时可使目标趋于负无穷，而 InfoNCE 非负；取值范围差异本身不排除两目标有相同最优解。；
**重要问题**：式（1）把未来目标位置记为 m=0，式（4）却把 e=E(X)⊙m 称为待重建部分，式（7）亦乘 m；式（5）的噪声系数为 1−α̂。按附录 A.3 所列单步高斯链及标准累计定义 α̂_k=∏_{i=1}^k(1−β_i)，累计标准差为 √(1−α̂_k)，与印刷公式不符。；
**重要问题**：图 9 和 §5.2 的归一化折损指标 0.83→0.59，按 0.83 为基线相对下降约 28.9%，而文中声称 40.7%；以 0.59 为分母得到约 40.7%，对应基线高于 UoMo 的相对差值；摘要将 27.85%、18.57% 分别对应长期、短期，正文 §4.2 则将同样数字分别称为短期、长期 MAE 的平均改善。；
**重要问题**：§3.3 与表 7 将移动用户数作为时间变化的上下文；正文和附录没有交代预测时段的用户数是如何取得，也未给出可追溯的时间切分、窗口重叠检查和各基线等信息输入。慕尼黑缺用户数时的处理亦未描述。；
**重要问题**：表 3 的北京、上海 UoMo 生成 JSD 为 0.2013、0.2259，表 5 分别为 0.2213、0.2202；两表条件是否一致未交代。表 5 预测列未说明为何北京、上海数值对应表 2 长期任务，而南昌对应表 1 短期任务。以表 5 的完整模型数值为分母，北京移除 POI 后预测 RMSE 增幅约 18.84%，表中 Δ 为 −23.82%；该 Δ 的计算方式未说明。；
**重要问题**：§5 的收益评估将预测送入平台求解，再用真实轨迹数据验证方案；§7 称模型已用于南宁实时预测。论文未报告这些收益来自在线随机对照；图 9 使用归一化指标，归一化细节未说明，且休眠方案的能耗高于 Lag-based 和 Hist-based。 | [评审](论文/P144/评审.md) · [评分数据](论文/P144/评审.json) · [复核](论文/P144/复核.md) |",
"raw_cells": [
"90",
"[P144 UoMo: A Universal Model of Mobile Traffic Forecasting for Wireless Network Optimization](https://doi.org/10.1145/3711896.3737272)",
"42.5",
"100%",
"55%",
"37.0–49.7",
"7",
"arXiv:2410.15322v4（本地 PDF，13 页；与会议 DOI 题名、作者一致，未核实与会议终稿逐项相同）",
"**重要问题**：附录 A.3 将互信息写成条件/边缘密度比；独立变量时互信息为 0 而密度比为 1，等式错误。式（18）至（19）缺少支撑近似的条件。式（11）的负误差项在无界且其余项不抵消时可使目标趋于负无穷，而 InfoNCE 非负；取值范围差异本身不排除两目标有相同最优解。；
**重要问题**：式（1）把未来目标位置记为 m=0，式（4）却把 e=E(X)⊙m 称为待重建部分，式（7）亦乘 m；式（5）的噪声系数为 1−α̂。按附录 A.3 所列单步高斯链及标准累计定义 α̂_k=∏_{i=1}^k(1−β_i)，累计标准差为 √(1−α̂_k)，与印刷公式不符。；
**重要问题**：图 9 和 §5.2 的归一化折损指标 0.83→0.59，按 0.83 为基线相对下降约 28.9%，而文中声称 40.7%；以 0.59 为分母得到约 40.7%，对应基线高于 UoMo 的相对差值；摘要将 27.85%、18.57% 分别对应长期、短期，正文 §4.2 则将同样数字分别称为短期、长期 MAE 的平均改善。；
**重要问题**：§3.3 与表 7 将移动用户数作为时间变化的上下文；正文和附录没有交代预测时段的用户数是如何取得，也未给出可追溯的时间切分、窗口重叠检查和各基线等信息输入。慕尼黑缺用户数时的处理亦未描述。；
**重要问题**：表 3 的北京、上海 UoMo 生成 JSD 为 0.2013、0.2259，表 5 分别为 0.2213、0.2202；两表条件是否一致未交代。表 5 预测列未说明为何北京、上海数值对应表 2 长期任务，而南昌对应表 1 短期任务。以表 5 的完整模型数值为分母，北京移除 POI 后预测 RMSE 增幅约 18.84%，表中 Δ 为 −23.82%；该 Δ 的计算方式未说明。；
**重要问题**：§5 的收益评估将预测送入平台求解，再用真实轨迹数据验证方案；§7 称模型已用于南宁实时预测。论文未报告这些收益来自在线随机对照；图 9 使用归一化指标，归一化细节未说明，且休眠方案的能耗高于 Lag-based 和 Hist-based。",
"[评审](论文/P144/评审.md) · [评分数据](论文/P144/评审.json) · [复核](论文/P144/复核.md)"
],
"normalized_title": "uomoauniversalmodelofmobiletrafficforecastingforwirelessnetworkoptimization",
"normalized_url": "doi.org/10.1145/3711896.3737272",
"record_type": "legacy_scored",
"legacy_rank": 90,
"legacy_score": "42.5",
"legacy_coverage": "100%",
"legacy_tentative_share": "55%",
"legacy_range": "37.0–49.7",
"citation_verbatim": "7",
"legacy_review_version": "arXiv:2410.15322v4（本地 PDF，13 页；与会议 DOI 题名、作者一致，未核实与会议终稿逐项相同）",
"legacy_risks_verbatim": "**重要问题**：附录 A.3 将互信息写成条件/边缘密度比；独立变量时互信息为 0 而密度比为 1，等式错误。式（18）至（19）缺少支撑近似的条件。式（11）的负误差项在无界且其余项不抵消时可使目标趋于负无穷，而 InfoNCE 非负；取值范围差异本身不排除两目标有相同最优解。；
**重要问题**：式（1）把未来目标位置记为 m=0，式（4）却把 e=E(X)⊙m 称为待重建部分，式（7）亦乘 m；式（5）的噪声系数为 1−α̂。按附录 A.3 所列单步高斯链及标准累计定义 α̂_k=∏_{i=1}^k(1−β_i)，累计标准差为 √(1−α̂_k)，与印刷公式不符。；
**重要问题**：图 9 和 §5.2 的归一化折损指标 0.83→0.59，按 0.83 为基线相对下降约 28.9%，而文中声称 40.7%；以 0.59 为分母得到约 40.7%，对应基线高于 UoMo 的相对差值；摘要将 27.85%、18.57% 分别对应长期、短期，正文 §4.2 则将同样数字分别称为短期、长期 MAE 的平均改善。；
**重要问题**：§3.3 与表 7 将移动用户数作为时间变化的上下文；正文和附录没有交代预测时段的用户数是如何取得，也未给出可追溯的时间切分、窗口重叠检查和各基线等信息输入。慕尼黑缺用户数时的处理亦未描述。；
**重要问题**：表 3 的北京、上海 UoMo 生成 JSD 为 0.2013、0.2259，表 5 分别为 0.2213、0.2202；两表条件是否一致未交代。表 5 预测列未说明为何北京、上海数值对应表 2 长期任务，而南昌对应表 1 短期任务。以表 5 的完整模型数值为分母，北京移除 POI 后预测 RMSE 增幅约 18.84%，表中 Δ 为 −23.82%；该 Δ 的计算方式未说明。；
**重要问题**：§5 的收益评估将预测送入平台求解，再用真实轨迹数据验证方案；§7 称模型已用于南宁实时预测。论文未报告这些收益来自在线随机对照；图 9 使用归一化指标，归一化细节未说明，且休眠方案的能耗高于 Lag-based 和 Hist-based。",
"legacy_material_links_verbatim": "[评审](论文/P144/评审.md) · [评分数据](论文/P144/评审.json) · [复核](论文/P144/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-046",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 107,
"source_number": 46,
"title": "Tri-Level Navigator: LLM-Empowered Tri-Level Learning for Time Series OOD Generalization",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/c78d9d035215039ab46e48c281e6a63d-Abstract-Conference.html",
"raw_line": "| 91 | [P046 Tri-Level Navigator: LLM-Empowered Tri-Level Learning for Time Series OOD Generalization](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c78d9d035215039ab46e48c281e6a63d-Abstract-Conference.html) | 42.4 | 95% | 53% | 34.0–52.0 | 4 | NeurIPS 2024 官方论文集 PDF，含正文及附录；本次未将 arXiv 版本等同于会议版 | **关键问题**：式(53)的未归一化梯度平方和上界在定理5所列假设及步长下存在反例：取 F=(θ+q)²/4，θ⁰=q⁰=1、δ任意，t₁=0、T₁=1、L=1、真实梯度、噪声方差0、α₁²=α₂²=1、α₃²=0，步长1；更新后 θ¹=q¹=0，F*=0，而式(53)左边为2、右边为1。式(54)—(56)由此得到的迭代界未被所给证明建立；该反例针对证明步骤，未反驳速率命题或具体训练结果。；
**重要问题**：表3 USC-HAD 域 A 为 2,3,4，域 B 为 4,5,6，同一编号4同时列入两域；第20页表2写 subjects 为4，表3编号却延伸到14，论文未清楚定义这些编号和原始受试者的对应关系。若两处编号4指同一受试者，其数据会跨域；该受试者的记录或窗口是否重复、是否再作互斥划分均未知。；
**重要问题**：附录 B.3 指出按每个域的均值和标准差标准化，未交代目标域统计量估计区间、可用时点及是否进入训练；若用整个留出域估计，则测试使用了无标签目标分布信息。文中未报告使用目标标签。；
**重要问题**：式(11)中的 max_{i,j} d(P_Si,P_Sj) 对给定源分布及假设类没有混合权重 α；附录 A.2.3 却称动态调权可最小化该项。目标域差异项未被直接观测；若作者意图让分布随表示变化，论文未在该处给出相应推导。；
**重要问题**：从式(14)到(15)，外层目标由对比损失 ℓ_con 变为对齐损失 ℓ_align；式(1)定义 ℓ_con 为 ℓ_align 与正则项之和。从式(12)到(13)未再看到 d(p,q′)≤τ 约束或对应惩罚。论文未解释这些变化与原三层目标的等价条件。；
**重要问题**：第3页将源域样本设为独立同分布；附录 B.3 使用窗口128、步长64，邻近窗口共享50%原始观测。论文未说明理论统计界如何处理这种窗口依赖。 | [评审](论文/P046/评审.md) · [评分数据](论文/P046/评审.json) · [复核](论文/P046/复核.md) |",
"raw_cells": [
"91",
"[P046 Tri-Level Navigator: LLM-Empowered Tri-Level Learning for Time Series OOD Generalization](https://proceedings.neurips.cc/paper_files/paper/2024/hash/c78d9d035215039ab46e48c281e6a63d-Abstract-Conference.html)",
"42.4",
"95%",
"53%",
"34.0–52.0",
"4",
"NeurIPS 2024 官方论文集 PDF，含正文及附录；本次未将 arXiv 版本等同于会议版",
"**关键问题**：式(53)的未归一化梯度平方和上界在定理5所列假设及步长下存在反例：取 F=(θ+q)²/4，θ⁰=q⁰=1、δ任意，t₁=0、T₁=1、L=1、真实梯度、噪声方差0、α₁²=α₂²=1、α₃²=0，步长1；更新后 θ¹=q¹=0，F*=0，而式(53)左边为2、右边为1。式(54)—(56)由此得到的迭代界未被所给证明建立；该反例针对证明步骤，未反驳速率命题或具体训练结果。；
**重要问题**：表3 USC-HAD 域 A 为 2,3,4，域 B 为 4,5,6，同一编号4同时列入两域；第20页表2写 subjects 为4，表3编号却延伸到14，论文未清楚定义这些编号和原始受试者的对应关系。若两处编号4指同一受试者，其数据会跨域；该受试者的记录或窗口是否重复、是否再作互斥划分均未知。；
**重要问题**：附录 B.3 指出按每个域的均值和标准差标准化，未交代目标域统计量估计区间、可用时点及是否进入训练；若用整个留出域估计，则测试使用了无标签目标分布信息。文中未报告使用目标标签。；
**重要问题**：式(11)中的 max_{i,j} d(P_Si,P_Sj) 对给定源分布及假设类没有混合权重 α；附录 A.2.3 却称动态调权可最小化该项。目标域差异项未被直接观测；若作者意图让分布随表示变化，论文未在该处给出相应推导。；
**重要问题**：从式(14)到(15)，外层目标由对比损失 ℓ_con 变为对齐损失 ℓ_align；式(1)定义 ℓ_con 为 ℓ_align 与正则项之和。从式(12)到(13)未再看到 d(p,q′)≤τ 约束或对应惩罚。论文未解释这些变化与原三层目标的等价条件。；
**重要问题**：第3页将源域样本设为独立同分布；附录 B.3 使用窗口128、步长64，邻近窗口共享50%原始观测。论文未说明理论统计界如何处理这种窗口依赖。",
"[评审](论文/P046/评审.md) · [评分数据](论文/P046/评审.json) · [复核](论文/P046/复核.md)"
],
"normalized_title": "trilevelnavigatorllmempoweredtrilevellearningfortimeseriesoodgeneralization",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/c78d9d035215039ab46e48c281e6a63d-abstract-conference.html",
"record_type": "legacy_scored",
"legacy_rank": 91,
"legacy_score": "42.4",
"legacy_coverage": "95%",
"legacy_tentative_share": "53%",
"legacy_range": "34.0–52.0",
"citation_verbatim": "4",
"legacy_review_version": "NeurIPS 2024 官方论文集 PDF，含正文及附录；本次未将 arXiv 版本等同于会议版",
"legacy_risks_verbatim": "**关键问题**：式(53)的未归一化梯度平方和上界在定理5所列假设及步长下存在反例：取 F=(θ+q)²/4，θ⁰=q⁰=1、δ任意，t₁=0、T₁=1、L=1、真实梯度、噪声方差0、α₁²=α₂²=1、α₃²=0，步长1；更新后 θ¹=q¹=0，F*=0，而式(53)左边为2、右边为1。式(54)—(56)由此得到的迭代界未被所给证明建立；该反例针对证明步骤，未反驳速率命题或具体训练结果。；
**重要问题**：表3 USC-HAD 域 A 为 2,3,4，域 B 为 4,5,6，同一编号4同时列入两域；第20页表2写 subjects 为4，表3编号却延伸到14，论文未清楚定义这些编号和原始受试者的对应关系。若两处编号4指同一受试者，其数据会跨域；该受试者的记录或窗口是否重复、是否再作互斥划分均未知。；
**重要问题**：附录 B.3 指出按每个域的均值和标准差标准化，未交代目标域统计量估计区间、可用时点及是否进入训练；若用整个留出域估计，则测试使用了无标签目标分布信息。文中未报告使用目标标签。；
**重要问题**：式(11)中的 max_{i,j} d(P_Si,P_Sj) 对给定源分布及假设类没有混合权重 α；附录 A.2.3 却称动态调权可最小化该项。目标域差异项未被直接观测；若作者意图让分布随表示变化，论文未在该处给出相应推导。；
**重要问题**：从式(14)到(15)，外层目标由对比损失 ℓ_con 变为对齐损失 ℓ_align；式(1)定义 ℓ_con 为 ℓ_align 与正则项之和。从式(12)到(13)未再看到 d(p,q′)≤τ 约束或对应惩罚。论文未解释这些变化与原三层目标的等价条件。；
**重要问题**：第3页将源域样本设为独立同分布；附录 B.3 使用窗口128、步长64，邻近窗口共享50%原始观测。论文未说明理论统计界如何处理这种窗口依赖。",
"legacy_material_links_verbatim": "[评审](论文/P046/评审.md) · [评分数据](论文/P046/评审.json) · [复核](论文/P046/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-145",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 108,
"source_number": 145,
"title": "SSL-STMFormer Self-Supervised Learning Spatio-Temporal Entanglement Transformer for Traffic Flow Prediction",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33321",
"raw_line": "| 92 | [P145 SSL-STMFormer Self-Supervised Learning Spatio-Temporal Entanglement Transformer for Traffic Flow Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33321) | 41.8 | 95% | 47% | 34.7–51.0 | 10 | AAAI-25 会议论文集出版版，9 页，DOI 10.1609/aaai.v39i11.33321 | **重要问题**：表 3 的 NYCTaxi 流入 RMSE、流出三指标和 T-Drive 流入 MAE 的 Improve 与所列基线及本模型误差不符。按六项流入流出的显示误差值分别算降幅再等权平均，网格 MAE/MAPE/RMSE 约为 11.51%/10.62%/4.83%；正文的 10.91%/9.86%/7.18% 与表中印出的 Improve 行的六项等权均值一致。论文未明确平均公式。；
**重要问题**：式 (6) 将二值图掩码乘到 softmax 之前的有限注意力分数：掩码为 0 的分量变为 0，softmax 后仍有正权重，与硬排除相应节点的解释存在差别。；
**重要问题**：所读正文给出切分比例和点误差，未报告独立重复、方差、置信区间、各基线调参预算、DTW 统计期或时间切分边界。 | [评审](论文/P145/评审.md) · [评分数据](论文/P145/评审.json) · [复核](论文/P145/复核.md) |",
"raw_cells": [
"92",
"[P145 SSL-STMFormer Self-Supervised Learning Spatio-Temporal Entanglement Transformer for Traffic Flow Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33321)",
"41.8",
"95%",
"47%",
"34.7–51.0",
"10",
"AAAI-25 会议论文集出版版，9 页，DOI 10.1609/aaai.v39i11.33321",
"**重要问题**：表 3 的 NYCTaxi 流入 RMSE、流出三指标和 T-Drive 流入 MAE 的 Improve 与所列基线及本模型误差不符。按六项流入流出的显示误差值分别算降幅再等权平均，网格 MAE/MAPE/RMSE 约为 11.51%/10.62%/4.83%；正文的 10.91%/9.86%/7.18% 与表中印出的 Improve 行的六项等权均值一致。论文未明确平均公式。；
**重要问题**：式 (6) 将二值图掩码乘到 softmax 之前的有限注意力分数：掩码为 0 的分量变为 0，softmax 后仍有正权重，与硬排除相应节点的解释存在差别。；
**重要问题**：所读正文给出切分比例和点误差，未报告独立重复、方差、置信区间、各基线调参预算、DTW 统计期或时间切分边界。",
"[评审](论文/P145/评审.md) · [评分数据](论文/P145/评审.json) · [复核](论文/P145/复核.md)"
],
"normalized_title": "sslstmformerselfsupervisedlearningspatiotemporalentanglementtransformerfortrafficflowprediction",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33321",
"record_type": "legacy_scored",
"legacy_rank": 92,
"legacy_score": "41.8",
"legacy_coverage": "95%",
"legacy_tentative_share": "47%",
"legacy_range": "34.7–51.0",
"citation_verbatim": "10",
"legacy_review_version": "AAAI-25 会议论文集出版版，9 页，DOI 10.1609/aaai.v39i11.33321",
"legacy_risks_verbatim": "**重要问题**：表 3 的 NYCTaxi 流入 RMSE、流出三指标和 T-Drive 流入 MAE 的 Improve 与所列基线及本模型误差不符。按六项流入流出的显示误差值分别算降幅再等权平均，网格 MAE/MAPE/RMSE 约为 11.51%/10.62%/4.83%；正文的 10.91%/9.86%/7.18% 与表中印出的 Improve 行的六项等权均值一致。论文未明确平均公式。；
**重要问题**：式 (6) 将二值图掩码乘到 softmax 之前的有限注意力分数：掩码为 0 的分量变为 0，softmax 后仍有正权重，与硬排除相应节点的解释存在差别。；
**重要问题**：所读正文给出切分比例和点误差，未报告独立重复、方差、置信区间、各基线调参预算、DTW 统计期或时间切分边界。",
"legacy_material_links_verbatim": "[评审](论文/P145/评审.md) · [评分数据](论文/P145/评审.json) · [复核](论文/P145/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-028",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 109,
"source_number": 28,
"title": "Large Scale Hierarchical Industrial Demand Time-Series Forecasting incorporating Sparsity",
"source_url": "https://doi.org/10.1145/3637528.3671632",
"raw_line": "| 93 | [P028 Large Scale Hierarchical Industrial Demand Time-Series Forecasting incorporating Sparsity](https://doi.org/10.1145/3637528.3671632) | 41.2 | 100% | 40% | 37.7–45.7 | 7 | arXiv:2407.02657v1（带 KDD 2024 会议信息的预印本；ACM 终稿未获得） | **关键问题**：针对所读 v1 的书面稀疏父子分支：式(5)无 φ 权重，且将无界表达称作有界标准 JSD。假设父节点 i=1、两子节点泊松均值各为1且权重各为1/2，父均值为1与目标均值一致但该式为 ln2，而父均值2时该式为0且偏离加权均值。均值一致是分布一致的必要条件。实际数据是否出现该具体节点配置及代码使用的损失未知。；
**重要问题**：v1 把 M5 描述为3914条、月度且预测28个月，并按训练均值加权；M5 官方指南列出42840条、日度28天与近期28天销售额权重。论文若作子集或时间聚合，具体映射未在所读版本解释。；
**重要问题**：所读 v1 以拟合高斯后的归一化 CRPS 及两个示例区间支持校准说法，未报告分层经验覆盖率或校准曲线；本文也未给出非交换共形覆盖结果。；
**重要问题**：v1 将稀疏层收益归因于泊松选择和 DCRS，但相对 ProfHiT 前作同时涉及基础模型和训练等多项设计；所读版本没有逐项消融和重复试验统计，表中 ProfHiT 的具体配置不明。；
**重要问题**：表2声称各层最好，但 L1、L3、L4、L8 的 HAILS 值并非最低；RMSSE展示公式从 n 求和至 n+N，包含 N+1 项，与 N 项预测定义有差异。正文称相对 ProfHiT 的 CRPS 改善超过40%但未给出计算口径；按表2总体行0.132至0.081计算约下降38.64%，两者口径关系未知。；
**重要问题**：所读 v1 对基线输入特征、调参资源、M5切分及 M5-Leader 在 CRPS 表中的产生方式未充分交代；各节点训练／验证比例为80:20，划分的时间顺序未说明，未来外生变量的可用时点也未知。 | [评审](论文/P028/评审.md) · [评分数据](论文/P028/评审.json) · [复核](论文/P028/复核.md) |",
"raw_cells": [
"93",
"[P028 Large Scale Hierarchical Industrial Demand Time-Series Forecasting incorporating Sparsity](https://doi.org/10.1145/3637528.3671632)",
"41.2",
"100%",
"40%",
"37.7–45.7",
"7",
"arXiv:2407.02657v1（带 KDD 2024 会议信息的预印本；ACM 终稿未获得）",
"**关键问题**：针对所读 v1 的书面稀疏父子分支：式(5)无 φ 权重，且将无界表达称作有界标准 JSD。假设父节点 i=1、两子节点泊松均值各为1且权重各为1/2，父均值为1与目标均值一致但该式为 ln2，而父均值2时该式为0且偏离加权均值。均值一致是分布一致的必要条件。实际数据是否出现该具体节点配置及代码使用的损失未知。；
**重要问题**：v1 把 M5 描述为3914条、月度且预测28个月，并按训练均值加权；M5 官方指南列出42840条、日度28天与近期28天销售额权重。论文若作子集或时间聚合，具体映射未在所读版本解释。；
**重要问题**：所读 v1 以拟合高斯后的归一化 CRPS 及两个示例区间支持校准说法，未报告分层经验覆盖率或校准曲线；本文也未给出非交换共形覆盖结果。；
**重要问题**：v1 将稀疏层收益归因于泊松选择和 DCRS，但相对 ProfHiT 前作同时涉及基础模型和训练等多项设计；所读版本没有逐项消融和重复试验统计，表中 ProfHiT 的具体配置不明。；
**重要问题**：表2声称各层最好，但 L1、L3、L4、L8 的 HAILS 值并非最低；RMSSE展示公式从 n 求和至 n+N，包含 N+1 项，与 N 项预测定义有差异。正文称相对 ProfHiT 的 CRPS 改善超过40%但未给出计算口径；按表2总体行0.132至0.081计算约下降38.64%，两者口径关系未知。；
**重要问题**：所读 v1 对基线输入特征、调参资源、M5切分及 M5-Leader 在 CRPS 表中的产生方式未充分交代；各节点训练／验证比例为80:20，划分的时间顺序未说明，未来外生变量的可用时点也未知。",
"[评审](论文/P028/评审.md) · [评分数据](论文/P028/评审.json) · [复核](论文/P028/复核.md)"
],
"normalized_title": "largescalehierarchicalindustrialdemandtimeseriesforecastingincorporatingsparsity",
"normalized_url": "doi.org/10.1145/3637528.3671632",
"record_type": "legacy_scored",
"legacy_rank": 93,
"legacy_score": "41.2",
"legacy_coverage": "100%",
"legacy_tentative_share": "40%",
"legacy_range": "37.7–45.7",
"citation_verbatim": "7",
"legacy_review_version": "arXiv:2407.02657v1（带 KDD 2024 会议信息的预印本；ACM 终稿未获得）",
"legacy_risks_verbatim": "**关键问题**：针对所读 v1 的书面稀疏父子分支：式(5)无 φ 权重，且将无界表达称作有界标准 JSD。假设父节点 i=1、两子节点泊松均值各为1且权重各为1/2，父均值为1与目标均值一致但该式为 ln2，而父均值2时该式为0且偏离加权均值。均值一致是分布一致的必要条件。实际数据是否出现该具体节点配置及代码使用的损失未知。；
**重要问题**：v1 把 M5 描述为3914条、月度且预测28个月，并按训练均值加权；M5 官方指南列出42840条、日度28天与近期28天销售额权重。论文若作子集或时间聚合，具体映射未在所读版本解释。；
**重要问题**：所读 v1 以拟合高斯后的归一化 CRPS 及两个示例区间支持校准说法，未报告分层经验覆盖率或校准曲线；本文也未给出非交换共形覆盖结果。；
**重要问题**：v1 将稀疏层收益归因于泊松选择和 DCRS，但相对 ProfHiT 前作同时涉及基础模型和训练等多项设计；所读版本没有逐项消融和重复试验统计，表中 ProfHiT 的具体配置不明。；
**重要问题**：表2声称各层最好，但 L1、L3、L4、L8 的 HAILS 值并非最低；RMSSE展示公式从 n 求和至 n+N，包含 N+1 项，与 N 项预测定义有差异。正文称相对 ProfHiT 的 CRPS 改善超过40%但未给出计算口径；按表2总体行0.132至0.081计算约下降38.64%，两者口径关系未知。；
**重要问题**：所读 v1 对基线输入特征、调参资源、M5切分及 M5-Leader 在 CRPS 表中的产生方式未充分交代；各节点训练／验证比例为80:20，划分的时间顺序未说明，未来外生变量的可用时点也未知。",
"legacy_material_links_verbatim": "[评审](论文/P028/评审.md) · [评分数据](论文/P028/评审.json) · [复核](论文/P028/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-156",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 110,
"source_number": 156,
"title": "Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals",
"source_url": "https://doi.org/10.1145/3696410.3714931",
"raw_line": "| 94 | [P156 Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals](https://doi.org/10.1145/3696410.3714931)
评审题名：From Orthogonality to Dependency: Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals；[版本核对](论文/P156/版本核对.md) | 40.5 | 95% | 58% | 32.7–49.7 | 7 | arXiv:2405.16083v1 预印本；非WWW 2025会议终稿 | **重要问题**：定理A2只写组内分量两两条件独立；式(30)将当期共享与私有分量全部写成仅以历史为条件的乘积。式(28)使用相邻时刻条件分布匹配，而定理文字明确写逐时跨模态联合分布匹配；式(34)–(35)中把含历史的条件密度混合偏导置零的理由不足。所举例子针对具体步骤，未证明满足全部定理假设。；
**重要问题**：表1的D1NAMO ResNet Accuracy/Macro-F1为91.88/65.00，附表A5为88.64/88.58；主表MATE行和附表MATED行均为93.31/73.72。附表A6.2.2称A5为完整分类结果。按A5所印数值，ResNet的Macro-F1高于MATED 14.86；表中未说明MATED名称及跨表条件差异。；
**重要问题**：正文式(3)写成无概率权重的对数项积分；式(4)与附录式(47)的证据下界左侧分别写p和log p。式(11)–(12)将对数项之和直接相乘作为密度；式(7)取log余弦，若最小化且权重为正，其方向与文字所称的对齐目标不一致，非正余弦下实数对数无定义。；
**重要问题**：第5节及A5–A6未清楚报告完整训练/测试分割、受试者隔离、重叠窗口处理、线性探测预训练的标签使用；表A4与A6对MIT-BIH、Motion和部分活动数据的具体描述不一致。D1NAMO的完整数据集描述和实验选取模态须区分。；
**重要问题**：表A5、A6有括号中的数值，本版文字未解释统计量和重复次数；表3–5未给出相应波动。多项优势幅度较小。；
**重要问题**：图A1比较多个损失消融，图A2展示t-SNE；此版未报告真实潜变量恢复、依赖强度控制或稳健性直接评测。稳健性出现在理论讨论的预期中。 | [评审](论文/P156/评审.md) · [评分数据](论文/P156/评审.json) · [复核](论文/P156/复核.md) |",
"raw_cells": [
"94",
"[P156 Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals](https://doi.org/10.1145/3696410.3714931)
评审题名：From Orthogonality to Dependency: Learning Disentangled Representation for Multi-Modal Time-Series Sensing Signals；[版本核对](论文/P156/版本核对.md)",
"40.5",
"95%",
"58%",
"32.7–49.7",
"7",
"arXiv:2405.16083v1 预印本；非WWW 2025会议终稿",
"**重要问题**：定理A2只写组内分量两两条件独立；式(30)将当期共享与私有分量全部写成仅以历史为条件的乘积。式(28)使用相邻时刻条件分布匹配，而定理文字明确写逐时跨模态联合分布匹配；式(34)–(35)中把含历史的条件密度混合偏导置零的理由不足。所举例子针对具体步骤，未证明满足全部定理假设。；
**重要问题**：表1的D1NAMO ResNet Accuracy/Macro-F1为91.88/65.00，附表A5为88.64/88.58；主表MATE行和附表MATED行均为93.31/73.72。附表A6.2.2称A5为完整分类结果。按A5所印数值，ResNet的Macro-F1高于MATED 14.86；表中未说明MATED名称及跨表条件差异。；
**重要问题**：正文式(3)写成无概率权重的对数项积分；式(4)与附录式(47)的证据下界左侧分别写p和log p。式(11)–(12)将对数项之和直接相乘作为密度；式(7)取log余弦，若最小化且权重为正，其方向与文字所称的对齐目标不一致，非正余弦下实数对数无定义。；
**重要问题**：第5节及A5–A6未清楚报告完整训练/测试分割、受试者隔离、重叠窗口处理、线性探测预训练的标签使用；表A4与A6对MIT-BIH、Motion和部分活动数据的具体描述不一致。D1NAMO的完整数据集描述和实验选取模态须区分。；
**重要问题**：表A5、A6有括号中的数值，本版文字未解释统计量和重复次数；表3–5未给出相应波动。多项优势幅度较小。；
**重要问题**：图A1比较多个损失消融，图A2展示t-SNE；此版未报告真实潜变量恢复、依赖强度控制或稳健性直接评测。稳健性出现在理论讨论的预期中。",
"[评审](论文/P156/评审.md) · [评分数据](论文/P156/评审.json) · [复核](论文/P156/复核.md)"
],
"normalized_title": "learningdisentangledrepresentationformultimodaltimeseriessensingsignals",
"normalized_url": "doi.org/10.1145/3696410.3714931",
"record_type": "legacy_scored",
"legacy_rank": 94,
"legacy_score": "40.5",
"legacy_coverage": "95%",
"legacy_tentative_share": "58%",
"legacy_range": "32.7–49.7",
"citation_verbatim": "7",
"legacy_review_version": "arXiv:2405.16083v1 预印本；非WWW 2025会议终稿",
"legacy_risks_verbatim": "**重要问题**：定理A2只写组内分量两两条件独立；式(30)将当期共享与私有分量全部写成仅以历史为条件的乘积。式(28)使用相邻时刻条件分布匹配，而定理文字明确写逐时跨模态联合分布匹配；式(34)–(35)中把含历史的条件密度混合偏导置零的理由不足。所举例子针对具体步骤，未证明满足全部定理假设。；
**重要问题**：表1的D1NAMO ResNet Accuracy/Macro-F1为91.88/65.00，附表A5为88.64/88.58；主表MATE行和附表MATED行均为93.31/73.72。附表A6.2.2称A5为完整分类结果。按A5所印数值，ResNet的Macro-F1高于MATED 14.86；表中未说明MATED名称及跨表条件差异。；
**重要问题**：正文式(3)写成无概率权重的对数项积分；式(4)与附录式(47)的证据下界左侧分别写p和log p。式(11)–(12)将对数项之和直接相乘作为密度；式(7)取log余弦，若最小化且权重为正，其方向与文字所称的对齐目标不一致，非正余弦下实数对数无定义。；
**重要问题**：第5节及A5–A6未清楚报告完整训练/测试分割、受试者隔离、重叠窗口处理、线性探测预训练的标签使用；表A4与A6对MIT-BIH、Motion和部分活动数据的具体描述不一致。D1NAMO的完整数据集描述和实验选取模态须区分。；
**重要问题**：表A5、A6有括号中的数值，本版文字未解释统计量和重复次数；表3–5未给出相应波动。多项优势幅度较小。；
**重要问题**：图A1比较多个损失消融，图A2展示t-SNE；此版未报告真实潜变量恢复、依赖强度控制或稳健性直接评测。稳健性出现在理论讨论的预期中。",
"legacy_material_links_verbatim": "[评审](论文/P156/评审.md) · [评分数据](论文/P156/评审.json) · [复核](论文/P156/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-159",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 111,
"source_number": 159,
"title": "STEM-LTS: Integrating Semantic-Temporal Dynamics in LLM-driven Time Series Analysis",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34447",
"raw_line": "| 95 | [P159 STEM-LTS: Integrating Semantic-Temporal Dynamics in LLM-driven Time Series Analysis](https://ojs.aaai.org/index.php/AAAI/article/view/34447) | 34.2 | 95% | 84% | 24.5–49.7 | 7 | AAAI-25 会议出版 PDF（9页，pp. 22858–22866） | **重要问题**：式(10)的输出维度与三分量协方差不符；若另假设3×3样本协方差，式(9)在满足式(7)的负相关分量例子上低于全零分解。维度问题对应发表的式(10)，条件性反例依据式(9)并额外假设3×3协方差。；
**重要问题**：论文称STEM-LTS逐数据集微调，且若干基线值与TEMPO前作零样本表一致；基线具体训练条件和结果来源未交代，条件匹配性待核实。数值一致本身未确认结果沿用或不公平。；
**重要问题**：正文图3未定义关联对象或展示语义质量测量，正文也未呈现去除对齐或语言预训练的匹配消融；附录未获得。；
**重要问题**：摘要声称噪声鲁棒性，但已读正文和图表没有噪声类型、扰动强度及对应误差比较；附录未获得。 | [评审](论文/P159/评审.md) · [评分数据](论文/P159/评审.json) · [复核](论文/P159/复核.md) |",
"raw_cells": [
"95",
"[P159 STEM-LTS: Integrating Semantic-Temporal Dynamics in LLM-driven Time Series Analysis](https://ojs.aaai.org/index.php/AAAI/article/view/34447)",
"34.2",
"95%",
"84%",
"24.5–49.7",
"7",
"AAAI-25 会议出版 PDF（9页，pp. 22858–22866）",
"**重要问题**：式(10)的输出维度与三分量协方差不符；若另假设3×3样本协方差，式(9)在满足式(7)的负相关分量例子上低于全零分解。维度问题对应发表的式(10)，条件性反例依据式(9)并额外假设3×3协方差。；
**重要问题**：论文称STEM-LTS逐数据集微调，且若干基线值与TEMPO前作零样本表一致；基线具体训练条件和结果来源未交代，条件匹配性待核实。数值一致本身未确认结果沿用或不公平。；
**重要问题**：正文图3未定义关联对象或展示语义质量测量，正文也未呈现去除对齐或语言预训练的匹配消融；附录未获得。；
**重要问题**：摘要声称噪声鲁棒性，但已读正文和图表没有噪声类型、扰动强度及对应误差比较；附录未获得。",
"[评审](论文/P159/评审.md) · [评分数据](论文/P159/评审.json) · [复核](论文/P159/复核.md)"
],
"normalized_title": "stemltsintegratingsemantictemporaldynamicsinllmdriventimeseriesanalysis",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34447",
"record_type": "legacy_scored",
"legacy_rank": 95,
"legacy_score": "34.2",
"legacy_coverage": "95%",
"legacy_tentative_share": "84%",
"legacy_range": "24.5–49.7",
"citation_verbatim": "7",
"legacy_review_version": "AAAI-25 会议出版 PDF（9页，pp. 22858–22866）",
"legacy_risks_verbatim": "**重要问题**：式(10)的输出维度与三分量协方差不符；若另假设3×3样本协方差，式(9)在满足式(7)的负相关分量例子上低于全零分解。维度问题对应发表的式(10)，条件性反例依据式(9)并额外假设3×3协方差。；
**重要问题**：论文称STEM-LTS逐数据集微调，且若干基线值与TEMPO前作零样本表一致；基线具体训练条件和结果来源未交代，条件匹配性待核实。数值一致本身未确认结果沿用或不公平。；
**重要问题**：正文图3未定义关联对象或展示语义质量测量，正文也未呈现去除对齐或语言预训练的匹配消融；附录未获得。；
**重要问题**：摘要声称噪声鲁棒性，但已读正文和图表没有噪声类型、扰动强度及对应误差比较；附录未获得。",
"legacy_material_links_verbatim": "[评审](论文/P159/评审.md) · [评分数据](论文/P159/评审.json) · [复核](论文/P159/复核.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-003",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 119,
"source_number": 3,
"title": "SolarCube: An Integrative Benchmark Dataset Harnessing Satellite and In-situ Observations for Large-scale Solar Energy Forecasting",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/06477eb61ea6b85c6608d42a222462df-Abstract-Datasets_and_Benchmarks_Track.html",
"raw_line": "| P003 | [SolarCube: An Integrative Benchmark Dataset Harnessing Satellite and In-situ Observations for Large-scale Solar Energy Forecasting](https://proceedings.neurips.cc/paper_files/paper/2024/hash/06477eb61ea6b85c6608d42a222462df-Abstract-Datasets_and_Benchmarks_Track.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P003",
"[SolarCube: An Integrative Benchmark Dataset Harnessing Satellite and In-situ Observations for Large-scale Solar Energy Forecasting](https://proceedings.neurips.cc/paper_files/paper/2024/hash/06477eb61ea6b85c6608d42a222462df-Abstract-Datasets_and_Benchmarks_Track.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "solarcubeanintegrativebenchmarkdatasetharnessingsatelliteandinsituobservationsforlargescalesolarenergyforecasting",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/06477eb61ea6b85c6608d42a222462df-abstract-datasets_and_benchmarks_track.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-009",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 120,
"source_number": 9,
"title": "Biased Temporal Convolution Graph Network for Time Series Forecasting with Missing Values",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/7f3b25e39e0f0330435d1ffc1d6fa193-Abstract-Conference.html",
"raw_line": "| P009 | [Biased Temporal Convolution Graph Network for Time Series Forecasting with Missing Values](https://proceedings.iclr.cc/paper_files/paper/2024/hash/7f3b25e39e0f0330435d1ffc1d6fa193-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P009",
"[Biased Temporal Convolution Graph Network for Time Series Forecasting with Missing Values](https://proceedings.iclr.cc/paper_files/paper/2024/hash/7f3b25e39e0f0330435d1ffc1d6fa193-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "biasedtemporalconvolutiongraphnetworkfortimeseriesforecastingwithmissingvalues",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/7f3b25e39e0f0330435d1ffc1d6fa193-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-010",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 121,
"source_number": 10,
"title": "STanHop: Sparse Tandem Hopfield Model for Memory-Enhanced Time Series Prediction",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/832b20b65f655587e9c0447860406a82-Abstract-Conference.html",
"raw_line": "| P010 | [STanHop: Sparse Tandem Hopfield Model for Memory-Enhanced Time Series Prediction](https://proceedings.iclr.cc/paper_files/paper/2024/hash/832b20b65f655587e9c0447860406a82-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P010",
"[STanHop: Sparse Tandem Hopfield Model for Memory-Enhanced Time Series Prediction](https://proceedings.iclr.cc/paper_files/paper/2024/hash/832b20b65f655587e9c0447860406a82-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "stanhopsparsetandemhopfieldmodelformemoryenhancedtimeseriesprediction",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/832b20b65f655587e9c0447860406a82-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-011",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 122,
"source_number": 11,
"title": "Reservoir Computing for Short High-Dimensional Time Series: an Application to SARS-CoV-2 Hospitalization Forecast",
"source_url": "https://proceedings.mlr.press/v235/ferte24a.html",
"raw_line": "| P011 | [Reservoir Computing for Short High-Dimensional Time Series: an Application to SARS-CoV-2 Hospitalization Forecast](https://proceedings.mlr.press/v235/ferte24a.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P011",
"[Reservoir Computing for Short High-Dimensional Time Series: an Application to SARS-CoV-2 Hospitalization Forecast](https://proceedings.mlr.press/v235/ferte24a.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "reservoircomputingforshorthighdimensionaltimeseriesanapplicationtosarscov2hospitalizationforecast",
"normalized_url": "proceedings.mlr.press/v235/ferte24a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-013",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 123,
"source_number": 13,
"title": "Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization",
"source_url": "https://proceedings.mlr.press/v235/liu24aw.html",
"raw_line": "| P013 | [Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization](https://proceedings.mlr.press/v235/liu24aw.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P013",
"[Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization](https://proceedings.mlr.press/v235/liu24aw.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "deepfunctionalfactormodelsforecastinghighdimensionalfunctionaltimeseriesviabayesiannonparametricfactorization",
"normalized_url": "proceedings.mlr.press/v235/liu24aw.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-014",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 124,
"source_number": 14,
"title": "Interpretable Sparse System Identification: Beyond Recent Deep Learning Techniques on Time-Series Prediction",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/3304c2bbd18129a3cfcef5ea886db4a7-Abstract-Conference.html",
"raw_line": "| P014 | [Interpretable Sparse System Identification: Beyond Recent Deep Learning Techniques on Time-Series Prediction](https://proceedings.iclr.cc/paper_files/paper/2024/hash/3304c2bbd18129a3cfcef5ea886db4a7-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P014",
"[Interpretable Sparse System Identification: Beyond Recent Deep Learning Techniques on Time-Series Prediction](https://proceedings.iclr.cc/paper_files/paper/2024/hash/3304c2bbd18129a3cfcef5ea886db4a7-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "interpretablesparsesystemidentificationbeyondrecentdeeplearningtechniquesontimeseriesprediction",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/3304c2bbd18129a3cfcef5ea886db4a7-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-015",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 125,
"source_number": 15,
"title": "Towards Transparent Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/b5414313268287725e91a8eae9ddea01-Abstract-Conference.html",
"raw_line": "| P015 | [Towards Transparent Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2024/hash/b5414313268287725e91a8eae9ddea01-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P015",
"[Towards Transparent Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2024/hash/b5414313268287725e91a8eae9ddea01-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "towardstransparenttimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/b5414313268287725e91a8eae9ddea01-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-021",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 126,
"source_number": 21,
"title": "Explainable and Interpretable Forecasts on Non-Smooth Multivariate Time Series for Responsible Gameplay",
"source_url": "https://doi.org/10.1145/3637528.3671657",
"raw_line": "| P021 | [Explainable and Interpretable Forecasts on Non-Smooth Multivariate Time Series for Responsible Gameplay](https://doi.org/10.1145/3637528.3671657) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P021",
"[Explainable and Interpretable Forecasts on Non-Smooth Multivariate Time Series for Responsible Gameplay](https://doi.org/10.1145/3637528.3671657)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "explainableandinterpretableforecastsonnonsmoothmultivariatetimeseriesforresponsiblegameplay",
"normalized_url": "doi.org/10.1145/3637528.3671657",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-022",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 127,
"source_number": 22,
"title": "CGS-Mask: Making Time Series Predictions Intuitive for All",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/29325",
"raw_line": "| P022 | [CGS-Mask: Making Time Series Predictions Intuitive for All](https://ojs.aaai.org/index.php/AAAI/article/view/29325) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P022",
"[CGS-Mask: Making Time Series Predictions Intuitive for All](https://ojs.aaai.org/index.php/AAAI/article/view/29325)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "cgsmaskmakingtimeseriespredictionsintuitiveforall",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/29325",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-029",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 128,
"source_number": 29,
"title": "Quantifying and Estimating the Predictability Upper Bound of Univariate Numeric Time Series",
"source_url": "https://doi.org/10.1145/3637528.3671995",
"raw_line": "| P029 | [Quantifying and Estimating the Predictability Upper Bound of Univariate Numeric Time Series](https://doi.org/10.1145/3637528.3671995) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P029",
"[Quantifying and Estimating the Predictability Upper Bound of Univariate Numeric Time Series](https://doi.org/10.1145/3637528.3671995)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "quantifyingandestimatingthepredictabilityupperboundofunivariatenumerictimeseries",
"normalized_url": "doi.org/10.1145/3637528.3671995",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-031",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 129,
"source_number": 31,
"title": "Towards a Self-contained Data-driven Global Weather Forecasting Framework",
"source_url": "https://proceedings.mlr.press/v235/xiao24a.html",
"raw_line": "| P031 | [Towards a Self-contained Data-driven Global Weather Forecasting Framework](https://proceedings.mlr.press/v235/xiao24a.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P031",
"[Towards a Self-contained Data-driven Global Weather Forecasting Framework](https://proceedings.mlr.press/v235/xiao24a.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "towardsaselfcontaineddatadrivenglobalweatherforecastingframework",
"normalized_url": "proceedings.mlr.press/v235/xiao24a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-036",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 130,
"source_number": 36,
"title": "IncomeSCM: From tabular data set to time-series simulator and causal estimation benchmark",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/f5c7fb98367440a5eb01d70294c180ca-Abstract-Datasets_and_Benchmarks_Track.html",
"raw_line": "| P036 | [IncomeSCM: From tabular data set to time-series simulator and causal estimation benchmark](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f5c7fb98367440a5eb01d70294c180ca-Abstract-Datasets_and_Benchmarks_Track.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P036",
"[IncomeSCM: From tabular data set to time-series simulator and causal estimation benchmark](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f5c7fb98367440a5eb01d70294c180ca-Abstract-Datasets_and_Benchmarks_Track.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "incomescmfromtabulardatasettotimeseriessimulatorandcausalestimationbenchmark",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/f5c7fb98367440a5eb01d70294c180ca-abstract-datasets_and_benchmarks_track.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-038",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 131,
"source_number": 38,
"title": "Disentangling Time Series Representations via Contrastive Independence-of-Support on l-Variational Inference",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2024/hash/099ba28b20462dac7bf057d21c27a011-Abstract-Conference.html",
"raw_line": "| P038 | [Disentangling Time Series Representations via Contrastive Independence-of-Support on l-Variational Inference](https://proceedings.iclr.cc/paper_files/paper/2024/hash/099ba28b20462dac7bf057d21c27a011-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P038",
"[Disentangling Time Series Representations via Contrastive Independence-of-Support on l-Variational Inference](https://proceedings.iclr.cc/paper_files/paper/2024/hash/099ba28b20462dac7bf057d21c27a011-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "disentanglingtimeseriesrepresentationsviacontrastiveindependenceofsupportonlvariationalinference",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2024/hash/099ba28b20462dac7bf057d21c27a011-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-041",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 132,
"source_number": 41,
"title": "Learning diverse causally emergent representations from time series data",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/d8398f4da88975e2a9c62ecaa5ba267b-Abstract-Conference.html",
"raw_line": "| P041 | [Learning diverse causally emergent representations from time series data](https://proceedings.neurips.cc/paper_files/paper/2024/hash/d8398f4da88975e2a9c62ecaa5ba267b-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P041",
"[Learning diverse causally emergent representations from time series data](https://proceedings.neurips.cc/paper_files/paper/2024/hash/d8398f4da88975e2a9c62ecaa5ba267b-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "learningdiversecausallyemergentrepresentationsfromtimeseriesdata",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/d8398f4da88975e2a9c62ecaa5ba267b-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-043",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 133,
"source_number": 43,
"title": "Exploiting Representation Curvature for Boundary Detection in Time Series",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/0b7f639ef28a9035a71f7e0c04c1d681-Abstract-Conference.html",
"raw_line": "| P043 | [Exploiting Representation Curvature for Boundary Detection in Time Series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/0b7f639ef28a9035a71f7e0c04c1d681-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P043",
"[Exploiting Representation Curvature for Boundary Detection in Time Series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/0b7f639ef28a9035a71f7e0c04c1d681-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "exploitingrepresentationcurvatureforboundarydetectionintimeseries",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/0b7f639ef28a9035a71f7e0c04c1d681-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-049",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 134,
"source_number": 49,
"title": "MF-CLR: Multi-Frequency Contrastive Learning Representation for Time Series",
"source_url": "https://proceedings.mlr.press/v235/duan24b.html",
"raw_line": "| P049 | [MF-CLR: Multi-Frequency Contrastive Learning Representation for Time Series](https://proceedings.mlr.press/v235/duan24b.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P049",
"[MF-CLR: Multi-Frequency Contrastive Learning Representation for Time Series](https://proceedings.mlr.press/v235/duan24b.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "mfclrmultifrequencycontrastivelearningrepresentationfortimeseries",
"normalized_url": "proceedings.mlr.press/v235/duan24b.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-052",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 135,
"source_number": 52,
"title": "Vector Quantization Pretraining for EEG Time Series with Random Projection and Phase Alignment",
"source_url": "https://proceedings.mlr.press/v235/gui24a.html",
"raw_line": "| P052 | [Vector Quantization Pretraining for EEG Time Series with Random Projection and Phase Alignment](https://proceedings.mlr.press/v235/gui24a.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P052",
"[Vector Quantization Pretraining for EEG Time Series with Random Projection and Phase Alignment](https://proceedings.mlr.press/v235/gui24a.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "vectorquantizationpretrainingforeegtimeserieswithrandomprojectionandphasealignment",
"normalized_url": "proceedings.mlr.press/v235/gui24a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-053",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 136,
"source_number": 53,
"title": "ChronoEpilogi: Scalable Time Series Selection with Multiple Solutions",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/f24e8cc1c1c06a689850ee766a7357b2-Abstract-Conference.html",
"raw_line": "| P053 | [ChronoEpilogi: Scalable Time Series Selection with Multiple Solutions](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f24e8cc1c1c06a689850ee766a7357b2-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P053",
"[ChronoEpilogi: Scalable Time Series Selection with Multiple Solutions](https://proceedings.neurips.cc/paper_files/paper/2024/hash/f24e8cc1c1c06a689850ee766a7357b2-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "chronoepilogiscalabletimeseriesselectionwithmultiplesolutions",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/f24e8cc1c1c06a689850ee766a7357b2-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-054",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 137,
"source_number": 54,
"title": "Reinforced Cross-Domain Knowledge Distillation on Time Series Data",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/5424f6b74f475ad738b54888d609283a-Abstract-Conference.html",
"raw_line": "| P054 | [Reinforced Cross-Domain Knowledge Distillation on Time Series Data](https://proceedings.neurips.cc/paper_files/paper/2024/hash/5424f6b74f475ad738b54888d609283a-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P054",
"[Reinforced Cross-Domain Knowledge Distillation on Time Series Data](https://proceedings.neurips.cc/paper_files/paper/2024/hash/5424f6b74f475ad738b54888d609283a-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "reinforcedcrossdomainknowledgedistillationontimeseriesdata",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/5424f6b74f475ad738b54888d609283a-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-057",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 138,
"source_number": 57,
"title": "Learning Causal Relations from Subsampled Time Series with Two Time-Slices",
"source_url": "https://proceedings.mlr.press/v235/wu24p.html",
"raw_line": "| P057 | [Learning Causal Relations from Subsampled Time Series with Two Time-Slices](https://proceedings.mlr.press/v235/wu24p.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P057",
"[Learning Causal Relations from Subsampled Time Series with Two Time-Slices](https://proceedings.mlr.press/v235/wu24p.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "learningcausalrelationsfromsubsampledtimeserieswithtwotimeslices",
"normalized_url": "proceedings.mlr.press/v235/wu24p.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-058",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 139,
"source_number": 58,
"title": "Rough Transformers: Lightweight and Continuous Time Series Modelling through Signature Patching",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/bfe167fee3be2d862a56af82dee77720-Abstract-Conference.html",
"raw_line": "| P058 | [Rough Transformers: Lightweight and Continuous Time Series Modelling through Signature Patching](https://proceedings.neurips.cc/paper_files/paper/2024/hash/bfe167fee3be2d862a56af82dee77720-Abstract-Conference.html) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P058",
"[Rough Transformers: Lightweight and Continuous Time Series Modelling through Signature Patching](https://proceedings.neurips.cc/paper_files/paper/2024/hash/bfe167fee3be2d862a56af82dee77720-Abstract-Conference.html)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "roughtransformerslightweightandcontinuoustimeseriesmodellingthroughsignaturepatching",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/bfe167fee3be2d862a56af82dee77720-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-060",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 140,
"source_number": 60,
"title": "Shape analysis for time series",
"source_url": "https://proceedings.neurips.cc/paper_files/paper/2024/hash/ad86418f7bdfa685cd089e028efd75cd-Abstract-Conference.html",
"raw_line": "| P060 | [Shape analysis for time series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/ad86418f7bdfa685cd089e028efd75cd-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P060",
"[Shape analysis for time series](https://proceedings.neurips.cc/paper_files/paper/2024/hash/ad86418f7bdfa685cd089e028efd75cd-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "shapeanalysisfortimeseries",
"normalized_url": "proceedings.neurips.cc/paper_files/paper/2024/hash/ad86418f7bdfa685cd089e028efd75cd-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-062",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 141,
"source_number": 62,
"title": "Physiome-ODE: A Benchmark for Irregularly Sampled Multivariate Time-Series Forecasting Based on Biological ODEs",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/565f995643da6329cec701f26f8579f5-Abstract-Conference.html",
"raw_line": "| P062 | [Physiome-ODE: A Benchmark for Irregularly Sampled Multivariate Time-Series Forecasting Based on Biological ODEs](https://proceedings.iclr.cc/paper_files/paper/2025/hash/565f995643da6329cec701f26f8579f5-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P062",
"[Physiome-ODE: A Benchmark for Irregularly Sampled Multivariate Time-Series Forecasting Based on Biological ODEs](https://proceedings.iclr.cc/paper_files/paper/2025/hash/565f995643da6329cec701f26f8579f5-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "physiomeodeabenchmarkforirregularlysampledmultivariatetimeseriesforecastingbasedonbiologicalodes",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/565f995643da6329cec701f26f8579f5-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-063",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 142,
"source_number": 63,
"title": "Shifting Time: Time-series Forecasting with Khatri-Rao Neural Operators",
"source_url": "https://proceedings.mlr.press/v267/dama25a.html",
"raw_line": "| P063 | [Shifting Time: Time-series Forecasting with Khatri-Rao Neural Operators](https://proceedings.mlr.press/v267/dama25a.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P063",
"[Shifting Time: Time-series Forecasting with Khatri-Rao Neural Operators](https://proceedings.mlr.press/v267/dama25a.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "shiftingtimetimeseriesforecastingwithkhatriraoneuraloperators",
"normalized_url": "proceedings.mlr.press/v267/dama25a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-067",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 143,
"source_number": 67,
"title": "Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting",
"source_url": "https://papers.neurips.cc/paper_files/paper/2025/hash/e4a04155055153f18b93824825cd6e3d-Abstract-Conference.html",
"raw_line": "| P067 | [Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/e4a04155055153f18b93824825cd6e3d-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P067",
"[Learning with Calibration: Exploring Test-Time Computing of Spatio-Temporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/e4a04155055153f18b93824825cd6e3d-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "learningwithcalibrationexploringtesttimecomputingofspatiotemporalforecasting",
"normalized_url": "papers.neurips.cc/paper_files/paper/2025/hash/e4a04155055153f18b93824825cd6e3d-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-068",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 144,
"source_number": 68,
"title": "Information Bottleneck-guided MLPs for Robust Spatial-temporal Forecasting",
"source_url": "https://proceedings.mlr.press/v267/chen25ax.html",
"raw_line": "| P068 | [Information Bottleneck-guided MLPs for Robust Spatial-temporal Forecasting](https://proceedings.mlr.press/v267/chen25ax.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P068",
"[Information Bottleneck-guided MLPs for Robust Spatial-temporal Forecasting](https://proceedings.mlr.press/v267/chen25ax.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "informationbottleneckguidedmlpsforrobustspatialtemporalforecasting",
"normalized_url": "proceedings.mlr.press/v267/chen25ax.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-069",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 145,
"source_number": 69,
"title": "FSTLLM: Spatio-Temporal LLM for Few Shot Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v267/jiang25a.html",
"raw_line": "| P069 | [FSTLLM: Spatio-Temporal LLM for Few Shot Time Series Forecasting](https://proceedings.mlr.press/v267/jiang25a.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P069",
"[FSTLLM: Spatio-Temporal LLM for Few Shot Time Series Forecasting](https://proceedings.mlr.press/v267/jiang25a.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "fstllmspatiotemporalllmforfewshottimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v267/jiang25a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-071",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 146,
"source_number": 71,
"title": "When Will It Fail?: Anomaly to Prompt for Forecasting Future Anomalies in Time Series",
"source_url": "https://proceedings.mlr.press/v267/park25e.html",
"raw_line": "| P071 | [When Will It Fail?: Anomaly to Prompt for Forecasting Future Anomalies in Time Series](https://proceedings.mlr.press/v267/park25e.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P071",
"[When Will It Fail?: Anomaly to Prompt for Forecasting Future Anomalies in Time Series](https://proceedings.mlr.press/v267/park25e.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "whenwillitfailanomalytopromptforforecastingfutureanomaliesintimeseries",
"normalized_url": "proceedings.mlr.press/v267/park25e.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-072",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 147,
"source_number": 72,
"title": "Investigating Pattern Neurons in Urban Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/1fb38e9c776b5ff18511a284570b15c8-Abstract-Conference.html",
"raw_line": "| P072 | [Investigating Pattern Neurons in Urban Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/1fb38e9c776b5ff18511a284570b15c8-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P072",
"[Investigating Pattern Neurons in Urban Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/1fb38e9c776b5ff18511a284570b15c8-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "investigatingpatternneuronsinurbantimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/1fb38e9c776b5ff18511a284570b15c8-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-073",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 148,
"source_number": 73,
"title": "Locally Connected Echo State Networks for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/d8bd445c2abe1343cce0e14b361b2fb3-Abstract-Conference.html",
"raw_line": "| P073 | [Locally Connected Echo State Networks for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/d8bd445c2abe1343cce0e14b361b2fb3-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P073",
"[Locally Connected Echo State Networks for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/d8bd445c2abe1343cce0e14b361b2fb3-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "locallyconnectedechostatenetworksfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/d8bd445c2abe1343cce0e14b361b2fb3-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-074",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 149,
"source_number": 74,
"title": "TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/45bd303ea32a8193d3c19de5d5eaf138-Abstract-Conference.html",
"raw_line": "| P074 | [TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/45bd303ea32a8193d3c19de5d5eaf138-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P074",
"[TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/45bd303ea32a8193d3c19de5d5eaf138-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "tslifatemporalsegmentspikingneuronnetworkfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/45bd303ea32a8193d3c19de5d5eaf138-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-076",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 150,
"source_number": 76,
"title": "Privacy Amplification by Structured Subsampling for Deep Differentially Private Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v267/schuchardt25a.html",
"raw_line": "| P076 | [Privacy Amplification by Structured Subsampling for Deep Differentially Private Time Series Forecasting](https://proceedings.mlr.press/v267/schuchardt25a.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P076",
"[Privacy Amplification by Structured Subsampling for Deep Differentially Private Time Series Forecasting](https://proceedings.mlr.press/v267/schuchardt25a.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "privacyamplificationbystructuredsubsamplingfordeepdifferentiallyprivatetimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v267/schuchardt25a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-077",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 151,
"source_number": 77,
"title": "KooNPro: A Variance-Aware Koopman Probabilistic Model Enhanced by Neural Process for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/4846257e355f6923fc2a1fbe35099e91-Abstract-Conference.html",
"raw_line": "| P077 | [KooNPro: A Variance-Aware Koopman Probabilistic Model Enhanced by Neural Process for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/4846257e355f6923fc2a1fbe35099e91-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P077",
"[KooNPro: A Variance-Aware Koopman Probabilistic Model Enhanced by Neural Process for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/4846257e355f6923fc2a1fbe35099e91-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "koonproavarianceawarekoopmanprobabilisticmodelenhancedbyneuralprocessfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/4846257e355f6923fc2a1fbe35099e91-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-078",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 152,
"source_number": 78,
"title": "CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/6ca155a091aedc939d289df8f16f6c75-Abstract-Conference.html",
"raw_line": "| P078 | [CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution](https://proceedings.iclr.cc/paper_files/paper/2025/hash/6ca155a091aedc939d289df8f16f6c75-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P078",
"[CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution](https://proceedings.iclr.cc/paper_files/paper/2025/hash/6ca155a091aedc939d289df8f16f6c75-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "comressemisupervisedtimeseriesforecastingutilizingconsensuspromotionofmultiresolution",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/6ca155a091aedc939d289df8f16f6c75-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-079",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 153,
"source_number": 79,
"title": "A Lightweight Sparse Interaction Network for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33452",
"raw_line": "| P079 | [A Lightweight Sparse Interaction Network for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33452) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P079",
"[A Lightweight Sparse Interaction Network for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/33452)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "alightweightsparseinteractionnetworkfortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33452",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-081",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 154,
"source_number": 81,
"title": "Probabilistic Hypergraph Recurrent Neural Networks for Time-series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709202",
"raw_line": "| P081 | [Probabilistic Hypergraph Recurrent Neural Networks for Time-series Forecasting](https://doi.org/10.1145/3690624.3709202) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P081",
"[Probabilistic Hypergraph Recurrent Neural Networks for Time-series Forecasting](https://doi.org/10.1145/3690624.3709202)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "probabilistichypergraphrecurrentneuralnetworksfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709202",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-082",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 155,
"source_number": 82,
"title": "CMA: A Unified Contextual Meta-Adaptation Methodology for Time-Series Denoising and Prediction",
"source_url": "https://doi.org/10.1145/3711896.3736881",
"raw_line": "| P082 | [CMA: A Unified Contextual Meta-Adaptation Methodology for Time-Series Denoising and Prediction](https://doi.org/10.1145/3711896.3736881) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P082",
"[CMA: A Unified Contextual Meta-Adaptation Methodology for Time-Series Denoising and Prediction](https://doi.org/10.1145/3711896.3736881)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "cmaaunifiedcontextualmetaadaptationmethodologyfortimeseriesdenoisingandprediction",
"normalized_url": "doi.org/10.1145/3711896.3736881",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-088",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 156,
"source_number": 88,
"title": "Apollo-Forecast: Overcoming Aliasing and Inference Speed Challenges in Language Models for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34371",
"raw_line": "| P088 | [Apollo-Forecast: Overcoming Aliasing and Inference Speed Challenges in Language Models for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34371) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P088",
"[Apollo-Forecast: Overcoming Aliasing and Inference Speed Challenges in Language Models for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34371)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "apolloforecastovercomingaliasingandinferencespeedchallengesinlanguagemodelsfortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34371",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-089",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 157,
"source_number": 89,
"title": "Quantum Time-index Models with Reservoir for Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3690624.3709228",
"raw_line": "| P089 | [Quantum Time-index Models with Reservoir for Time Series Forecasting](https://doi.org/10.1145/3690624.3709228) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P089",
"[Quantum Time-index Models with Reservoir for Time Series Forecasting](https://doi.org/10.1145/3690624.3709228)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "quantumtimeindexmodelswithreservoirfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3690624.3709228",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-090",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 158,
"source_number": 90,
"title": "Probabilistic Forecasting of Irregularly Sampled Time Series with Missing Values via Conditional Normalizing Flows",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/35494",
"raw_line": "| P090 | [Probabilistic Forecasting of Irregularly Sampled Time Series with Missing Values via Conditional Normalizing Flows](https://ojs.aaai.org/index.php/AAAI/article/view/35494) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P090",
"[Probabilistic Forecasting of Irregularly Sampled Time Series with Missing Values via Conditional Normalizing Flows](https://ojs.aaai.org/index.php/AAAI/article/view/35494)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "probabilisticforecastingofirregularlysampledtimeserieswithmissingvaluesviaconditionalnormalizingflows",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/35494",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-091",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 159,
"source_number": 91,
"title": "Test-Time Training with Diversified Local Aggregation Consistency for Mortality Prediction using Clinical Time Series",
"source_url": "https://doi.org/10.1145/3711896.3737151",
"raw_line": "| P091 | [Test-Time Training with Diversified Local Aggregation Consistency for Mortality Prediction using Clinical Time Series](https://doi.org/10.1145/3711896.3737151) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P091",
"[Test-Time Training with Diversified Local Aggregation Consistency for Mortality Prediction using Clinical Time Series](https://doi.org/10.1145/3711896.3737151)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "testtimetrainingwithdiversifiedlocalaggregationconsistencyformortalitypredictionusingclinicaltimeseries",
"normalized_url": "doi.org/10.1145/3711896.3737151",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-094",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 160,
"source_number": 94,
"title": "Timing is important: Risk-aware Fund Allocation based on Time-Series Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737268",
"raw_line": "| P094 | [Timing is important: Risk-aware Fund Allocation based on Time-Series Forecasting](https://doi.org/10.1145/3711896.3737268) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P094",
"[Timing is important: Risk-aware Fund Allocation based on Time-Series Forecasting](https://doi.org/10.1145/3711896.3737268)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "timingisimportantriskawarefundallocationbasedontimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737268",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-099",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 161,
"source_number": 99,
"title": "Semantics-Aware Patch Encoding and Hierarchical Dependency Modeling for Long-Term Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737123",
"raw_line": "| P099 | [Semantics-Aware Patch Encoding and Hierarchical Dependency Modeling for Long-Term Time Series Forecasting](https://doi.org/10.1145/3711896.3737123) | 4 | 失败 | 评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（4 次，低于 10 次门槛） |",
"raw_cells": [
"P099",
"[Semantics-Aware Patch Encoding and Hierarchical Dependency Modeling for Long-Term Time Series Forecasting](https://doi.org/10.1145/3711896.3737123)",
"4",
"失败",
"评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（4 次，低于 10 次门槛）"
],
"normalized_title": "semanticsawarepatchencodingandhierarchicaldependencymodelingforlongtermtimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737123",
"record_type": "legacy_unranked",
"citation_verbatim": "4",
"legacy_status_verbatim": "失败",
"legacy_unranked_reason_verbatim": "评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（4 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-105",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 162,
"source_number": 105,
"title": "HyperMixer: Specializable Hypergraph Channel Mixing for Long-term Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34302",
"raw_line": "| P105 | [HyperMixer: Specializable Hypergraph Channel Mixing for Long-term Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34302) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P105",
"[HyperMixer: Specializable Hypergraph Channel Mixing for Long-term Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/34302)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "hypermixerspecializablehypergraphchannelmixingforlongtermmultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34302",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-107",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 163,
"source_number": 107,
"title": "Beyond Fixed Variables: Expanding-variate Time Series Forecasting via Flat Scheme and Spatio-temporal Focal Learning",
"source_url": "https://doi.org/10.1145/3711896.3736854",
"raw_line": "| P107 | [Beyond Fixed Variables: Expanding-variate Time Series Forecasting via Flat Scheme and Spatio-temporal Focal Learning](https://doi.org/10.1145/3711896.3736854) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P107",
"[Beyond Fixed Variables: Expanding-variate Time Series Forecasting via Flat Scheme and Spatio-temporal Focal Learning](https://doi.org/10.1145/3711896.3736854)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "beyondfixedvariablesexpandingvariatetimeseriesforecastingviaflatschemeandspatiotemporalfocallearning",
"normalized_url": "doi.org/10.1145/3711896.3736854",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-109",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 164,
"source_number": 109,
"title": "Stochastic Diffusion: A Diffusion Based Model for Stochastic Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737137",
"raw_line": "| P109 | [Stochastic Diffusion: A Diffusion Based Model for Stochastic Time Series Forecasting](https://doi.org/10.1145/3711896.3737137) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P109",
"[Stochastic Diffusion: A Diffusion Based Model for Stochastic Time Series Forecasting](https://doi.org/10.1145/3711896.3737137)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "stochasticdiffusionadiffusionbasedmodelforstochastictimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737137",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-116",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 165,
"source_number": 116,
"title": "Mesh Interpolation Graph Network for Dynamic and Spatially Irregular Global Weather Forecasting",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/23cf4f3fd33c2fb071fc40aee0ec2884-Abstract-Conference.html",
"raw_line": "| P116 | [Mesh Interpolation Graph Network for Dynamic and Spatially Irregular Global Weather Forecasting](https://papers.nips.cc/paper_files/paper/2025/hash/23cf4f3fd33c2fb071fc40aee0ec2884-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P116",
"[Mesh Interpolation Graph Network for Dynamic and Spatially Irregular Global Weather Forecasting](https://papers.nips.cc/paper_files/paper/2025/hash/23cf4f3fd33c2fb071fc40aee0ec2884-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "meshinterpolationgraphnetworkfordynamicandspatiallyirregularglobalweatherforecasting",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/23cf4f3fd33c2fb071fc40aee0ec2884-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-117",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 166,
"source_number": 117,
"title": "DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/da8a39bc39ae1c89dd6ebb1e3bcbb3f3-Abstract-Conference.html",
"raw_line": "| P117 | [DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space](https://papers.nips.cc/paper_files/paper/2025/hash/da8a39bc39ae1c89dd6ebb1e3bcbb3f3-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P117",
"[DAWP: A framework for global observation forecasting via Data Assimilation and Weather Prediction in satellite observation space](https://papers.nips.cc/paper_files/paper/2025/hash/da8a39bc39ae1c89dd6ebb1e3bcbb3f3-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "dawpaframeworkforglobalobservationforecastingviadataassimilationandweatherpredictioninsatelliteobservationspace",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/da8a39bc39ae1c89dd6ebb1e3bcbb3f3-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-118",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 167,
"source_number": 118,
"title": "Decision-aware Training of Spatiotemporal Forecasting Models to Select a Top-K Subset of Sites for Intervention",
"source_url": "https://proceedings.mlr.press/v267/heuton25a.html",
"raw_line": "| P118 | [Decision-aware Training of Spatiotemporal Forecasting Models to Select a Top-K Subset of Sites for Intervention](https://proceedings.mlr.press/v267/heuton25a.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P118",
"[Decision-aware Training of Spatiotemporal Forecasting Models to Select a Top-K Subset of Sites for Intervention](https://proceedings.mlr.press/v267/heuton25a.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "decisionawaretrainingofspatiotemporalforecastingmodelstoselectatopksubsetofsitesforintervention",
"normalized_url": "proceedings.mlr.press/v267/heuton25a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-119",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 168,
"source_number": 119,
"title": "HR-Extreme: A High-Resolution Dataset for Extreme Weather Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/3236f661fc6f00e8b16ff42db99acaaf-Abstract-Conference.html",
"raw_line": "| P119 | [HR-Extreme: A High-Resolution Dataset for Extreme Weather Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/3236f661fc6f00e8b16ff42db99acaaf-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P119",
"[HR-Extreme: A High-Resolution Dataset for Extreme Weather Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/3236f661fc6f00e8b16ff42db99acaaf-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "hrextremeahighresolutiondatasetforextremeweatherforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/3236f661fc6f00e8b16ff42db99acaaf-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-120",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 169,
"source_number": 120,
"title": "TCP-Diffusion: A Multi-modal Diffusion Model for Global Tropical Cyclone Precipitation Forecasting with Change Awareness",
"source_url": "https://proceedings.mlr.press/v267/huang25y.html",
"raw_line": "| P120 | [TCP-Diffusion: A Multi-modal Diffusion Model for Global Tropical Cyclone Precipitation Forecasting with Change Awareness](https://proceedings.mlr.press/v267/huang25y.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P120",
"[TCP-Diffusion: A Multi-modal Diffusion Model for Global Tropical Cyclone Precipitation Forecasting with Change Awareness](https://proceedings.mlr.press/v267/huang25y.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "tcpdiffusionamultimodaldiffusionmodelforglobaltropicalcycloneprecipitationforecastingwithchangeawareness",
"normalized_url": "proceedings.mlr.press/v267/huang25y.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-121",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 170,
"source_number": 121,
"title": "LoRA-EnVar: Parameter-Efficient Hybrid Ensemble Variational Assimilation for Weather Forecasting",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/413885e70482b95dcbeeddc1daf39177-Abstract-Conference.html",
"raw_line": "| P121 | [LoRA-EnVar: Parameter-Efficient Hybrid Ensemble Variational Assimilation for Weather Forecasting](https://papers.nips.cc/paper_files/paper/2025/hash/413885e70482b95dcbeeddc1daf39177-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P121",
"[LoRA-EnVar: Parameter-Efficient Hybrid Ensemble Variational Assimilation for Weather Forecasting](https://papers.nips.cc/paper_files/paper/2025/hash/413885e70482b95dcbeeddc1daf39177-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "loraenvarparameterefficienthybridensemblevariationalassimilationforweatherforecasting",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/413885e70482b95dcbeeddc1daf39177-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-122",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 171,
"source_number": 122,
"title": "FlowMixer: A Depth-Agnostic Neural Architecture for Interpretable Spatiotemporal Forecasting",
"source_url": "https://papers.neurips.cc/paper_files/paper/2025/hash/803485352e61e3ebf41221e4776c9fd4-Abstract-Conference.html",
"raw_line": "| P122 | [FlowMixer: A Depth-Agnostic Neural Architecture for Interpretable Spatiotemporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/803485352e61e3ebf41221e4776c9fd4-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P122",
"[FlowMixer: A Depth-Agnostic Neural Architecture for Interpretable Spatiotemporal Forecasting](https://papers.neurips.cc/paper_files/paper/2025/hash/803485352e61e3ebf41221e4776c9fd4-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "flowmixeradepthagnosticneuralarchitectureforinterpretablespatiotemporalforecasting",
"normalized_url": "papers.neurips.cc/paper_files/paper/2025/hash/803485352e61e3ebf41221e4776c9fd4-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-125",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 172,
"source_number": 125,
"title": "OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/fb0b57dfb8686c357f96e41b880ce7ed-Abstract-Conference.html",
"raw_line": "| P125 | [OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales](https://papers.nips.cc/paper_files/paper/2025/hash/fb0b57dfb8686c357f96e41b880ce7ed-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P125",
"[OmniCast: A Masked Latent Diffusion Model for Weather Forecasting Across Time Scales](https://papers.nips.cc/paper_files/paper/2025/hash/fb0b57dfb8686c357f96e41b880ce7ed-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "omnicastamaskedlatentdiffusionmodelforweatherforecastingacrosstimescales",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/fb0b57dfb8686c357f96e41b880ce7ed-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-126",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 173,
"source_number": 126,
"title": "RadarQA: Multi-modal Quality Analysis of Weather Radar Forecasts",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/beb667eee2fecb594dc3154950ef7aa4-Abstract-Conference.html",
"raw_line": "| P126 | [RadarQA: Multi-modal Quality Analysis of Weather Radar Forecasts](https://papers.nips.cc/paper_files/paper/2025/hash/beb667eee2fecb594dc3154950ef7aa4-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P126",
"[RadarQA: Multi-modal Quality Analysis of Weather Radar Forecasts](https://papers.nips.cc/paper_files/paper/2025/hash/beb667eee2fecb594dc3154950ef7aa4-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "radarqamultimodalqualityanalysisofweatherradarforecasts",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/beb667eee2fecb594dc3154950ef7aa4-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-127",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 174,
"source_number": 127,
"title": "Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/cb2266111eadcfa2c02187ace64e2183-Abstract-Conference.html",
"raw_line": "| P127 | [Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/cb2266111eadcfa2c02187ace64e2183-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P127",
"[Expand and Compress: Exploring Tuning Principles for Continual Spatio-Temporal Graph Forecasting](https://proceedings.iclr.cc/paper_files/paper/2025/hash/cb2266111eadcfa2c02187ace64e2183-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "expandandcompressexploringtuningprinciplesforcontinualspatiotemporalgraphforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/cb2266111eadcfa2c02187ace64e2183-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-128",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 175,
"source_number": 128,
"title": "STKOpt: Automated Spatio-Temporal Knowledge Optimization for Traffic Prediction",
"source_url": "https://doi.org/10.1145/3696410.3714598",
"raw_line": "| P128 | [STKOpt: Automated Spatio-Temporal Knowledge Optimization for Traffic Prediction](https://doi.org/10.1145/3696410.3714598) | 5 | 失败 | 评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（5 次，低于 10 次门槛）；[下载核对](论文/P128/下载核对.md) |",
"raw_cells": [
"P128",
"[STKOpt: Automated Spatio-Temporal Knowledge Optimization for Traffic Prediction](https://doi.org/10.1145/3696410.3714598)",
"5",
"失败",
"评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（5 次，低于 10 次门槛）；[下载核对](论文/P128/下载核对.md)"
],
"normalized_title": "stkoptautomatedspatiotemporalknowledgeoptimizationfortrafficprediction",
"normalized_url": "doi.org/10.1145/3696410.3714598",
"record_type": "legacy_unranked",
"citation_verbatim": "5",
"legacy_status_verbatim": "失败",
"legacy_unranked_reason_verbatim": "评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（5 次，低于 10 次门槛）；[下载核对](论文/P128/下载核对.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-130",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 176,
"source_number": 130,
"title": "NRFormer: Nationwide Nuclear Radiation Forecasting with Spatio-Temporal Transformer",
"source_url": "https://doi.org/10.1145/3711896.3737252",
"raw_line": "| P130 | [NRFormer: Nationwide Nuclear Radiation Forecasting with Spatio-Temporal Transformer](https://doi.org/10.1145/3711896.3737252) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P130",
"[NRFormer: Nationwide Nuclear Radiation Forecasting with Spatio-Temporal Transformer](https://doi.org/10.1145/3711896.3737252)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "nrformernationwidenuclearradiationforecastingwithspatiotemporaltransformer",
"normalized_url": "doi.org/10.1145/3711896.3737252",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-132",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 177,
"source_number": 132,
"title": "ASTNet: Asynchronous Spatio-Temporal Network for Large-Scale Chemical Sensor Forecasting",
"source_url": "https://doi.org/10.1145/3711896.3737194",
"raw_line": "| P132 | [ASTNet: Asynchronous Spatio-Temporal Network for Large-Scale Chemical Sensor Forecasting](https://doi.org/10.1145/3711896.3737194) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P132",
"[ASTNet: Asynchronous Spatio-Temporal Network for Large-Scale Chemical Sensor Forecasting](https://doi.org/10.1145/3711896.3737194)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "astnetasynchronousspatiotemporalnetworkforlargescalechemicalsensorforecasting",
"normalized_url": "doi.org/10.1145/3711896.3737194",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-139",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 178,
"source_number": 139,
"title": "Spatial-Temporal Heterogenous Graph Contrastive Learning for Microservice Workload Prediction",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/33271",
"raw_line": "| P139 | [Spatial-Temporal Heterogenous Graph Contrastive Learning for Microservice Workload Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33271) | 4 | 待处理 | 尚未评审；引用较低（4 次，低于 10 次门槛） |",
"raw_cells": [
"P139",
"[Spatial-Temporal Heterogenous Graph Contrastive Learning for Microservice Workload Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/33271)",
"4",
"待处理",
"尚未评审；引用较低（4 次，低于 10 次门槛）"
],
"normalized_title": "spatialtemporalheterogenousgraphcontrastivelearningformicroserviceworkloadprediction",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/33271",
"record_type": "legacy_unranked",
"citation_verbatim": "4",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（4 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-141",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 179,
"source_number": 141,
"title": "Does Vector Quantization Fail in Spatio-Temporal Forecasting? Exploring a Differentiable Sparse Soft-Vector Quantization Approach",
"source_url": "https://doi.org/10.1145/3711896.3736914",
"raw_line": "| P141 | [Does Vector Quantization Fail in Spatio-Temporal Forecasting? Exploring a Differentiable Sparse Soft-Vector Quantization Approach](https://doi.org/10.1145/3711896.3736914) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P141",
"[Does Vector Quantization Fail in Spatio-Temporal Forecasting? Exploring a Differentiable Sparse Soft-Vector Quantization Approach](https://doi.org/10.1145/3711896.3736914)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "doesvectorquantizationfailinspatiotemporalforecastingexploringadifferentiablesparsesoftvectorquantizationapproach",
"normalized_url": "doi.org/10.1145/3711896.3736914",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-143",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 180,
"source_number": 143,
"title": "FairTP: A Prolonged Fairness Framework for Traffic Prediction",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/34838",
"raw_line": "| P143 | [FairTP: A Prolonged Fairness Framework for Traffic Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/34838) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P143",
"[FairTP: A Prolonged Fairness Framework for Traffic Prediction](https://ojs.aaai.org/index.php/AAAI/article/view/34838)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "fairtpaprolongedfairnessframeworkfortrafficprediction",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/34838",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-146",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 181,
"source_number": 146,
"title": "Toward Interpretable Evaluation Measures for Time Series Segmentation",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/09425891e393e64b0535194a81ba15b7-Abstract-Conference.html",
"raw_line": "| P146 | [Toward Interpretable Evaluation Measures for Time Series Segmentation](https://papers.nips.cc/paper_files/paper/2025/hash/09425891e393e64b0535194a81ba15b7-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P146",
"[Toward Interpretable Evaluation Measures for Time Series Segmentation](https://papers.nips.cc/paper_files/paper/2025/hash/09425891e393e64b0535194a81ba15b7-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "towardinterpretableevaluationmeasuresfortimeseriessegmentation",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/09425891e393e64b0535194a81ba15b7-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-147",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 182,
"source_number": 147,
"title": "CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/a205fda871b0f6c1e18a7ad7325eb6cf-Abstract-Conference.html",
"raw_line": "| P147 | [CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series](https://proceedings.iclr.cc/paper_files/paper/2025/hash/a205fda871b0f6c1e18a7ad7325eb6cf-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P147",
"[CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series](https://proceedings.iclr.cc/paper_files/paper/2025/hash/a205fda871b0f6c1e18a7ad7325eb6cf-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "causalriversscalingupbenchmarkingofcausaldiscoveryforrealworldtimeseries",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/a205fda871b0f6c1e18a7ad7325eb6cf-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-148",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 183,
"source_number": 148,
"title": "Real-Time Calibration Model for Low-Cost Sensor in Fine-Grained Time Series",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/31974",
"raw_line": "| P148 | [Real-Time Calibration Model for Low-Cost Sensor in Fine-Grained Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/31974) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P148",
"[Real-Time Calibration Model for Low-Cost Sensor in Fine-Grained Time Series](https://ojs.aaai.org/index.php/AAAI/article/view/31974)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "realtimecalibrationmodelforlowcostsensorinfinegrainedtimeseries",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/31974",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-150",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 184,
"source_number": 150,
"title": "Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/b05616c9f202380cd3075ddabb6bf86d-Abstract-Conference.html",
"raw_line": "| P150 | [Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections](https://papers.nips.cc/paper_files/paper/2025/hash/b05616c9f202380cd3075ddabb6bf86d-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P150",
"[Learning Without Augmenting: Unsupervised Time Series Representation Learning via Frame Projections](https://papers.nips.cc/paper_files/paper/2025/hash/b05616c9f202380cd3075ddabb6bf86d-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "learningwithoutaugmentingunsupervisedtimeseriesrepresentationlearningviaframeprojections",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/b05616c9f202380cd3075ddabb6bf86d-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-151",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 185,
"source_number": 151,
"title": "AdaTS: Learning Adaptive Time Series Representations via Dynamic Soft Contrasts",
"source_url": "https://papers.nips.cc/paper_files/paper/2025/hash/6c66efb65c3031bd09b2201ca9a28a74-Abstract-Conference.html",
"raw_line": "| P151 | [AdaTS: Learning Adaptive Time Series Representations via Dynamic Soft Contrasts](https://papers.nips.cc/paper_files/paper/2025/hash/6c66efb65c3031bd09b2201ca9a28a74-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P151",
"[AdaTS: Learning Adaptive Time Series Representations via Dynamic Soft Contrasts](https://papers.nips.cc/paper_files/paper/2025/hash/6c66efb65c3031bd09b2201ca9a28a74-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "adatslearningadaptivetimeseriesrepresentationsviadynamicsoftcontrasts",
"normalized_url": "papers.nips.cc/paper_files/paper/2025/hash/6c66efb65c3031bd09b2201ca9a28a74-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-152",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 186,
"source_number": 152,
"title": "Time Series Representations with Hard-Coded Invariances",
"source_url": "https://proceedings.mlr.press/v267/germain25a.html",
"raw_line": "| P152 | [Time Series Representations with Hard-Coded Invariances](https://proceedings.mlr.press/v267/germain25a.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P152",
"[Time Series Representations with Hard-Coded Invariances](https://proceedings.mlr.press/v267/germain25a.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "timeseriesrepresentationswithhardcodedinvariances",
"normalized_url": "proceedings.mlr.press/v267/germain25a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-154",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 187,
"source_number": 154,
"title": "FAT: Frequency-Aware Pretraining for Enhanced Time-Series Representation Learning",
"source_url": "https://doi.org/10.1145/3711896.3736952",
"raw_line": "| P154 | [FAT: Frequency-Aware Pretraining for Enhanced Time-Series Representation Learning](https://doi.org/10.1145/3711896.3736952) | 6 | 失败 | 评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（6 次，低于 10 次门槛）；[下载核对](论文/P154/下载核对.md) |",
"raw_cells": [
"P154",
"[FAT: Frequency-Aware Pretraining for Enhanced Time-Series Representation Learning](https://doi.org/10.1145/3711896.3736952)",
"6",
"失败",
"评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（6 次，低于 10 次门槛）；[下载核对](论文/P154/下载核对.md)"
],
"normalized_title": "fatfrequencyawarepretrainingforenhancedtimeseriesrepresentationlearning",
"normalized_url": "doi.org/10.1145/3711896.3736952",
"record_type": "legacy_unranked",
"citation_verbatim": "6",
"legacy_status_verbatim": "失败",
"legacy_unranked_reason_verbatim": "评审未完成：该篇论文的下载不可得（执行器退出码 10）；引用较低（6 次，低于 10 次门槛）；[下载核对](论文/P154/下载核对.md)",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-155",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 188,
"source_number": 155,
"title": "Pre-training Time Series Models with Stock Data Customization",
"source_url": "https://doi.org/10.1145/3711896.3737005",
"raw_line": "| P155 | [Pre-training Time Series Models with Stock Data Customization](https://doi.org/10.1145/3711896.3737005) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P155",
"[Pre-training Time Series Models with Stock Data Customization](https://doi.org/10.1145/3711896.3737005)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "pretrainingtimeseriesmodelswithstockdatacustomization",
"normalized_url": "doi.org/10.1145/3711896.3737005",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-160",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 189,
"source_number": 160,
"title": "DyCAST: Learning Dynamic Causal Structure from Time Series",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2025/hash/3918dd28e51ce20c38c8a1c05758e4a0-Abstract-Conference.html",
"raw_line": "| P160 | [DyCAST: Learning Dynamic Causal Structure from Time Series](https://proceedings.iclr.cc/paper_files/paper/2025/hash/3918dd28e51ce20c38c8a1c05758e4a0-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P160",
"[DyCAST: Learning Dynamic Causal Structure from Time Series](https://proceedings.iclr.cc/paper_files/paper/2025/hash/3918dd28e51ce20c38c8a1c05758e4a0-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "dycastlearningdynamiccausalstructurefromtimeseries",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2025/hash/3918dd28e51ce20c38c8a1c05758e4a0-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-162",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 190,
"source_number": 162,
"title": "Does Synthetic Data Help? Empirical Evidence from Deep Learning Time Series Forecasters",
"source_url": "https://neurips.cc/virtual/2026/poster/154722",
"raw_line": "| P162 | [Does Synthetic Data Help? Empirical Evidence from Deep Learning Time Series Forecasters](https://neurips.cc/virtual/2026/poster/154722) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P162",
"[Does Synthetic Data Help? Empirical Evidence from Deep Learning Time Series Forecasters](https://neurips.cc/virtual/2026/poster/154722)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "doessyntheticdatahelpempiricalevidencefromdeeplearningtimeseriesforecasters",
"normalized_url": "neurips.cc/virtual/2026/poster/154722",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-163",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 191,
"source_number": 163,
"title": "GlucoFM-Bench: Benchmarking Time-Series Foundation Models for Blood Glucose Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/139533",
"raw_line": "| P163 | [GlucoFM-Bench: Benchmarking Time-Series Foundation Models for Blood Glucose Forecasting](https://neurips.cc/virtual/2026/poster/139533) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P163",
"[GlucoFM-Bench: Benchmarking Time-Series Foundation Models for Blood Glucose Forecasting](https://neurips.cc/virtual/2026/poster/139533)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "glucofmbenchbenchmarkingtimeseriesfoundationmodelsforbloodglucoseforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/139533",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-164",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 192,
"source_number": 164,
"title": "Differencing the Diffusion Trajectory toward Uncertain Components for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154330",
"raw_line": "| P164 | [Differencing the Diffusion Trajectory toward Uncertain Components for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154330) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P164",
"[Differencing the Diffusion Trajectory toward Uncertain Components for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154330)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "differencingthediffusiontrajectorytowarduncertaincomponentsfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154330",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-165",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 193,
"source_number": 165,
"title": "Nuwa: Evaluation-Grounded Agentic Construction of Time Series Forecasting Systems",
"source_url": "https://neurips.cc/virtual/2026/poster/153330",
"raw_line": "| P165 | [Nuwa: Evaluation-Grounded Agentic Construction of Time Series Forecasting Systems](https://neurips.cc/virtual/2026/poster/153330) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P165",
"[Nuwa: Evaluation-Grounded Agentic Construction of Time Series Forecasting Systems](https://neurips.cc/virtual/2026/poster/153330)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "nuwaevaluationgroundedagenticconstructionoftimeseriesforecastingsystems",
"normalized_url": "neurips.cc/virtual/2026/poster/153330",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-166",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 194,
"source_number": 166,
"title": "FISC: Time-Series Forecasting via First-Layer Statistical Calibration Constraints",
"source_url": "https://neurips.cc/virtual/2026/poster/149568",
"raw_line": "| P166 | [FISC: Time-Series Forecasting via First-Layer Statistical Calibration Constraints](https://neurips.cc/virtual/2026/poster/149568) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P166",
"[FISC: Time-Series Forecasting via First-Layer Statistical Calibration Constraints](https://neurips.cc/virtual/2026/poster/149568)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "fisctimeseriesforecastingviafirstlayerstatisticalcalibrationconstraints",
"normalized_url": "neurips.cc/virtual/2026/poster/149568",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-168",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 195,
"source_number": 168,
"title": "Robust Residual Correction via Selective Deployment for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154644",
"raw_line": "| P168 | [Robust Residual Correction via Selective Deployment for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154644) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P168",
"[Robust Residual Correction via Selective Deployment for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154644)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "robustresidualcorrectionviaselectivedeploymentfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154644",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-169",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 196,
"source_number": 169,
"title": "PAMod: Modeling Cyclical Shifts via Phase-Amplitude Modulation for Non-stationary Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/155815",
"raw_line": "| P169 | [PAMod: Modeling Cyclical Shifts via Phase-Amplitude Modulation for Non-stationary Time Series Forecasting](https://neurips.cc/virtual/2026/poster/155815) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P169",
"[PAMod: Modeling Cyclical Shifts via Phase-Amplitude Modulation for Non-stationary Time Series Forecasting](https://neurips.cc/virtual/2026/poster/155815)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "pamodmodelingcyclicalshiftsviaphaseamplitudemodulationfornonstationarytimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/155815",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-171",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 197,
"source_number": 171,
"title": "PhyTS: A Benchmark for Scientific Time Series",
"source_url": "https://neurips.cc/virtual/2026/poster/139305",
"raw_line": "| P171 | [PhyTS: A Benchmark for Scientific Time Series](https://neurips.cc/virtual/2026/poster/139305) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P171",
"[PhyTS: A Benchmark for Scientific Time Series](https://neurips.cc/virtual/2026/poster/139305)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "phytsabenchmarkforscientifictimeseries",
"normalized_url": "neurips.cc/virtual/2026/poster/139305",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-172",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 198,
"source_number": 172,
"title": "Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models",
"source_url": "https://neurips.cc/virtual/2026/poster/155883",
"raw_line": "| P172 | [Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models](https://neurips.cc/virtual/2026/poster/155883) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P172",
"[Revisiting the Generic Transformer: Deconstructing a Strong Baseline for Time Series Foundation Models](https://neurips.cc/virtual/2026/poster/155883)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "revisitingthegenerictransformerdeconstructingastrongbaselinefortimeseriesfoundationmodels",
"normalized_url": "neurips.cc/virtual/2026/poster/155883",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-173",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 199,
"source_number": 173,
"title": "Why Do Time Series Models Need Long Context Windows?",
"source_url": "https://neurips.cc/virtual/2026/poster/149819",
"raw_line": "| P173 | [Why Do Time Series Models Need Long Context Windows?](https://neurips.cc/virtual/2026/poster/149819) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P173",
"[Why Do Time Series Models Need Long Context Windows?](https://neurips.cc/virtual/2026/poster/149819)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "whydotimeseriesmodelsneedlongcontextwindows",
"normalized_url": "neurips.cc/virtual/2026/poster/149819",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-174",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 200,
"source_number": 174,
"title": "Filter Banks: from Low-Rank Representations to Deep Models for Efficient Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152048",
"raw_line": "| P174 | [Filter Banks: from Low-Rank Representations to Deep Models for Efficient Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152048) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P174",
"[Filter Banks: from Low-Rank Representations to Deep Models for Efficient Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152048)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "filterbanksfromlowrankrepresentationstodeepmodelsforefficienttimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152048",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-175",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 201,
"source_number": 175,
"title": "When to Trust Memory: Retrieval-Guided Probabilistic Spatiotemporal Forecasting under Distribution Shift",
"source_url": "https://neurips.cc/virtual/2026/poster/151106",
"raw_line": "| P175 | [When to Trust Memory: Retrieval-Guided Probabilistic Spatiotemporal Forecasting under Distribution Shift](https://neurips.cc/virtual/2026/poster/151106) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P175",
"[When to Trust Memory: Retrieval-Guided Probabilistic Spatiotemporal Forecasting under Distribution Shift](https://neurips.cc/virtual/2026/poster/151106)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "whentotrustmemoryretrievalguidedprobabilisticspatiotemporalforecastingunderdistributionshift",
"normalized_url": "neurips.cc/virtual/2026/poster/151106",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-176",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 202,
"source_number": 176,
"title": "OrangeTree: A Linear and Tree-based Time Series Forecasting Model Supporting Multiple Input and Output Lengths",
"source_url": "https://neurips.cc/virtual/2026/poster/151255",
"raw_line": "| P176 | [OrangeTree: A Linear and Tree-based Time Series Forecasting Model Supporting Multiple Input and Output Lengths](https://neurips.cc/virtual/2026/poster/151255) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P176",
"[OrangeTree: A Linear and Tree-based Time Series Forecasting Model Supporting Multiple Input and Output Lengths](https://neurips.cc/virtual/2026/poster/151255)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "orangetreealinearandtreebasedtimeseriesforecastingmodelsupportingmultipleinputandoutputlengths",
"normalized_url": "neurips.cc/virtual/2026/poster/151255",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-177",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 203,
"source_number": 177,
"title": "CVTA: Cross-Variable Temporal Attention for Multivariate Irregular Time Series Prediction",
"source_url": "https://neurips.cc/virtual/2026/poster/152025",
"raw_line": "| P177 | [CVTA: Cross-Variable Temporal Attention for Multivariate Irregular Time Series Prediction](https://neurips.cc/virtual/2026/poster/152025) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P177",
"[CVTA: Cross-Variable Temporal Attention for Multivariate Irregular Time Series Prediction](https://neurips.cc/virtual/2026/poster/152025)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "cvtacrossvariabletemporalattentionformultivariateirregulartimeseriesprediction",
"normalized_url": "neurips.cc/virtual/2026/poster/152025",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-178",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 204,
"source_number": 178,
"title": "TimeES: Probabilistic and Deterministic Time Series Forecasting via Evolutionary Spectra",
"source_url": "https://neurips.cc/virtual/2026/poster/154967",
"raw_line": "| P178 | [TimeES: Probabilistic and Deterministic Time Series Forecasting via Evolutionary Spectra](https://neurips.cc/virtual/2026/poster/154967) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P178",
"[TimeES: Probabilistic and Deterministic Time Series Forecasting via Evolutionary Spectra](https://neurips.cc/virtual/2026/poster/154967)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "timeesprobabilisticanddeterministictimeseriesforecastingviaevolutionaryspectra",
"normalized_url": "neurips.cc/virtual/2026/poster/154967",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-179",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 205,
"source_number": 179,
"title": "BucpTSF: Breaking the Uniform Computation Paradigm in Time-Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/148238",
"raw_line": "| P179 | [BucpTSF: Breaking the Uniform Computation Paradigm in Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/148238) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P179",
"[BucpTSF: Breaking the Uniform Computation Paradigm in Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/148238)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "bucptsfbreakingtheuniformcomputationparadigmintimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/148238",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-180",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 206,
"source_number": 180,
"title": "Hermes: A Multi-Scale Spatial-Temporal Hypergraph Network for Stock Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/149780",
"raw_line": "| P180 | [Hermes: A Multi-Scale Spatial-Temporal Hypergraph Network for Stock Time Series Forecasting](https://neurips.cc/virtual/2026/poster/149780) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P180",
"[Hermes: A Multi-Scale Spatial-Temporal Hypergraph Network for Stock Time Series Forecasting](https://neurips.cc/virtual/2026/poster/149780)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "hermesamultiscalespatialtemporalhypergraphnetworkforstocktimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/149780",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-181",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 207,
"source_number": 181,
"title": "OTformer: Non-Stationarity-Aware Adaptive Optimal Transport Attention for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/148743",
"raw_line": "| P181 | [OTformer: Non-Stationarity-Aware Adaptive Optimal Transport Attention for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148743) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P181",
"[OTformer: Non-Stationarity-Aware Adaptive Optimal Transport Attention for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148743)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "otformernonstationarityawareadaptiveoptimaltransportattentionfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/148743",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-182",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 208,
"source_number": 182,
"title": "HAMSTAR: Hamiltonian Structured Inter-Period Refinement for Long-Term Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/150931",
"raw_line": "| P182 | [HAMSTAR: Hamiltonian Structured Inter-Period Refinement for Long-Term Time Series Forecasting](https://neurips.cc/virtual/2026/poster/150931) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P182",
"[HAMSTAR: Hamiltonian Structured Inter-Period Refinement for Long-Term Time Series Forecasting](https://neurips.cc/virtual/2026/poster/150931)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "hamstarhamiltonianstructuredinterperiodrefinementforlongtermtimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/150931",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-183",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 209,
"source_number": 183,
"title": "Dirichlet-Guided Group Forecasting for Alleviating Over-smoothing in Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154424",
"raw_line": "| P183 | [Dirichlet-Guided Group Forecasting for Alleviating Over-smoothing in Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154424) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P183",
"[Dirichlet-Guided Group Forecasting for Alleviating Over-smoothing in Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154424)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "dirichletguidedgroupforecastingforalleviatingoversmoothingintimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154424",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-184",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 210,
"source_number": 184,
"title": "TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/149443",
"raw_line": "| P184 | [TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/149443) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P184",
"[TStruct: Learning Shared Temporal Structures for Long-Term Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/149443)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "tstructlearningsharedtemporalstructuresforlongtermtimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/149443",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-185",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 211,
"source_number": 185,
"title": "Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series",
"source_url": "https://neurips.cc/virtual/2026/poster/153183",
"raw_line": "| P185 | [Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series](https://neurips.cc/virtual/2026/poster/153183) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P185",
"[Predictive Surprise as Self-Grounding Concept Bottleneck for Interpretable Time Series](https://neurips.cc/virtual/2026/poster/153183)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "predictivesurpriseasselfgroundingconceptbottleneckforinterpretabletimeseries",
"normalized_url": "neurips.cc/virtual/2026/poster/153183",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-186",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 212,
"source_number": 186,
"title": "Beyond Isolation: Neighbor-Consistent Data Pruning for Multivariate Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154771",
"raw_line": "| P186 | [Beyond Isolation: Neighbor-Consistent Data Pruning for Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154771) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P186",
"[Beyond Isolation: Neighbor-Consistent Data Pruning for Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154771)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "beyondisolationneighborconsistentdatapruningformultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154771",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-187",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 213,
"source_number": 187,
"title": "Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154087",
"raw_line": "| P187 | [Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154087) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P187",
"[Paloma: Phase-Conditioned Residual Modulation for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154087)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "palomaphaseconditionedresidualmodulationfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154087",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-188",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 214,
"source_number": 188,
"title": "FLAME: Flow Enhanced Legendre Memory Models for General Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/151862",
"raw_line": "| P188 | [FLAME: Flow Enhanced Legendre Memory Models for General Time Series Forecasting](https://neurips.cc/virtual/2026/poster/151862) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P188",
"[FLAME: Flow Enhanced Legendre Memory Models for General Time Series Forecasting](https://neurips.cc/virtual/2026/poster/151862)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "flameflowenhancedlegendrememorymodelsforgeneraltimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/151862",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-189",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 215,
"source_number": 189,
"title": "QWaveNet: Quantum-Enhanced Wavelet Network for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152829",
"raw_line": "| P189 | [QWaveNet: Quantum-Enhanced Wavelet Network for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152829) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P189",
"[QWaveNet: Quantum-Enhanced Wavelet Network for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152829)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "qwavenetquantumenhancedwaveletnetworkfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152829",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-190",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 216,
"source_number": 190,
"title": "To See Far, Look Close: Evolutionary Forecasting for Long-term Time Series",
"source_url": "https://neurips.cc/virtual/2026/poster/152940",
"raw_line": "| P190 | [To See Far, Look Close: Evolutionary Forecasting for Long-term Time Series](https://neurips.cc/virtual/2026/poster/152940) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P190",
"[To See Far, Look Close: Evolutionary Forecasting for Long-term Time Series](https://neurips.cc/virtual/2026/poster/152940)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "toseefarlookcloseevolutionaryforecastingforlongtermtimeseries",
"normalized_url": "neurips.cc/virtual/2026/poster/152940",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-191",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 217,
"source_number": 191,
"title": "Multivariate Time Series Forecasting needs Cross Variable Loss",
"source_url": "https://neurips.cc/virtual/2026/poster/150730",
"raw_line": "| P191 | [Multivariate Time Series Forecasting needs Cross Variable Loss](https://neurips.cc/virtual/2026/poster/150730) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P191",
"[Multivariate Time Series Forecasting needs Cross Variable Loss](https://neurips.cc/virtual/2026/poster/150730)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "multivariatetimeseriesforecastingneedscrossvariableloss",
"normalized_url": "neurips.cc/virtual/2026/poster/150730",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-192",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 218,
"source_number": 192,
"title": "Beyond Linear Decoders: Dynamic Expert-Coupled Optimal Decoding for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153816",
"raw_line": "| P192 | [Beyond Linear Decoders: Dynamic Expert-Coupled Optimal Decoding for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153816) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P192",
"[Beyond Linear Decoders: Dynamic Expert-Coupled Optimal Decoding for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153816)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "beyondlineardecodersdynamicexpertcoupledoptimaldecodingfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153816",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-193",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 219,
"source_number": 193,
"title": "Spectral Energy Allocation Enables Source-Free Domain Adaptation in Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/oral/154915",
"raw_line": "| P193 | [Spectral Energy Allocation Enables Source-Free Domain Adaptation in Time Series Forecasting](https://neurips.cc/virtual/2026/oral/154915) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P193",
"[Spectral Energy Allocation Enables Source-Free Domain Adaptation in Time Series Forecasting](https://neurips.cc/virtual/2026/oral/154915)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "spectralenergyallocationenablessourcefreedomainadaptationintimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/oral/154915",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-194",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 220,
"source_number": 194,
"title": "Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153768",
"raw_line": "| P194 | [Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153768) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P194",
"[Rethinking Post-Training Recipes for Multimodal Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153768)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "rethinkingposttrainingrecipesformultimodaltimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153768",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-195",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 221,
"source_number": 195,
"title": "Rubato: Signature Attention for Irregular Multivariate Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153092",
"raw_line": "| P195 | [Rubato: Signature Attention for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153092) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P195",
"[Rubato: Signature Attention for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153092)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "rubatosignatureattentionforirregularmultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153092",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-196",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 222,
"source_number": 196,
"title": "When to Trust, How to Distill: Multi-Foundation Model Guidance for Lightweight, Robust Scientific Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3819018",
"raw_line": "| P196 | [When to Trust, How to Distill: Multi-Foundation Model Guidance for Lightweight, Robust Scientific Time Series Forecasting](https://doi.org/10.1145/3770855.3819018) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P196",
"[When to Trust, How to Distill: Multi-Foundation Model Guidance for Lightweight, Robust Scientific Time Series Forecasting](https://doi.org/10.1145/3770855.3819018)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "whentotrusthowtodistillmultifoundationmodelguidanceforlightweightrobustscientifictimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3770855.3819018",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-197",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 223,
"source_number": 197,
"title": "SEER: Transformer-based Robust Time Series Forecasting via Automated Patch Enhancement and Replacement",
"source_url": "https://proceedings.mlr.press/v306/qiu26n.html",
"raw_line": "| P197 | [SEER: Transformer-based Robust Time Series Forecasting via Automated Patch Enhancement and Replacement](https://proceedings.mlr.press/v306/qiu26n.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P197",
"[SEER: Transformer-based Robust Time Series Forecasting via Automated Patch Enhancement and Replacement](https://proceedings.mlr.press/v306/qiu26n.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "seertransformerbasedrobusttimeseriesforecastingviaautomatedpatchenhancementandreplacement",
"normalized_url": "proceedings.mlr.press/v306/qiu26n.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-199",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 224,
"source_number": 199,
"title": "Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/96d328a1f6d8396d8c8a62f2beee252a-Abstract-Conference.html",
"raw_line": "| P199 | [Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift](https://proceedings.iclr.cc/paper_files/paper/2026/hash/96d328a1f6d8396d8c8a62f2beee252a-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P199",
"[Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift](https://proceedings.iclr.cc/paper_files/paper/2026/hash/96d328a1f6d8396d8c8a62f2beee252a-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "tacklingtimeseriesforecastinggeneralizationviamitigatingconceptdrift",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/96d328a1f6d8396d8c8a62f2beee252a-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-200",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 225,
"source_number": 200,
"title": "Robust Inter-Series Dependency Modeling for Time Series Forecasting via Information-Theoretic Alignment",
"source_url": "https://proceedings.mlr.press/v306/yu26n.html",
"raw_line": "| P200 | [Robust Inter-Series Dependency Modeling for Time Series Forecasting via Information-Theoretic Alignment](https://proceedings.mlr.press/v306/yu26n.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P200",
"[Robust Inter-Series Dependency Modeling for Time Series Forecasting via Information-Theoretic Alignment](https://proceedings.mlr.press/v306/yu26n.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "robustinterseriesdependencymodelingfortimeseriesforecastingviainformationtheoreticalignment",
"normalized_url": "proceedings.mlr.press/v306/yu26n.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-201",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 226,
"source_number": 201,
"title": "Taming the Recent-Data Bias: Towards Robust Time Series Forecasting with Global Context",
"source_url": "https://proceedings.mlr.press/v306/xu26ab.html",
"raw_line": "| P201 | [Taming the Recent-Data Bias: Towards Robust Time Series Forecasting with Global Context](https://proceedings.mlr.press/v306/xu26ab.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P201",
"[Taming the Recent-Data Bias: Towards Robust Time Series Forecasting with Global Context](https://proceedings.mlr.press/v306/xu26ab.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "tamingtherecentdatabiastowardsrobusttimeseriesforecastingwithglobalcontext",
"normalized_url": "proceedings.mlr.press/v306/xu26ab.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-202",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 227,
"source_number": 202,
"title": "Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/f924dab7a06d0463fe238904678acaee-Abstract-Conference.html",
"raw_line": "| P202 | [Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f924dab7a06d0463fe238904678acaee-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P202",
"[Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f924dab7a06d0463fe238904678acaee-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "towardsrobustrealworldmultivariatetimeseriesforecastingaunifiedframeworkfordependencyasynchronyandmissingness",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/f924dab7a06d0463fe238904678acaee-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-203",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 228,
"source_number": 203,
"title": "Beyond Point Predictions: Manifold Expansion and Dual Alignment for Robust Time Series Distillation",
"source_url": "https://proceedings.mlr.press/v306/hong26e.html",
"raw_line": "| P203 | [Beyond Point Predictions: Manifold Expansion and Dual Alignment for Robust Time Series Distillation](https://proceedings.mlr.press/v306/hong26e.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P203",
"[Beyond Point Predictions: Manifold Expansion and Dual Alignment for Robust Time Series Distillation](https://proceedings.mlr.press/v306/hong26e.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "beyondpointpredictionsmanifoldexpansionanddualalignmentforrobusttimeseriesdistillation",
"normalized_url": "proceedings.mlr.press/v306/hong26e.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-204",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 229,
"source_number": 204,
"title": "Benchmarking Physics-Informed Time-Series Models for Operational Global Station Weather Forecasting",
"source_url": "https://proceedings.mlr.press/v306/han26v.html",
"raw_line": "| P204 | [Benchmarking Physics-Informed Time-Series Models for Operational Global Station Weather Forecasting](https://proceedings.mlr.press/v306/han26v.html) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P204",
"[Benchmarking Physics-Informed Time-Series Models for Operational Global Station Weather Forecasting](https://proceedings.mlr.press/v306/han26v.html)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "benchmarkingphysicsinformedtimeseriesmodelsforoperationalglobalstationweatherforecasting",
"normalized_url": "proceedings.mlr.press/v306/han26v.html",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-205",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 230,
"source_number": 205,
"title": "DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/zhong26l.html",
"raw_line": "| P205 | [DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting](https://proceedings.mlr.press/v306/zhong26l.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P205",
"[DropoutTS: Sample-Adaptive Dropout for Robust Time Series Forecasting](https://proceedings.mlr.press/v306/zhong26l.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "dropouttssampleadaptivedropoutforrobusttimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/zhong26l.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-206",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 231,
"source_number": 206,
"title": "ResCP: Reservoir Conformal Prediction for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/b0a978efeb20d33c15e31a5beee9f6df-Abstract-Conference.html",
"raw_line": "| P206 | [ResCP: Reservoir Conformal Prediction for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/b0a978efeb20d33c15e31a5beee9f6df-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P206",
"[ResCP: Reservoir Conformal Prediction for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/b0a978efeb20d33c15e31a5beee9f6df-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "rescpreservoirconformalpredictionfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/b0a978efeb20d33c15e31a5beee9f6df-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-207",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 232,
"source_number": 207,
"title": "Beyond Holistic Models: Systematic Component-level Benchmarking of Deep Multivariate Time-Series Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3817551",
"raw_line": "| P207 | [Beyond Holistic Models: Systematic Component-level Benchmarking of Deep Multivariate Time-Series Forecasting](https://doi.org/10.1145/3770855.3817551) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P207",
"[Beyond Holistic Models: Systematic Component-level Benchmarking of Deep Multivariate Time-Series Forecasting](https://doi.org/10.1145/3770855.3817551)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "beyondholisticmodelssystematiccomponentlevelbenchmarkingofdeepmultivariatetimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3770855.3817551",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-208",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 233,
"source_number": 208,
"title": "Online Irregular Multivariate Time Series Forecasting via Uncertainty-Driven Dual-Expert Calibration",
"source_url": "https://doi.org/10.1145/3770855.3817729",
"raw_line": "| P208 | [Online Irregular Multivariate Time Series Forecasting via Uncertainty-Driven Dual-Expert Calibration](https://doi.org/10.1145/3770855.3817729) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P208",
"[Online Irregular Multivariate Time Series Forecasting via Uncertainty-Driven Dual-Expert Calibration](https://doi.org/10.1145/3770855.3817729)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "onlineirregularmultivariatetimeseriesforecastingviauncertaintydrivendualexpertcalibration",
"normalized_url": "doi.org/10.1145/3770855.3817729",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-209",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 234,
"source_number": 209,
"title": "FinStressTS: A Parametric Synthetic Benchmark for Time-Series Forecasting in Finance",
"source_url": "https://doi.org/10.1145/3770855.3817578",
"raw_line": "| P209 | [FinStressTS: A Parametric Synthetic Benchmark for Time-Series Forecasting in Finance](https://doi.org/10.1145/3770855.3817578) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P209",
"[FinStressTS: A Parametric Synthetic Benchmark for Time-Series Forecasting in Finance](https://doi.org/10.1145/3770855.3817578)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "finstresstsaparametricsyntheticbenchmarkfortimeseriesforecastinginfinance",
"normalized_url": "doi.org/10.1145/3770855.3817578",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-210",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 235,
"source_number": 210,
"title": "APT: Affine Prototype-Timestamp for Time Series Forecasting Under Distribution Shift",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/38542",
"raw_line": "| P210 | [APT: Affine Prototype-Timestamp for Time Series Forecasting Under Distribution Shift](https://ojs.aaai.org/index.php/AAAI/article/view/38542) | 3 | 待处理 | 尚未评审；引用较低（3 次，低于 10 次门槛） |",
"raw_cells": [
"P210",
"[APT: Affine Prototype-Timestamp for Time Series Forecasting Under Distribution Shift](https://ojs.aaai.org/index.php/AAAI/article/view/38542)",
"3",
"待处理",
"尚未评审；引用较低（3 次，低于 10 次门槛）"
],
"normalized_title": "aptaffineprototypetimestampfortimeseriesforecastingunderdistributionshift",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/38542",
"record_type": "legacy_unranked",
"citation_verbatim": "3",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（3 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-211",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 236,
"source_number": 211,
"title": "Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions",
"source_url": "https://doi.org/10.1145/3770855.3817522",
"raw_line": "| P211 | [Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions](https://doi.org/10.1145/3770855.3817522) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P211",
"[Benchmarking Counterfactual Prediction in Epidemic Time Series with Time-Varying Interventions](https://doi.org/10.1145/3770855.3817522)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "benchmarkingcounterfactualpredictioninepidemictimeserieswithtimevaryinginterventions",
"normalized_url": "doi.org/10.1145/3770855.3817522",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-212",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 237,
"source_number": 212,
"title": "SEED: Spectral Entropy-Guided Evaluation of Spatial-Temporal Dependencies for Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/39930",
"raw_line": "| P212 | [SEED: Spectral Entropy-Guided Evaluation of Spatial-Temporal Dependencies for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39930) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P212",
"[SEED: Spectral Entropy-Guided Evaluation of Spatial-Temporal Dependencies for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39930)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "seedspectralentropyguidedevaluationofspatialtemporaldependenciesformultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/39930",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-213",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 238,
"source_number": 213,
"title": "ReCast: Reliability-aware Codebook-assisted Lightweight Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/39610",
"raw_line": "| P213 | [ReCast: Reliability-aware Codebook-assisted Lightweight Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39610) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P213",
"[ReCast: Reliability-aware Codebook-assisted Lightweight Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39610)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "recastreliabilityawarecodebookassistedlightweighttimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/39610",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-215",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 239,
"source_number": 215,
"title": "TimeMRA: LLM-Empowered Time Series Forecasting via Multi-Scale Retrieval-Augmented Representations",
"source_url": "https://proceedings.mlr.press/v306/shang26b.html",
"raw_line": "| P215 | [TimeMRA: LLM-Empowered Time Series Forecasting via Multi-Scale Retrieval-Augmented Representations](https://proceedings.mlr.press/v306/shang26b.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P215",
"[TimeMRA: LLM-Empowered Time Series Forecasting via Multi-Scale Retrieval-Augmented Representations](https://proceedings.mlr.press/v306/shang26b.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "timemrallmempoweredtimeseriesforecastingviamultiscaleretrievalaugmentedrepresentations",
"normalized_url": "proceedings.mlr.press/v306/shang26b.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-216",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 240,
"source_number": 216,
"title": "Channel Adapter for Time Series Foundation Models in Zero-Shot Multivariate Forecasting",
"source_url": "https://proceedings.mlr.press/v306/li26bu.html",
"raw_line": "| P216 | [Channel Adapter for Time Series Foundation Models in Zero-Shot Multivariate Forecasting](https://proceedings.mlr.press/v306/li26bu.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P216",
"[Channel Adapter for Time Series Foundation Models in Zero-Shot Multivariate Forecasting](https://proceedings.mlr.press/v306/li26bu.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "channeladapterfortimeseriesfoundationmodelsinzeroshotmultivariateforecasting",
"normalized_url": "proceedings.mlr.press/v306/li26bu.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-217",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 241,
"source_number": 217,
"title": "Invariant Representation Learning for Source-Free Time Series Forecasting with LLM-Centric Proxy Denoising",
"source_url": "https://proceedings.mlr.press/v306/yan26l.html",
"raw_line": "| P217 | [Invariant Representation Learning for Source-Free Time Series Forecasting with LLM-Centric Proxy Denoising](https://proceedings.mlr.press/v306/yan26l.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P217",
"[Invariant Representation Learning for Source-Free Time Series Forecasting with LLM-Centric Proxy Denoising](https://proceedings.mlr.press/v306/yan26l.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "invariantrepresentationlearningforsourcefreetimeseriesforecastingwithllmcentricproxydenoising",
"normalized_url": "proceedings.mlr.press/v306/yan26l.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-218",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 242,
"source_number": 218,
"title": "See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling",
"source_url": "https://proceedings.mlr.press/v306/xu26ac.html",
"raw_line": "| P218 | [See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling](https://proceedings.mlr.press/v306/xu26ac.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P218",
"[See More, Forecast Better and Faster: Enhancing Time Series Foundation Models via Inference-Time Plug-and-Play Downsampling](https://proceedings.mlr.press/v306/xu26ac.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "seemoreforecastbetterandfasterenhancingtimeseriesfoundationmodelsviainferencetimeplugandplaydownsampling",
"normalized_url": "proceedings.mlr.press/v306/xu26ac.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-219",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 243,
"source_number": 219,
"title": "LagLLM: LLM-empowered lead–lag dependency learning for spatial-temporal time series forecasting",
"source_url": "https://proceedings.mlr.press/v306/wu26ci.html",
"raw_line": "| P219 | [LagLLM: LLM-empowered lead–lag dependency learning for spatial-temporal time series forecasting](https://proceedings.mlr.press/v306/wu26ci.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P219",
"[LagLLM: LLM-empowered lead–lag dependency learning for spatial-temporal time series forecasting](https://proceedings.mlr.press/v306/wu26ci.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "lagllmllmempoweredleadlagdependencylearningforspatialtemporaltimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wu26ci.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-220",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 244,
"source_number": 220,
"title": "Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/410dacbcfda0388f69b8983ed771c44e-Abstract-Conference.html",
"raw_line": "| P220 | [Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/410dacbcfda0388f69b8983ed771c44e-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P220",
"[Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/410dacbcfda0388f69b8983ed771c44e-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "learningrecursivemultiscalerepresentationsforirregularmultivariatetimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/410dacbcfda0388f69b8983ed771c44e-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-221",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 245,
"source_number": 221,
"title": "FireSentry: A Multi-Modal Spatio-temporal Benchmark Dataset for Fine-Grained Wildfire Spread Forecasting",
"source_url": "https://doi.org/10.1145/3770854.3785696",
"raw_line": "| P221 | [FireSentry: A Multi-Modal Spatio-temporal Benchmark Dataset for Fine-Grained Wildfire Spread Forecasting](https://doi.org/10.1145/3770854.3785696) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P221",
"[FireSentry: A Multi-Modal Spatio-temporal Benchmark Dataset for Fine-Grained Wildfire Spread Forecasting](https://doi.org/10.1145/3770854.3785696)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "firesentryamultimodalspatiotemporalbenchmarkdatasetforfinegrainedwildfirespreadforecasting",
"normalized_url": "doi.org/10.1145/3770854.3785696",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-223",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 246,
"source_number": 223,
"title": "LoFT-LLM: Low-Frequency Time-series Forecasting with Large Language Models",
"source_url": "https://doi.org/10.1145/3770854.3780245",
"raw_line": "| P223 | [LoFT-LLM: Low-Frequency Time-series Forecasting with Large Language Models](https://doi.org/10.1145/3770854.3780245) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P223",
"[LoFT-LLM: Low-Frequency Time-series Forecasting with Large Language Models](https://doi.org/10.1145/3770854.3780245)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "loftllmlowfrequencytimeseriesforecastingwithlargelanguagemodels",
"normalized_url": "doi.org/10.1145/3770854.3780245",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-224",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 247,
"source_number": 224,
"title": "Efficient Test-Time Scaling for LLM-based Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3818201",
"raw_line": "| P224 | [Efficient Test-Time Scaling for LLM-based Time Series Forecasting](https://doi.org/10.1145/3770855.3818201) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P224",
"[Efficient Test-Time Scaling for LLM-based Time Series Forecasting](https://doi.org/10.1145/3770855.3818201)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "efficienttesttimescalingforllmbasedtimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3770855.3818201",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-225",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 248,
"source_number": 225,
"title": "Energy-Efficient Training-Free Zero-Inflation Correction for Rainfall Forecasting with Time-Series Foundation Models",
"source_url": "https://doi.org/10.1145/3774904.3792998",
"raw_line": "| P225 | [Energy-Efficient Training-Free Zero-Inflation Correction for Rainfall Forecasting with Time-Series Foundation Models](https://doi.org/10.1145/3774904.3792998) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P225",
"[Energy-Efficient Training-Free Zero-Inflation Correction for Rainfall Forecasting with Time-Series Foundation Models](https://doi.org/10.1145/3774904.3792998)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "energyefficienttrainingfreezeroinflationcorrectionforrainfallforecastingwithtimeseriesfoundationmodels",
"normalized_url": "doi.org/10.1145/3774904.3792998",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-226",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 249,
"source_number": 226,
"title": "Unifying Channel Independence and Mixing: Multi-Scale Patch Recursion for Global–Local Representation Synergy in Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/40072",
"raw_line": "| P226 | [Unifying Channel Independence and Mixing: Multi-Scale Patch Recursion for Global–Local Representation Synergy in Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/40072) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P226",
"[Unifying Channel Independence and Mixing: Multi-Scale Patch Recursion for Global–Local Representation Synergy in Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/40072)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "unifyingchannelindependenceandmixingmultiscalepatchrecursionforgloballocalrepresentationsynergyinmultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/40072",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-227",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 250,
"source_number": 227,
"title": "M3Time: LLM-Enhanced Multi-Modal, Multi-Scale, and Multi-Frequency Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/39383",
"raw_line": "| P227 | [M3Time: LLM-Enhanced Multi-Modal, Multi-Scale, and Multi-Frequency Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39383) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P227",
"[M3Time: LLM-Enhanced Multi-Modal, Multi-Scale, and Multi-Frequency Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39383)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "m3timellmenhancedmultimodalmultiscaleandmultifrequencymultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/39383",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-228",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 251,
"source_number": 228,
"title": "Unified Spatio-Temporal Tokens are Bases for Generalizable Traffic Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3817746",
"raw_line": "| P228 | [Unified Spatio-Temporal Tokens are Bases for Generalizable Traffic Forecasting](https://doi.org/10.1145/3770855.3817746) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P228",
"[Unified Spatio-Temporal Tokens are Bases for Generalizable Traffic Forecasting](https://doi.org/10.1145/3770855.3817746)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "unifiedspatiotemporaltokensarebasesforgeneralizabletrafficforecasting",
"normalized_url": "doi.org/10.1145/3770855.3817746",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-229",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 252,
"source_number": 229,
"title": "One Step Closer to Ground Truth: A Multi-Scale Residual-Aware Representation Learning Pipeline for Predicting Time Series Data",
"source_url": "https://doi.org/10.1145/3770855.3817960",
"raw_line": "| P229 | [One Step Closer to Ground Truth: A Multi-Scale Residual-Aware Representation Learning Pipeline for Predicting Time Series Data](https://doi.org/10.1145/3770855.3817960) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P229",
"[One Step Closer to Ground Truth: A Multi-Scale Residual-Aware Representation Learning Pipeline for Predicting Time Series Data](https://doi.org/10.1145/3770855.3817960)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "onestepclosertogroundtruthamultiscaleresidualawarerepresentationlearningpipelineforpredictingtimeseriesdata",
"normalized_url": "doi.org/10.1145/3770855.3817960",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-230",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 253,
"source_number": 230,
"title": "Scalable Uncertainty Quantification for Extreme Weather Forecasting via Empirical Neural Tangent Kernels",
"source_url": "https://doi.org/10.1145/3770855.3818106",
"raw_line": "| P230 | [Scalable Uncertainty Quantification for Extreme Weather Forecasting via Empirical Neural Tangent Kernels](https://doi.org/10.1145/3770855.3818106) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P230",
"[Scalable Uncertainty Quantification for Extreme Weather Forecasting via Empirical Neural Tangent Kernels](https://doi.org/10.1145/3770855.3818106)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "scalableuncertaintyquantificationforextremeweatherforecastingviaempiricalneuraltangentkernels",
"normalized_url": "doi.org/10.1145/3770855.3818106",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-231",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 254,
"source_number": 231,
"title": "GraFT: Infusing Pre-trained Transformers with Relational Structure for Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/40029",
"raw_line": "| P231 | [GraFT: Infusing Pre-trained Transformers with Relational Structure for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/40029) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P231",
"[GraFT: Infusing Pre-trained Transformers with Relational Structure for Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/40029)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "graftinfusingpretrainedtransformerswithrelationalstructurefortimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/40029",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-232",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 255,
"source_number": 232,
"title": "Lightweight Wrappers for Adapting Time Series Foundation Models to Regional Drought Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3818835",
"raw_line": "| P232 | [Lightweight Wrappers for Adapting Time Series Foundation Models to Regional Drought Forecasting](https://doi.org/10.1145/3770855.3818835) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P232",
"[Lightweight Wrappers for Adapting Time Series Foundation Models to Regional Drought Forecasting](https://doi.org/10.1145/3770855.3818835)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "lightweightwrappersforadaptingtimeseriesfoundationmodelstoregionaldroughtforecasting",
"normalized_url": "doi.org/10.1145/3770855.3818835",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-233",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 256,
"source_number": 233,
"title": "RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting",
"source_url": "https://doi.org/10.1145/3770855.3817683",
"raw_line": "| P233 | [RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting](https://doi.org/10.1145/3770855.3817683) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P233",
"[RePatch: Learning Entropy-Guided Patch Structures with Quantized Representations for Time Series Forecasting](https://doi.org/10.1145/3770855.3817683)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "repatchlearningentropyguidedpatchstructureswithquantizedrepresentationsfortimeseriesforecasting",
"normalized_url": "doi.org/10.1145/3770855.3817683",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-234",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 257,
"source_number": 234,
"title": "PatchCLE: Breaking the Linear Representation Bottleneck in Time Series Forecasting via Soft Contrastive Learning",
"source_url": "https://doi.org/10.1145/3770855.3818023",
"raw_line": "| P234 | [PatchCLE: Breaking the Linear Representation Bottleneck in Time Series Forecasting via Soft Contrastive Learning](https://doi.org/10.1145/3770855.3818023) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P234",
"[PatchCLE: Breaking the Linear Representation Bottleneck in Time Series Forecasting via Soft Contrastive Learning](https://doi.org/10.1145/3770855.3818023)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "patchclebreakingthelinearrepresentationbottleneckintimeseriesforecastingviasoftcontrastivelearning",
"normalized_url": "doi.org/10.1145/3770855.3818023",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-235",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 258,
"source_number": 235,
"title": "TimeCAP: A Channel-Aware Pre-Training Framework for Multivariate Time Series Forecasting",
"source_url": "https://ojs.aaai.org/index.php/AAAI/article/view/39700",
"raw_line": "| P235 | [TimeCAP: A Channel-Aware Pre-Training Framework for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39700) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P235",
"[TimeCAP: A Channel-Aware Pre-Training Framework for Multivariate Time Series Forecasting](https://ojs.aaai.org/index.php/AAAI/article/view/39700)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "timecapachannelawarepretrainingframeworkformultivariatetimeseriesforecasting",
"normalized_url": "ojs.aaai.org/index.php/aaai/article/view/39700",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-236",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 259,
"source_number": 236,
"title": "TimeGuard: Channel-wise Pool Training for Backdoor Defense in Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/nguyen26g.html",
"raw_line": "| P236 | [TimeGuard: Channel-wise Pool Training for Backdoor Defense in Time Series Forecasting](https://proceedings.mlr.press/v306/nguyen26g.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P236",
"[TimeGuard: Channel-wise Pool Training for Backdoor Defense in Time Series Forecasting](https://proceedings.mlr.press/v306/nguyen26g.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "timeguardchannelwisepooltrainingforbackdoordefenseintimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/nguyen26g.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-237",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 260,
"source_number": 237,
"title": "Bridging Time and Frequency: A Joint Modeling Framework for Irregular Multivariate Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/qiu26t.html",
"raw_line": "| P237 | [Bridging Time and Frequency: A Joint Modeling Framework for Irregular Multivariate Time Series Forecasting](https://proceedings.mlr.press/v306/qiu26t.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P237",
"[Bridging Time and Frequency: A Joint Modeling Framework for Irregular Multivariate Time Series Forecasting](https://proceedings.mlr.press/v306/qiu26t.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "bridgingtimeandfrequencyajointmodelingframeworkforirregularmultivariatetimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/qiu26t.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-238",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 261,
"source_number": 238,
"title": "L-Drive: Beyond a Single Mapping—Latent Context Drives Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/zhang26p.html",
"raw_line": "| P238 | [L-Drive: Beyond a Single Mapping—Latent Context Drives Time Series Forecasting](https://proceedings.mlr.press/v306/zhang26p.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P238",
"[L-Drive: Beyond a Single Mapping—Latent Context Drives Time Series Forecasting](https://proceedings.mlr.press/v306/zhang26p.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "ldrivebeyondasinglemappinglatentcontextdrivestimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/zhang26p.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-239",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 262,
"source_number": 239,
"title": "Reviving Error Correction in Modern Deep Time-Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/nguyen26u.html",
"raw_line": "| P239 | [Reviving Error Correction in Modern Deep Time-Series Forecasting](https://proceedings.mlr.press/v306/nguyen26u.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P239",
"[Reviving Error Correction in Modern Deep Time-Series Forecasting](https://proceedings.mlr.press/v306/nguyen26u.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "revivingerrorcorrectioninmoderndeeptimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/nguyen26u.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-240",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 263,
"source_number": 240,
"title": "KUMA: A Novel Framework with Koopman Separation and Efficient Multilevel Extraction in Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/xiong26j.html",
"raw_line": "| P240 | [KUMA: A Novel Framework with Koopman Separation and Efficient Multilevel Extraction in Time Series Forecasting](https://proceedings.mlr.press/v306/xiong26j.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P240",
"[KUMA: A Novel Framework with Koopman Separation and Efficient Multilevel Extraction in Time Series Forecasting](https://proceedings.mlr.press/v306/xiong26j.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "kumaanovelframeworkwithkoopmanseparationandefficientmultilevelextractionintimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/xiong26j.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-241",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 264,
"source_number": 241,
"title": "KITE: Knowledge-Guided Probabilistic Modeling for Time Series Forecasting with Exogenous Variables",
"source_url": "https://proceedings.mlr.press/v306/cheng26t.html",
"raw_line": "| P241 | [KITE: Knowledge-Guided Probabilistic Modeling for Time Series Forecasting with Exogenous Variables](https://proceedings.mlr.press/v306/cheng26t.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P241",
"[KITE: Knowledge-Guided Probabilistic Modeling for Time Series Forecasting with Exogenous Variables](https://proceedings.mlr.press/v306/cheng26t.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "kiteknowledgeguidedprobabilisticmodelingfortimeseriesforecastingwithexogenousvariables",
"normalized_url": "proceedings.mlr.press/v306/cheng26t.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-242",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 265,
"source_number": 242,
"title": "FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/090d636f951ac0bb18fd5610c95c6a76-Abstract-Conference.html",
"raw_line": "| P242 | [FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/090d636f951ac0bb18fd5610c95c6a76-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P242",
"[FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/090d636f951ac0bb18fd5610c95c6a76-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "factfinegrainedacrossvariableconvolutionformultivariatetimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/090d636f951ac0bb18fd5610c95c6a76-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-243",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 266,
"source_number": 243,
"title": "CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/ebba182cb97864368fdb6ae00773a5e4-Abstract-Conference.html",
"raw_line": "| P243 | [CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/ebba182cb97864368fdb6ae00773a5e4-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P243",
"[CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/ebba182cb97864368fdb6ae00773a5e4-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "cpirichannelpermutationinvariantrelationalinteractionformultivariatetimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/ebba182cb97864368fdb6ae00773a5e4-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-244",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 267,
"source_number": 244,
"title": "ConFlux: Multivariate Time Series in Flux, One Unified Forecast in Confluence",
"source_url": "https://proceedings.mlr.press/v306/wang26fy.html",
"raw_line": "| P244 | [ConFlux: Multivariate Time Series in Flux, One Unified Forecast in Confluence](https://proceedings.mlr.press/v306/wang26fy.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P244",
"[ConFlux: Multivariate Time Series in Flux, One Unified Forecast in Confluence](https://proceedings.mlr.press/v306/wang26fy.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "confluxmultivariatetimeseriesinfluxoneunifiedforecastinconfluence",
"normalized_url": "proceedings.mlr.press/v306/wang26fy.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-245",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 268,
"source_number": 245,
"title": "Revealing Scaling Paradox in Large-scale Time Series Models: Implications for More Efficient and Accurate Forecasting",
"source_url": "https://proceedings.mlr.press/v306/qiu26q.html",
"raw_line": "| P245 | [Revealing Scaling Paradox in Large-scale Time Series Models: Implications for More Efficient and Accurate Forecasting](https://proceedings.mlr.press/v306/qiu26q.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P245",
"[Revealing Scaling Paradox in Large-scale Time Series Models: Implications for More Efficient and Accurate Forecasting](https://proceedings.mlr.press/v306/qiu26q.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "revealingscalingparadoxinlargescaletimeseriesmodelsimplicationsformoreefficientandaccurateforecasting",
"normalized_url": "proceedings.mlr.press/v306/qiu26q.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-246",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 269,
"source_number": 246,
"title": "Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/zhu26ad.html",
"raw_line": "| P246 | [Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/zhu26ad.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P246",
"[Dynamic TMoE: A Drift-Aware Dynamic Mixture of Experts Framework for Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/zhu26ad.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "dynamictmoeadriftawaredynamicmixtureofexpertsframeworkfornonstationarytimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/zhu26ad.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-247",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 270,
"source_number": 247,
"title": "Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/494ad6d9b36d9a1528dbf3baeb4e8a72-Abstract-Conference.html",
"raw_line": "| P247 | [Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition](https://proceedings.iclr.cc/paper_files/paper/2026/hash/494ad6d9b36d9a1528dbf3baeb4e8a72-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P247",
"[Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition](https://proceedings.iclr.cc/paper_files/paper/2026/hash/494ad6d9b36d9a1528dbf3baeb4e8a72-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "routingchannelpatchdependenciesintimeseriesforecastingwithgraphspectraldecomposition",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/494ad6d9b36d9a1528dbf3baeb4e8a72-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-248",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 271,
"source_number": 248,
"title": "Crisp: A Spectral-Based Interaction Strategy for Multivariate Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/wang26ec.html",
"raw_line": "| P248 | [Crisp: A Spectral-Based Interaction Strategy for Multivariate Time Series Forecasting](https://proceedings.mlr.press/v306/wang26ec.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P248",
"[Crisp: A Spectral-Based Interaction Strategy for Multivariate Time Series Forecasting](https://proceedings.mlr.press/v306/wang26ec.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "crispaspectralbasedinteractionstrategyformultivariatetimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wang26ec.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-249",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 272,
"source_number": 249,
"title": "PULSE: Generative Phase Evolution for Non-Stationary Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/liu26dn.html",
"raw_line": "| P249 | [PULSE: Generative Phase Evolution for Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/liu26dn.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P249",
"[PULSE: Generative Phase Evolution for Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/liu26dn.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "pulsegenerativephaseevolutionfornonstationarytimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/liu26dn.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-250",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 273,
"source_number": 250,
"title": "From Observations to States: Latent Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/yang26y.html",
"raw_line": "| P250 | [From Observations to States: Latent Time Series Forecasting](https://proceedings.mlr.press/v306/yang26y.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P250",
"[From Observations to States: Latent Time Series Forecasting](https://proceedings.mlr.press/v306/yang26y.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "fromobservationstostateslatenttimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/yang26y.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-251",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 274,
"source_number": 251,
"title": "MoRGen: Mixture-of-Resolutions Generative Forecasting for Irregularly Sampled Medical Time-Series Data",
"source_url": "https://proceedings.mlr.press/v306/oufattole26a.html",
"raw_line": "| P251 | [MoRGen: Mixture-of-Resolutions Generative Forecasting for Irregularly Sampled Medical Time-Series Data](https://proceedings.mlr.press/v306/oufattole26a.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P251",
"[MoRGen: Mixture-of-Resolutions Generative Forecasting for Irregularly Sampled Medical Time-Series Data](https://proceedings.mlr.press/v306/oufattole26a.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "morgenmixtureofresolutionsgenerativeforecastingforirregularlysampledmedicaltimeseriesdata",
"normalized_url": "proceedings.mlr.press/v306/oufattole26a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-252",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 275,
"source_number": 252,
"title": "COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/2a8ce71baac4c89bf9ff479d8240c7d9-Abstract-Conference.html",
"raw_line": "| P252 | [COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/2a8ce71baac4c89bf9ff479d8240c7d9-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P252",
"[COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/2a8ce71baac4c89bf9ff479d8240c7d9-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "cosacontextawareoutputspaceadapterfortesttimeadaptationintimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/2a8ce71baac4c89bf9ff479d8240c7d9-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-253",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 276,
"source_number": 253,
"title": "ReNF: Rethinking the Design of Neural Long-Term Time Series Forecasters",
"source_url": "https://proceedings.mlr.press/v306/lu26w.html",
"raw_line": "| P253 | [ReNF: Rethinking the Design of Neural Long-Term Time Series Forecasters](https://proceedings.mlr.press/v306/lu26w.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P253",
"[ReNF: Rethinking the Design of Neural Long-Term Time Series Forecasters](https://proceedings.mlr.press/v306/lu26w.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "renfrethinkingthedesignofneurallongtermtimeseriesforecasters",
"normalized_url": "proceedings.mlr.press/v306/lu26w.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-254",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 277,
"source_number": 254,
"title": "StretchTime: Adaptive Time Series Forecasting via Symplectic Attention",
"source_url": "https://proceedings.mlr.press/v306/kim26bk.html",
"raw_line": "| P254 | [StretchTime: Adaptive Time Series Forecasting via Symplectic Attention](https://proceedings.mlr.press/v306/kim26bk.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P254",
"[StretchTime: Adaptive Time Series Forecasting via Symplectic Attention](https://proceedings.mlr.press/v306/kim26bk.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "stretchtimeadaptivetimeseriesforecastingviasymplecticattention",
"normalized_url": "proceedings.mlr.press/v306/kim26bk.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-255",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 278,
"source_number": 255,
"title": "Byte Pair Encoding for Efficient Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/gotz26a.html",
"raw_line": "| P255 | [Byte Pair Encoding for Efficient Time Series Forecasting](https://proceedings.mlr.press/v306/gotz26a.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P255",
"[Byte Pair Encoding for Efficient Time Series Forecasting](https://proceedings.mlr.press/v306/gotz26a.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "bytepairencodingforefficienttimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/gotz26a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-256",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 279,
"source_number": 256,
"title": "PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/dbb5180957513805ebeea787b8c66ac9-Abstract-Conference.html",
"raw_line": "| P256 | [PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/dbb5180957513805ebeea787b8c66ac9-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P256",
"[PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/dbb5180957513805ebeea787b8c66ac9-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "phatmodelingperiodheterogeneityformultivariatetimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/dbb5180957513805ebeea787b8c66ac9-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-257",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 280,
"source_number": 257,
"title": "FIPN: Forward Self-Organizing Interpretable Polynomial Networks for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/wang26hv.html",
"raw_line": "| P257 | [FIPN: Forward Self-Organizing Interpretable Polynomial Networks for Time Series Forecasting](https://proceedings.mlr.press/v306/wang26hv.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P257",
"[FIPN: Forward Self-Organizing Interpretable Polynomial Networks for Time Series Forecasting](https://proceedings.mlr.press/v306/wang26hv.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "fipnforwardselforganizinginterpretablepolynomialnetworksfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wang26hv.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-258",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 281,
"source_number": 258,
"title": "PESD-TSF: A Period-Aware and Explicit Structured Decomposition Framework for Long-Term Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/wang26cs.html",
"raw_line": "| P258 | [PESD-TSF: A Period-Aware and Explicit Structured Decomposition Framework for Long-Term Time Series Forecasting](https://proceedings.mlr.press/v306/wang26cs.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P258",
"[PESD-TSF: A Period-Aware and Explicit Structured Decomposition Framework for Long-Term Time Series Forecasting](https://proceedings.mlr.press/v306/wang26cs.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "pesdtsfaperiodawareandexplicitstructureddecompositionframeworkforlongtermtimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wang26cs.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-259",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 282,
"source_number": 259,
"title": "SDHilb: Schrödinger Dynamics-Guided Neural Network with Adaptive Multi-Scale Hilbert Transform for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152741",
"raw_line": "| P259 | [SDHilb: Schrödinger Dynamics-Guided Neural Network with Adaptive Multi-Scale Hilbert Transform for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152741) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P259",
"[SDHilb: Schrödinger Dynamics-Guided Neural Network with Adaptive Multi-Scale Hilbert Transform for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152741)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "sdhilbschrödingerdynamicsguidedneuralnetworkwithadaptivemultiscalehilberttransformfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152741",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-260",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 283,
"source_number": 260,
"title": "DSENet: A Novel Dual-Stream Enhancement Network for Multi-Scale Non-Stationary Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/wang26lu.html",
"raw_line": "| P260 | [DSENet: A Novel Dual-Stream Enhancement Network for Multi-Scale Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/wang26lu.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P260",
"[DSENet: A Novel Dual-Stream Enhancement Network for Multi-Scale Non-Stationary Time Series Forecasting](https://proceedings.mlr.press/v306/wang26lu.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "dsenetanoveldualstreamenhancementnetworkformultiscalenonstationarytimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wang26lu.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-261",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 284,
"source_number": 261,
"title": "Ellipsoidal Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/wang26ax.html",
"raw_line": "| P261 | [Ellipsoidal Time Series Forecasting](https://proceedings.mlr.press/v306/wang26ax.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P261",
"[Ellipsoidal Time Series Forecasting](https://proceedings.mlr.press/v306/wang26ax.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "ellipsoidaltimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/wang26ax.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-262",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 285,
"source_number": 262,
"title": "Stability-Constrained Regime-Aware Forecasting for Heterogeneous Panel Time Series",
"source_url": "https://neurips.cc/virtual/2026/poster/153846",
"raw_line": "| P262 | [Stability-Constrained Regime-Aware Forecasting for Heterogeneous Panel Time Series](https://neurips.cc/virtual/2026/poster/153846) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P262",
"[Stability-Constrained Regime-Aware Forecasting for Heterogeneous Panel Time Series](https://neurips.cc/virtual/2026/poster/153846)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "stabilityconstrainedregimeawareforecastingforheterogeneouspaneltimeseries",
"normalized_url": "neurips.cc/virtual/2026/poster/153846",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-263",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 286,
"source_number": 263,
"title": "TimeSeed: Effective Time Series Forecasting with Sparse Endogenous Variables",
"source_url": "https://proceedings.mlr.press/v306/wu26l.html",
"raw_line": "| P263 | [TimeSeed: Effective Time Series Forecasting with Sparse Endogenous Variables](https://proceedings.mlr.press/v306/wu26l.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P263",
"[TimeSeed: Effective Time Series Forecasting with Sparse Endogenous Variables](https://proceedings.mlr.press/v306/wu26l.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "timeseedeffectivetimeseriesforecastingwithsparseendogenousvariables",
"normalized_url": "proceedings.mlr.press/v306/wu26l.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-264",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 287,
"source_number": 264,
"title": "Parameter Decorrelation via Transition-Variance Alignment for Multivariate Time-series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/choi26j.html",
"raw_line": "| P264 | [Parameter Decorrelation via Transition-Variance Alignment for Multivariate Time-series Forecasting](https://proceedings.mlr.press/v306/choi26j.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P264",
"[Parameter Decorrelation via Transition-Variance Alignment for Multivariate Time-series Forecasting](https://proceedings.mlr.press/v306/choi26j.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "parameterdecorrelationviatransitionvariancealignmentformultivariatetimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/choi26j.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-265",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 288,
"source_number": 265,
"title": "Time–Frequency Non-Stationary Modeling for Multivariate Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152338",
"raw_line": "| P265 | [Time–Frequency Non-Stationary Modeling for Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152338) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P265",
"[Time–Frequency Non-Stationary Modeling for Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152338)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "timefrequencynonstationarymodelingformultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152338",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-266",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 289,
"source_number": 266,
"title": "Information Geometry Loss for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/fang26m.html",
"raw_line": "| P266 | [Information Geometry Loss for Time Series Forecasting](https://proceedings.mlr.press/v306/fang26m.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P266",
"[Information Geometry Loss for Time Series Forecasting](https://proceedings.mlr.press/v306/fang26m.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "informationgeometrylossfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/fang26m.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-267",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 290,
"source_number": 267,
"title": "Existence Precedes Value: Joint Modeling of Observational Existence and Evolving States in Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/148905",
"raw_line": "| P267 | [Existence Precedes Value: Joint Modeling of Observational Existence and Evolving States in Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148905) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P267",
"[Existence Precedes Value: Joint Modeling of Observational Existence and Evolving States in Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148905)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "existenceprecedesvaluejointmodelingofobservationalexistenceandevolvingstatesintimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/148905",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-268",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 291,
"source_number": 268,
"title": "Echoes of Error: Residual-Directional Local Rollout Consistency for Timeseries Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152212",
"raw_line": "| P268 | [Echoes of Error: Residual-Directional Local Rollout Consistency for Timeseries Forecasting](https://neurips.cc/virtual/2026/poster/152212) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P268",
"[Echoes of Error: Residual-Directional Local Rollout Consistency for Timeseries Forecasting](https://neurips.cc/virtual/2026/poster/152212)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "echoesoferrorresidualdirectionallocalrolloutconsistencyfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152212",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-269",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 292,
"source_number": 269,
"title": "Numerion: A Multi-Hypercomplex Model for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/5eca2e4fe7858cbbfef4e08573cfcb25-Abstract-Conference.html",
"raw_line": "| P269 | [Numerion: A Multi-Hypercomplex Model for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/5eca2e4fe7858cbbfef4e08573cfcb25-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P269",
"[Numerion: A Multi-Hypercomplex Model for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/5eca2e4fe7858cbbfef4e08573cfcb25-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "numerionamultihypercomplexmodelfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/5eca2e4fe7858cbbfef4e08573cfcb25-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-270",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 293,
"source_number": 270,
"title": "Anti-Aliasing Matters: A Dynamic Network for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/zhou26aq.html",
"raw_line": "| P270 | [Anti-Aliasing Matters: A Dynamic Network for Time Series Forecasting](https://proceedings.mlr.press/v306/zhou26aq.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P270",
"[Anti-Aliasing Matters: A Dynamic Network for Time Series Forecasting](https://proceedings.mlr.press/v306/zhou26aq.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "antialiasingmattersadynamicnetworkfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/zhou26aq.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-271",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 294,
"source_number": 271,
"title": "Online time series prediction using feature adjustment",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/f418594e90047a10f4c158f70d6701cc-Abstract-Conference.html",
"raw_line": "| P271 | [Online time series prediction using feature adjustment](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f418594e90047a10f4c158f70d6701cc-Abstract-Conference.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P271",
"[Online time series prediction using feature adjustment](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f418594e90047a10f4c158f70d6701cc-Abstract-Conference.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "onlinetimeseriespredictionusingfeatureadjustment",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/f418594e90047a10f4c158f70d6701cc-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-272",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 295,
"source_number": 272,
"title": "MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with $0.1K$ Parameters",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/7db2348b5bfeca620aa7327df815adcc-Abstract-Conference.html",
"raw_line": "| P272 | [MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with $0.1K$ Parameters](https://proceedings.iclr.cc/paper_files/paper/2026/hash/7db2348b5bfeca620aa7327df815adcc-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P272",
"[MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with $0.1K$ Parameters](https://proceedings.iclr.cc/paper_files/paper/2026/hash/7db2348b5bfeca620aa7327df815adcc-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "mixlinearextremelowresourcemultivariatetimeseriesforecastingwith01kparameters",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/7db2348b5bfeca620aa7327df815adcc-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-273",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 296,
"source_number": 273,
"title": "SpikeSTAG: A Dendritic Compartmental Spiking Graph Network for Multivariate Time-Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153490",
"raw_line": "| P273 | [SpikeSTAG: A Dendritic Compartmental Spiking Graph Network for Multivariate Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153490) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P273",
"[SpikeSTAG: A Dendritic Compartmental Spiking Graph Network for Multivariate Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153490)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "spikestagadendriticcompartmentalspikinggraphnetworkformultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153490",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-274",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 297,
"source_number": 274,
"title": "AME-TS: Anchored Mixture-of-Experts for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153106",
"raw_line": "| P274 | [AME-TS: Anchored Mixture-of-Experts for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153106) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P274",
"[AME-TS: Anchored Mixture-of-Experts for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/153106)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "ametsanchoredmixtureofexpertsfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153106",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-275",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 298,
"source_number": 275,
"title": "Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/7a3284ce53f620d244fca0c79d94c478-Abstract-Conference.html",
"raw_line": "| P275 | [Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval](https://proceedings.iclr.cc/paper_files/paper/2026/hash/7a3284ce53f620d244fca0c79d94c478-Abstract-Conference.html) | 2 | 待处理 | 尚未评审；引用较低（2 次，低于 10 次门槛） |",
"raw_cells": [
"P275",
"[Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval](https://proceedings.iclr.cc/paper_files/paper/2026/hash/7a3284ce53f620d244fca0c79d94c478-Abstract-Conference.html)",
"2",
"待处理",
"尚未评审；引用较低（2 次，低于 10 次门槛）"
],
"normalized_title": "enhancingmultivariatetimeseriesforecastingwithglobaltemporalretrieval",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/7a3284ce53f620d244fca0c79d94c478-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "2",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（2 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-276",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 299,
"source_number": 276,
"title": "Not All Frequencies Are Equal: Energy-Adaptive Diffusion for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/qin26e.html",
"raw_line": "| P276 | [Not All Frequencies Are Equal: Energy-Adaptive Diffusion for Time Series Forecasting](https://proceedings.mlr.press/v306/qin26e.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P276",
"[Not All Frequencies Are Equal: Energy-Adaptive Diffusion for Time Series Forecasting](https://proceedings.mlr.press/v306/qin26e.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "notallfrequenciesareequalenergyadaptivediffusionfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/qin26e.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-277",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 300,
"source_number": 277,
"title": "Hi-Time: Hierarchical Latent Prediction for Multivariate Time Series Classification",
"source_url": "https://proceedings.mlr.press/v306/zeng26s.html",
"raw_line": "| P277 | [Hi-Time: Hierarchical Latent Prediction for Multivariate Time Series Classification](https://proceedings.mlr.press/v306/zeng26s.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P277",
"[Hi-Time: Hierarchical Latent Prediction for Multivariate Time Series Classification](https://proceedings.mlr.press/v306/zeng26s.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "hitimehierarchicallatentpredictionformultivariatetimeseriesclassification",
"normalized_url": "proceedings.mlr.press/v306/zeng26s.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-278",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 301,
"source_number": 278,
"title": "DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/d591ef4c631b33ac4e08dd88f1ee1dff-Abstract-Conference.html",
"raw_line": "| P278 | [DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment](https://proceedings.iclr.cc/paper_files/paper/2026/hash/d591ef4c631b33ac4e08dd88f1ee1dff-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P278",
"[DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment](https://proceedings.iclr.cc/paper_files/paper/2026/hash/d591ef4c631b33ac4e08dd88f1ee1dff-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "distdftimeseriesforecastingneedsjointdistributionwassersteinalignment",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/d591ef4c631b33ac4e08dd88f1ee1dff-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-279",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 302,
"source_number": 279,
"title": "MDK-MoE: Multi-view Decomposed Kalman Mixture of Experts Framework for Non-stationary Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/151139",
"raw_line": "| P279 | [MDK-MoE: Multi-view Decomposed Kalman Mixture of Experts Framework for Non-stationary Time Series Forecasting](https://neurips.cc/virtual/2026/poster/151139) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P279",
"[MDK-MoE: Multi-view Decomposed Kalman Mixture of Experts Framework for Non-stationary Time Series Forecasting](https://neurips.cc/virtual/2026/poster/151139)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "mdkmoemultiviewdecomposedkalmanmixtureofexpertsframeworkfornonstationarytimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/151139",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-280",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 303,
"source_number": 280,
"title": "Parametric Prior Mapping Framework for Non-stationary Probabilistic Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/li26fu.html",
"raw_line": "| P280 | [Parametric Prior Mapping Framework for Non-stationary Probabilistic Time Series Forecasting](https://proceedings.mlr.press/v306/li26fu.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P280",
"[Parametric Prior Mapping Framework for Non-stationary Probabilistic Time Series Forecasting](https://proceedings.mlr.press/v306/li26fu.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "parametricpriormappingframeworkfornonstationaryprobabilistictimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/li26fu.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-281",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 304,
"source_number": 281,
"title": "Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/a832b65da26c4867a8581cb13d46ad6d-Abstract-Conference.html",
"raw_line": "| P281 | [Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling](https://proceedings.iclr.cc/paper_files/paper/2026/hash/a832b65da26c4867a8581cb13d46ad6d-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P281",
"[Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling](https://proceedings.iclr.cc/paper_files/paper/2026/hash/a832b65da26c4867a8581cb13d46ad6d-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "areglobaldependenciesnecessaryscalabletimeseriesforecastingvialocalcrossvariatemodeling",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/a832b65da26c4867a8581cb13d46ad6d-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-282",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 305,
"source_number": 282,
"title": "IMTS-Tokenizer: Time-Aware Tokenization for Irregular Multivariate Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154661",
"raw_line": "| P282 | [IMTS-Tokenizer: Time-Aware Tokenization for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154661) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P282",
"[IMTS-Tokenizer: Time-Aware Tokenization for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154661)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "imtstokenizertimeawaretokenizationforirregularmultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154661",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-283",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 306,
"source_number": 283,
"title": "MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning",
"source_url": "https://proceedings.mlr.press/v306/tao26a.html",
"raw_line": "| P283 | [MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning](https://proceedings.mlr.press/v306/tao26a.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P283",
"[MemCast: Memory-Driven Time Series Forecasting with Experience-Conditioned Reasoning](https://proceedings.mlr.press/v306/tao26a.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "memcastmemorydriventimeseriesforecastingwithexperienceconditionedreasoning",
"normalized_url": "proceedings.mlr.press/v306/tao26a.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-284",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 307,
"source_number": 284,
"title": "Nested Spatio-Temporal Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/ai26f.html",
"raw_line": "| P284 | [Nested Spatio-Temporal Time Series Forecasting](https://proceedings.mlr.press/v306/ai26f.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P284",
"[Nested Spatio-Temporal Time Series Forecasting](https://proceedings.mlr.press/v306/ai26f.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "nestedspatiotemporaltimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/ai26f.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-285",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 308,
"source_number": 285,
"title": "MSAR: Next-Scale Autoregressive Forecasting for Time Series via Modular Multi-Scale Decoupling",
"source_url": "https://neurips.cc/virtual/2026/poster/149231",
"raw_line": "| P285 | [MSAR: Next-Scale Autoregressive Forecasting for Time Series via Modular Multi-Scale Decoupling](https://neurips.cc/virtual/2026/poster/149231) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P285",
"[MSAR: Next-Scale Autoregressive Forecasting for Time Series via Modular Multi-Scale Decoupling](https://neurips.cc/virtual/2026/poster/149231)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "msarnextscaleautoregressiveforecastingfortimeseriesviamodularmultiscaledecoupling",
"normalized_url": "neurips.cc/virtual/2026/poster/149231",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-286",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 309,
"source_number": 286,
"title": "ToLD: Efficient Time Series Forecasting via Tokenized Truncated Latent Diffusion",
"source_url": "https://neurips.cc/virtual/2026/poster/156116",
"raw_line": "| P286 | [ToLD: Efficient Time Series Forecasting via Tokenized Truncated Latent Diffusion](https://neurips.cc/virtual/2026/poster/156116) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P286",
"[ToLD: Efficient Time Series Forecasting via Tokenized Truncated Latent Diffusion](https://neurips.cc/virtual/2026/poster/156116)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "toldefficienttimeseriesforecastingviatokenizedtruncatedlatentdiffusion",
"normalized_url": "neurips.cc/virtual/2026/poster/156116",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-287",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 310,
"source_number": 287,
"title": "ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/65e17c5eb95d72ad6540e8f862f38ab1-Abstract-Conference.html",
"raw_line": "| P287 | [ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/65e17c5eb95d72ad6540e8f862f38ab1-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P287",
"[ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/65e17c5eb95d72ad6540e8f862f38ab1-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "prototslearninghierarchicalprototypesforexplainabletimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/65e17c5eb95d72ad6540e8f862f38ab1-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-288",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 311,
"source_number": 288,
"title": "DiffPTS: Rethinking Diffusion ELBO for Probabilistic Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/148542",
"raw_line": "| P288 | [DiffPTS: Rethinking Diffusion ELBO for Probabilistic Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148542) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P288",
"[DiffPTS: Rethinking Diffusion ELBO for Probabilistic Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148542)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "diffptsrethinkingdiffusionelboforprobabilistictimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/148542",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-289",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 312,
"source_number": 289,
"title": "Koopman Generative Operators for Efficient Probabilistic Time-Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/153665",
"raw_line": "| P289 | [Koopman Generative Operators for Efficient Probabilistic Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153665) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P289",
"[Koopman Generative Operators for Efficient Probabilistic Time-Series Forecasting](https://neurips.cc/virtual/2026/poster/153665)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "koopmangenerativeoperatorsforefficientprobabilistictimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/153665",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-290",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 313,
"source_number": 290,
"title": "Probabilistic Circuits for Irregular Multivariate Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152394",
"raw_line": "| P290 | [Probabilistic Circuits for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152394) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P290",
"[Probabilistic Circuits for Irregular Multivariate Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152394)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "probabilisticcircuitsforirregularmultivariatetimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152394",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-291",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 314,
"source_number": 291,
"title": "Beyond Extrapolation: Knowledge Utilization Paradigm with Bidirectional Inspiration for Time Series Forecasting",
"source_url": "https://proceedings.mlr.press/v306/chong26c.html",
"raw_line": "| P291 | [Beyond Extrapolation: Knowledge Utilization Paradigm with Bidirectional Inspiration for Time Series Forecasting](https://proceedings.mlr.press/v306/chong26c.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P291",
"[Beyond Extrapolation: Knowledge Utilization Paradigm with Bidirectional Inspiration for Time Series Forecasting](https://proceedings.mlr.press/v306/chong26c.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "beyondextrapolationknowledgeutilizationparadigmwithbidirectionalinspirationfortimeseriesforecasting",
"normalized_url": "proceedings.mlr.press/v306/chong26c.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-292",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 315,
"source_number": 292,
"title": "Local Geometry Attention for Time Series Forecasting under Realistic Corruptions",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/476ab8f369e489c04187ba84f68cfa68-Abstract-Conference.html",
"raw_line": "| P292 | [Local Geometry Attention for Time Series Forecasting under Realistic Corruptions](https://proceedings.iclr.cc/paper_files/paper/2026/hash/476ab8f369e489c04187ba84f68cfa68-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P292",
"[Local Geometry Attention for Time Series Forecasting under Realistic Corruptions](https://proceedings.iclr.cc/paper_files/paper/2026/hash/476ab8f369e489c04187ba84f68cfa68-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "localgeometryattentionfortimeseriesforecastingunderrealisticcorruptions",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/476ab8f369e489c04187ba84f68cfa68-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-293",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 316,
"source_number": 293,
"title": "TimeSeg: An Information-Theoretic Segment-Wise Explainer for Time-Series Predictions",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/f4fba41b554f9aaa013c4062a1c40518-Abstract-Conference.html",
"raw_line": "| P293 | [TimeSeg: An Information-Theoretic Segment-Wise Explainer for Time-Series Predictions](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f4fba41b554f9aaa013c4062a1c40518-Abstract-Conference.html) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P293",
"[TimeSeg: An Information-Theoretic Segment-Wise Explainer for Time-Series Predictions](https://proceedings.iclr.cc/paper_files/paper/2026/hash/f4fba41b554f9aaa013c4062a1c40518-Abstract-Conference.html)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "timeseganinformationtheoreticsegmentwiseexplainerfortimeseriespredictions",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/f4fba41b554f9aaa013c4062a1c40518-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-294",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 317,
"source_number": 294,
"title": "WILD: Widely Linear Conditioning for Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/148890",
"raw_line": "| P294 | [WILD: Widely Linear Conditioning for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148890) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P294",
"[WILD: Widely Linear Conditioning for Time Series Forecasting](https://neurips.cc/virtual/2026/poster/148890)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "wildwidelylinearconditioningfortimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/148890",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-295",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 318,
"source_number": 295,
"title": "CombinationTS: A Modular Framework for Understanding Time-Series Forecasting Models",
"source_url": "https://proceedings.mlr.press/v306/wang26ay.html",
"raw_line": "| P295 | [CombinationTS: A Modular Framework for Understanding Time-Series Forecasting Models](https://proceedings.mlr.press/v306/wang26ay.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P295",
"[CombinationTS: A Modular Framework for Understanding Time-Series Forecasting Models](https://proceedings.mlr.press/v306/wang26ay.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "combinationtsamodularframeworkforunderstandingtimeseriesforecastingmodels",
"normalized_url": "proceedings.mlr.press/v306/wang26ay.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-296",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 319,
"source_number": 296,
"title": "What if Tomorrow is the World Cup Final? Counterfactual Time Series Forecasting with Textual Conditions",
"source_url": "https://proceedings.mlr.press/v306/gu26r.html",
"raw_line": "| P296 | [What if Tomorrow is the World Cup Final? Counterfactual Time Series Forecasting with Textual Conditions](https://proceedings.mlr.press/v306/gu26r.html) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P296",
"[What if Tomorrow is the World Cup Final? Counterfactual Time Series Forecasting with Textual Conditions](https://proceedings.mlr.press/v306/gu26r.html)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "whatiftomorrowistheworldcupfinalcounterfactualtimeseriesforecastingwithtextualconditions",
"normalized_url": "proceedings.mlr.press/v306/gu26r.html",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-297",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 320,
"source_number": 297,
"title": "DSR-TSF: Spectrum-Driven Dynamic Routing for Efficient Long-Horizon Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/152845",
"raw_line": "| P297 | [DSR-TSF: Spectrum-Driven Dynamic Routing for Efficient Long-Horizon Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152845) | 未知 | 待处理 | 尚未评审；引用数未知，未判为低引用 |",
"raw_cells": [
"P297",
"[DSR-TSF: Spectrum-Driven Dynamic Routing for Efficient Long-Horizon Time Series Forecasting](https://neurips.cc/virtual/2026/poster/152845)",
"未知",
"待处理",
"尚未评审；引用数未知，未判为低引用"
],
"normalized_title": "dsrtsfspectrumdrivendynamicroutingforefficientlonghorizontimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/152845",
"record_type": "legacy_unranked",
"citation_verbatim": "未知",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用数未知，未判为低引用",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-298",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 321,
"source_number": 298,
"title": "Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/34a1fc7890141f1ada3d8bc6199cce07-Abstract-Conference.html",
"raw_line": "| P298 | [Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/34a1fc7890141f1ada3d8bc6199cce07-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P298",
"[Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting](https://proceedings.iclr.cc/paper_files/paper/2026/hash/34a1fc7890141f1ada3d8bc6199cce07-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "unlockingthevalueoftexteventdrivenreasoningandmultilevelalignmentfortimeseriesforecasting",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/34a1fc7890141f1ada3d8bc6199cce07-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-299",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 322,
"source_number": 299,
"title": "SteerCast: Retrieval-Based Latent Steering for Decoder-Only Time Series Forecasting",
"source_url": "https://neurips.cc/virtual/2026/poster/154457",
"raw_line": "| P299 | [SteerCast: Retrieval-Based Latent Steering for Decoder-Only Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154457) | 0 | 待处理 | 尚未评审；引用较低（0 次，低于 10 次门槛） |",
"raw_cells": [
"P299",
"[SteerCast: Retrieval-Based Latent Steering for Decoder-Only Time Series Forecasting](https://neurips.cc/virtual/2026/poster/154457)",
"0",
"待处理",
"尚未评审；引用较低（0 次，低于 10 次门槛）"
],
"normalized_title": "steercastretrievalbasedlatentsteeringfordecoderonlytimeseriesforecasting",
"normalized_url": "neurips.cc/virtual/2026/poster/154457",
"record_type": "legacy_unranked",
"citation_verbatim": "0",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（0 次，低于 10 次门槛）",
"counts_as_independent_review": false
},
{
"source_record_id": "S5-300",
"source_id": "S5",
"source_library_file_id": "libfile_849759d66a808191813dbe9bd1605bb4",
"source_filename": "已粘贴的文本(5).txt",
"source_line": 323,
"source_number": 300,
"title": "Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models",
"source_url": "https://proceedings.iclr.cc/paper_files/paper/2026/hash/d56a1a2b24b1906345fc2cd8276b11ed-Abstract-Conference.html",
"raw_line": "| P300 | [Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models](https://proceedings.iclr.cc/paper_files/paper/2026/hash/d56a1a2b24b1906345fc2cd8276b11ed-Abstract-Conference.html) | 1 | 待处理 | 尚未评审；引用较低（1 次，低于 10 次门槛） |",
"raw_cells": [
"P300",
"[Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models](https://proceedings.iclr.cc/paper_files/paper/2026/hash/d56a1a2b24b1906345fc2cd8276b11ed-Abstract-Conference.html)",
"1",
"待处理",
"尚未评审；引用较低（1 次，低于 10 次门槛）"
],
"normalized_title": "quadraticdirectforecastfortrainingmultisteptimeseriesforecastmodels",
"normalized_url": "proceedings.iclr.cc/paper_files/paper/2026/hash/d56a1a2b24b1906345fc2cd8276b11ed-abstract-conference.html",
"record_type": "legacy_unranked",
"citation_verbatim": "1",
"legacy_status_verbatim": "待处理",
"legacy_unranked_reason_verbatim": "尚未评审；引用较低（1 次，低于 10 次门槛）",
"counts_as_independent_review": false
}
]