{
"candidate_id": "ATT-0097",
"reviewer": "D",
"rubric": "PAPER-RUBRIC-1.0",
"review_status": "frozen",
"identity": {
"candidate_id": "ATT-0097",
"title": "Adaptive Multi-Scale Decomposition Framework for Time Series Forecasting",
"authors": [
"Hu, Yifan",
"Liu, Peiyuan",
"Zhu, Peng",
"Cheng, Dawei",
"Dai, Tao"
],
"arxiv_id": "2406.03751",
"arxiv_version": "v2",
"first_arxiv_date": "2024/06/06",
"latest_arxiv_date": "2025/04/15",
"source_urls": [
"https://ojs.aaai.org/index.php/AAAI/article/view/33908"
],
"publication_status": "AAAI 2025, DOI 10.1609/aaai.v39i16.33908 verified",
"full_text_path": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0097/ATT-0097_paper.pdf",
"metadata_source_path": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0097/ATT-0097_arxiv.html",
"full_text_sha256": "02aad9b5b876e3b53d6c34114d6e658eb557898d57f20402baf512b3e264b05e",
"review_state": "ready_for_five_new_independent_reviews",
"completed_new_reviews": 0,
"source_versions_not_asserted_identical": true
},
"analysis_date": "2026-11-08",
"reading_record": "2026-11-08 UTC，先读旧风险摘要，再读14页全文（正文1–7、参考8–9、证明9–10、附录算法/配置/结果11–14），视觉复核式10、证明12–32、参数表6、资源图5、敏感性图6–7。旧两项证明疑点重新构造反例，不读其他新评。",
"freeze_time_utc": "2026-11-08T02:46:45.676494+00:00",
"review_type": "independent_full_text_review",
"independence": "未读取其他新评；先读取旧历史条目，再检查本版本全文和附录。",
"legacy": {
"full_review_available": false,
"counts_as_review": false,
"source_sha256": "6de5413cf2ea9ef97ef6aa33f864a935510f08b9c4047d54215032656af6e175",
"records": [
{
"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
}
]
},
"claims": [
"多尺度混合、双依赖建模和密集专家加权提供自适应预测。",
"经验消融较完整，理论误差界证明与门控记号存在需要修正的问题。"
],
"review_body": "AMD 把多尺度平均池化与自顶向下混合、时间和通道依赖、输入相关的多预测器加权组合到一个 MLP 框架中。其问题动机合理：不同时间分辨率的模式对未来影响不同，固定相加未必最合适。相比仅做趋势季节分解，多个尺度与动态门控提供了更灵活的表示。但 TimeMixer、多尺度混合、线性专家和 MoE 门控都是紧密先例，主要新意应定位于具体模块组织与选择器，而不是首次发现多尺度或动态专家。论文引用了 Mixture-of-Linear-Experts，却未将它纳入核心直接比较，限制了对选择器独立贡献和原创性的判断。\n\n长期预测表覆盖常用九个具名数据集及四种长度，短期预测覆盖四个 PEMS 集。TimeMixer 作为直接近邻确实出现，且作者明确说所有基线重跑并搜索六种输入长度，不能沿用“只挑弱基线”的概括。五次重复也在正文说明，附录表七给六个长期数据集的五种子均值和小标准差，这是稳定性证据的实际优点。尚不清楚的是最优输入长度根据哪个划分选择、各模型其他超参的搜索预算是否一致，以及基线重复的方差。本文只能说选择程序和公平预算未充分公开，不能据此断言用了测试集调参。\n\n结果体现竞争力而非稳定统治。ETTh1 的中长预测、ETTm1 的短预测、Solar 的部分设置较好，短期 PEMS 的均方误差也明显有利。但 ETTh2 七百二十步 AMD 为零点四零二，TimeMixer 为零点三四二；Exchange 最长预测 DLinear 为零点六四三，AMD 为零点七五零；Traffic 多项也落后。PEMS07 的 MSE 虽优，MAE 却不及表列 PatchTST。作者反复写“始终最先进”，没有充分覆盖这些重要逆例。短期表还没有列出长期最直接对手 TimeMixer 或有图结构的专门交通预测模型，因此应区分一般骨干上的短期表现和交通领域最强比较。\n\n消融是本文比较扎实的部分。Weather 和 ECL 上分别去除 MDM、DDI、AMS，改变通道混合强度，比较稀疏专家、随机顺序和平均权重，并删除负载平衡损失。平均权重和随机顺序比完整门控差，确实支持输入相关组合与顺序信息的价值，比单纯删除模块更接近机制对照。五个尺度层数及降采样倍率的附表、预测器数和权重正则敏感性也充分。但多数组件删除会改变容量；随机打乱还破坏输入分布，不能独自证明“收益并非参数增多”。需要一个参数量匹配的常规 MLP、多头线性预测器以及同样门控的 TimeMixer 对照。选择器权重图说明模型如何分配权重，不证明每个专家已识别可解释的真实时间尺度。\n\n旧历史摘要提出两个证明疑点，本次重新逐行和视觉核对后确认。第一，附录第十页式十三到十四把分母从索引差变为降采样率乘索引差，却保留原来的前置平均因子，漏掉倍率。用原序列 f(j)=j 和降采样率二即可看出：池化后的粗索引斜率是二，不是仍小于等于一。正确结论至多是 Lipschitz 常数随粗索引尺度变化，不能直接沿用原常数。第二，式二十一到二十二把不同列权重下的加权值差，用原信号差乘最大权重上界代替，忽略了权重列本身的变化。取常数信号、零 Lipschitz 常数，但两个输出列权重不同，左侧可以非零而右侧为零，便构成反例。\n\n理论还有更深的目标错位。证明中的预测构造使用周期加一的位置，并把输入终点和周期联系起来，却没有保证该位置总在已知历史内；非周期令周期趋于无穷的处理更不能直接实现成有限历史线性预测器。多尺度混合中任意权重虽是训练后常数矩阵，也不保证对时间索引保留所写的平滑界。即使修复为一个有限但宽松的误差界，对有限时域的平滑序列，简单末值保持已经可获得随预测距离增长的界，仍不能证明分解提高精度或选择器学到了主导模式。因此本版定理证明不能当作经验优势的理论保障，应该与可运行的工程模块分开评分。\n\n进一步检查实现描述发现一个需澄清的门控边界。式十上分支写成系数乘指数再减一，下分支是系数乘对数；表六多个数据集系数为零。按公式直读，此时原本较大的候选分量会映射到负一，较小分量映射到零，随后 softmax 反而降低前者权重，与“强化主导模式”方向相反。可能是系数命名或括号排版没有对应实际实现，但当前文档无法自行消解。本文没有执行仓库，不能认定实现也犯同一错误；应要求代码级对应说明，而非凭意图代替公式。\n\n可复现材料包括官方仓库、分割大小、逐数据集补丁长度、批量、训练轮次、学习率、隐藏宽度和层数，基础准备较充分。资源测量在同一 Weather 设置、相同批量下比较训练每迭代时间与显存，图上八预测器 AMD 约十七毫秒、一千三百四十九兆字节，低于较重 Transformer，却不比 DLinear 的七毫秒和九百九十九兆字节更轻。因此合理结论是精度资源折中，不是对所有 MLP 同时更快更省。没有推理延迟、完整调参总成本以及跨数据维度扩展曲线，插件实验所谓相同资源也缺对应实测表。MDM 与 AMS 接到 DLinear、MTS-Mixers 后平均提升有价值，但个别任务退化必须保留。\n\n综合而言，多尺度混合加动态预测器是一条有实证基础的路线，消融与种子报告优于不少同类论文；主要扣分来自无法成立的理论链条、门控文档歧义和过度概括的优势主张。后续最有价值的是把门控定义、理论条件和等容量近邻比较核清，而不是继续堆更多标准数据集。旧完整评语没有取得，历史分数不计入本次结论；本次阅读覆盖当前版本全文与附录，没有训练或推理执行。",
"review_chinese_character_count": 1749,
"dimensions": [
{
"dimension": "A1",
"name": "问题重要性",
"score": 80,
"status": "rated",
"confidence": "high",
"reason": "多尺度和依赖权重是有意义的预测问题",
"evidence_pages": "1–3",
"judgment_range": null
},
{
"dimension": "A2",
"name": "原创性",
"score": 60,
"status": "tentative",
"confidence": "mid",
"reason": "多尺度与MoE组合，缺线性专家直接近邻充分对照",
"evidence_pages": "2–4,8",
"judgment_range": [
50,
70
]
},
{
"dimension": "A3",
"name": "实质推进程度",
"score": 65,
"status": "rated",
"confidence": "high",
"reason": "常用长期和短期任务有收益，非一致最优",
"evidence_pages": "5–7",
"judgment_range": null
},
{
"dimension": "A4",
"name": "知识与解释增量",
"score": 45,
"status": "rated",
"confidence": "high",
"reason": "经验门控有价值，理论证明错误削弱解释增量",
"evidence_pages": "3–4,9–10",
"judgment_range": null
},
{
"dimension": "B1",
"name": "问题定义与评价有效性",
"score": 75,
"status": "rated",
"confidence": "high",
"reason": "MSE/MAE与长度明确，短期任务最强对照不足",
"evidence_pages": "3,5–6",
"judgment_range": null
},
{
"dimension": "B2",
"name": "技术与论证正确性",
"score": 35,
"status": "rated",
"confidence": "high",
"reason": "降采样倍率及不同列权重不等式有反例，门控零系数需解释",
"evidence_pages": "4,10,12",
"judgment_range": null
},
{
"dimension": "B3",
"name": "数据与评测流程完整性",
"score": 70,
"status": "rated",
"confidence": "mid",
"reason": "分割大小及配置可查，选择划分与测试末批未定位",
"evidence_pages": "5,11–12",
"judgment_range": null
},
{
"dimension": "B4",
"name": "比较对象与比较公平性",
"score": 65,
"status": "rated",
"confidence": "high",
"reason": "基线重跑搜索输入长度是优点，其他预算与近邻MoLE缺失",
"evidence_pages": "5–6,8,11–12",
"judgment_range": null
},
{
"dimension": "C1",
"name": "核心主张的证据充分度",
"score": 60,
"status": "rated",
"confidence": "high",
"reason": "若干优势确立但始终最优及理论支撑不成立",
"evidence_pages": "5–7,10",
"judgment_range": null
},
{
"dimension": "C2",
"name": "机制归因与竞争解释区分度",
"score": 70,
"status": "rated",
"confidence": "high",
"reason": "平均权重、顺序、稀疏专家、删除及超参实验丰富，容量未完全匹配",
"evidence_pages": "6–7,12–14",
"judgment_range": null
},
{
"dimension": "C3",
"name": "结果稳定性与不确定性处理",
"score": 75,
"status": "rated",
"confidence": "high",
"reason": "五次重复及六数据标准差，小优势仍缺基线配对检验",
"evidence_pages": "5,12,14",
"judgment_range": null
},
{
"dimension": "C4",
"name": "适用范围与边界验证",
"score": 65,
"status": "rated",
"confidence": "high",
"reason": "常用多域长期与PEMS短期，缺噪声漂移压力验证",
"evidence_pages": "5–6,11,14",
"judgment_range": null
},
{
"dimension": "D1",
"name": "过程透明度",
"score": 65,
"status": "rated",
"confidence": "high",
"reason": "详细配置与伪码好，门控系数和部分索引影响精确重建",
"evidence_pages": "4,11–12",
"judgment_range": null
},
{
"dimension": "D2",
"name": "研究资源可获得性",
"score": 65,
"status": "tentative",
"confidence": "mid",
"reason": "官方代码链接，未检查提交、运行和复现",
"evidence_pages": "1,11",
"judgment_range": [
50,
80
]
},
{
"dimension": "D3",
"name": "主张克制与报告一致性",
"score": 45,
"status": "rated",
"confidence": "high",
"reason": "一致最优、理论保证及同资源插件表述超出表证据",
"evidence_pages": "1–2,5–7,10",
"judgment_range": null
},
{
"dimension": "D4",
"name": "表达清晰度",
"score": 60,
"status": "rated",
"confidence": "high",
"reason": "模块清楚，但数学与参数歧义不小",
"evidence_pages": "3–4,10–12",
"judgment_range": null
},
{
"dimension": "E1",
"name": "复杂度合理性",
"score": 65,
"status": "rated",
"confidence": "mid",
"reason": "三模块有消融，八专家成本需与简单同容量模型权衡",
"evidence_pages": "6–7",
"judgment_range": null
},
{
"dimension": "E2",
"name": "资源效率与成本收益",
"score": 65,
"status": "tentative",
"confidence": "mid",
"reason": "同批Weather训练显存/时间有证据，不领先DLinear、缺推理总成本",
"evidence_pages": "7",
"judgment_range": [
50,
75
]
},
{
"dimension": "E3",
"name": "可迁移与可复用程度",
"score": 70,
"status": "rated",
"confidence": "high",
"reason": "两个外部骨干插件有效但存在局部退化",
"evidence_pages": "7",
"judgment_range": null
},
{
"dimension": "E4",
"name": "后续研究启发性",
"score": 70,
"status": "rated",
"confidence": "high",
"reason": "门控、尺度分解和依赖权衡可启发后续研究",
"evidence_pages": "6–7,12–14",
"judgment_range": null
},
{
"dimension": "F1",
"name": "与当前问题的相关性",
"score": null,
"status": "unknown",
"confidence": "low",
"reason": "当前问题未知",
"evidence_pages": "用户背景未提供",
"judgment_range": null
},
{
"dimension": "F2",
"name": "相对于已有知识的学习收益",
"score": null,
"status": "unknown",
"confidence": "low",
"reason": "已有知识未知",
"evidence_pages": "用户背景未提供",
"judgment_range": null
},
{
"dimension": "F3",
"name": "当前使用可行性",
"score": null,
"status": "unknown",
"confidence": "low",
"reason": "当前算力及任务约束未知",
"evidence_pages": "用户背景未提供",
"judgment_range": null
},
{
"dimension": "F4",
"name": "当前决策信息价值",
"score": null,
"status": "unknown",
"confidence": "low",
"reason": "当前决策未知",
"evidence_pages": "用户背景未提供",
"judgment_range": null
}
],
"missing_and_limits": [
"仅有旧摘要与评分行，完整旧评缺失；旧结论经独立全文复核。",
"未运行代码；式十与表六冲突只能确认文档层面。",
"当前arXiv v2未与AAAI最终版逐字比对。",
"代码 https://github.com/TROUBADOUR000/AMD ，未核验执行。",
"最佳输入长度选择划分、预算及基线方差未完全公开。",
"F组背景不足全部未知。"
],
"experiments_executed": false,
"aggregate_score": null
}