{ "rubric_version": "PAPER-RUBRIC-1.0", "paper_id": "ATT-0100", "paper_title": "xPatch: Dual-Stream Time Series Forecasting with Exponential Seasonal-Trend Decomposition", "reviewer": "B", "analysis_mode": "standard", "paper_version": "2412.17323v3", "version_date": "2025/02/11", "first_public_date": "2024/12/23", "assessment_date": "2026-11-08", "reading_started_at": "2026-11-08T02:47:51.896970+00:00", "reading_completed_at": "2026-11-08T02:50:05.763811+00:00", "read_physical_pages": [ 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25 ], "visual_checks_physical_pages": [ 5, 11, 17, 21, 22, 23, 24, 25 ], "pdf_sha256": "032d723a6df6df4d66df52ada8ce695594a21b49493374439f0890fc030c1765", "legacy_review_read_first": true, "legacy_full_review_available": false, "legacy_status": "仅有历史清单行或风险摘要；未提供完整旧评，不能声称复读完整旧评。", "legacy_records": [ { "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 } ], "new_peer_reviews_read": false, "review_text": "先阅读历史摘要，再通读2412.17323v3的二十五页，包括全部实验表与预测图。完整旧评未提供。本次首先修正两项旧风险的适用范围：第16—17页明确说明表13—15取三个随机种子平均，不能继续说未发现种子重复；第17页逐表列明哪些结果转载、哪些模型重跑，也给出不丢弃最后批次的TFB比较。这些材料比历史摘要丰富。当前官方README同样区分发表结果与公平协议，二者不构成已证实的直接来源冲突；仍未提供波动区间和每格完整运行来源。\n\n核心思想是用指数移动平均拆出平滑信号与残差，分别经MLP和分块卷积，再合并预测。它比只换一个骨干更关注训练损失、学习率与预处理的整体作用，且把EMA、损失、调度器移植到多种基线作消融，工程可复用价值较高。EMA、双流、深度可分离卷积和MAE均已有明确背景，原创性主要是适合预测任务的组合，而非这些工具首次出现。把残差命名为季节性不保证其周期或平稳，趋势也未必线性；这些名称只能视为操作性分量定义。\n\n理论和方法表述有几处实质缺陷。第11页把一个长度随输入增长的加权点积称为O(1)，标准顺序计算模型中点积仍需O(n)乘加，向量化不等于常数总运算量。对每个前缀分别点积还可能增加总工作量；论文没有给出能把完整分解变成常数复杂度的算法。因此暂停O(1)这一特定主张，不能由此否定EMA递推本身正确。第4页所谓线性流含依赖输入统计量的LayerNorm，整体并非严格线性映射；第5页式8把N×P结果与N×P²残差直接相加，按印刷维度不相容，框图存在残差全连接层，说明更可能漏写投影，实际代码是否正确本次未知。最后固定线性层融合两流，也不能直接称输入条件化的动态门控。\n\n历史关于ADF的警示成立且需要讲清原因。附录C先对多个长度720的片段做单位根检验，再平均p值并用0.05判断整体平稳性；平均p值没有这里所需的校准，拒绝单位根不是一般严格平稳的充分证明，不拒绝也不等于已证明非平稳。把Traffic滤波后更高p值解释为成功提取弱非平稳趋势，混合了平滑导致自相关和检验功效的变化。EMA对弱平稳输入的固定稳定线性滤波在稳态下仍可保持弱平稳，窗口初始化造成的瞬态不等于发现真实趋势。因此这组检验支持描述差异，不足以证明所称成分机制，更不足以据此唯一选出α=0.3。\n\n损失的“反正切”作用在预测步位置而非残差幅度，实质是正权重MAE。其极限权重为1-π/4，约0.215，所以比CARD的远期衰减保留更多长预测监督；抗大残差的性质主要来自MAE，论文没有污染强度或异常点压力测试来验证新的抗异常机制。附录把i的负幂叫指数函数也不准确。表9多数MAE改善，但MSE存在退化，如DLinear在Traffic的96步从0.650变0.687而MAE下降，这体现目标取舍，不能概括成所有指标一致改善。多处损失公式把预测点索引和整条向量混用，需与文字意图区分。\n\n经验结果仍然有价值。固定输入长度与搜索输入长度分开报告，附录还有完整四长度结果，避免只凭均值判断。表2中xPatch在多组数据领先，但Exchange平均MSE为0.366，不及DLinear的0.297；Traffic为0.392，不及iTransformer的0.378。更可比较的无末批丢弃表15里，Weather、Traffic、Electricity等仍较强，ETTh1平均MSE0.413落后PDF的0.407，ETTh2的0.341落后FITS的0.335。ILI从搜索表的1.281变为公平协议的1.633，说明协议选择会显著改变绝对成绩，不能把两个表的胜出比例与幅度无条件混用；这也不是证明差异仅由drop-last造成，训练配置与数据覆盖可能共同改变。\n\n消融详尽是本篇强项：EMA替换、多流反转、单流、分位不涉及但损失衰减、学习率参数及RevIN都有对照，且部分改进在其他模型可迁移。不过单流与双流未等容量，损失改变和网络设计的作用不完全隔离，三种子均值缺标准差，不支持把千分位差异当稳定排序。表11的Weather完整模型项与表13同设置主结果不同，可能涉及RevIN配置；应视为配置映射未解释，而不是任意挑一个数字作结果造假判断。长输入比较也不能证明整个Transformer类别无法保留顺序，带位置结构的注意力模型不是简单的无序集合运算。\n\n资源方面，官方README提供环境、统一/搜索/公平三类脚本和消融入口，本次只读说明，没有安装或运行。表3报告每步时间，xPatch训练3.099毫秒、推理1.303毫秒，较列出的Transformer快，却明显慢于DLinear的0.420、0.310毫秒；缺批量、计时预热和方差等完整说明，不能解释成全生命周期成本最优。总体适合保留作非Transformer强基线，避免把复杂度、平稳性和动态融合的术语等同于已证机制。值得思考的是收益主要来自分解、预测位置权重还是训练预算；若替换损失和调度器后架构优势显著缩小，结构独立贡献的解释就应收窄。优先亲读第17—20页的结果来源与公平协议，再对照第10—11页ADF和复杂度论证。这些是判断边界，不是给用户的实验清单。", "review_hanzi_count": 1524, "claims": [ { "id": "CL01", "claim": "双流预测在多个长预测基准具有竞争力", "status": "supported_with_scope", "pages": [ 18, 19, 20 ] }, { "id": "CL02", "claim": "EMA向量点积把完整分解变成O(1)", "status": "suspended", "pages": [ 11 ] }, { "id": "CL03", "claim": "平均ADF p值证明一般平稳分量分离", "status": "not_supported", "pages": [ 10, 11 ] }, { "id": "CL04", "claim": "未报告种子重复或来源完全矛盾的旧风险", "status": "legacy_partly_resolved", "pages": [ 16, 17 ] } ], "scores": [ { "dimension_id": "A1", "dimension_name": "问题重要性", "score": 80, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "分解与高效非Transformer预测具有实际价值。", "evidence_pages": [ 1, 2 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "A2", "dimension_name": "原创性", "score": 55, "status": "tentative", "confidence": "中", "judgment_range": [ 45, 65 ], "rationale": "已知EMA、双流卷积及加权MAE的任务化组合，先行全文核查有限。", "evidence_pages": [ 2, 3, 5 ], "missing_information": "见理由与材料边界", "source_ids": [ "S01" ] }, { "dimension_id": "A3", "dimension_name": "实质推进程度", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "强预测表现及可移植训练组件，但多项机制表述过度。", "evidence_pages": [ 7, 12, 14, 16, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "A4", "dimension_name": "知识与解释增量", "score": 55, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "丰富消融可学习，平稳性解释与复杂度论证错误削弱增量。", "evidence_pages": [ 10, 11, 13 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "B1", "dimension_name": "问题定义与评价有效性", "score": 75, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "统一、搜索和无drop-last协议分列，目标指标清楚。", "evidence_pages": [ 6, 7, 17, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "B2", "dimension_name": "技术与论证正确性", "score": 45, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "O(1)错误、残差维度漏投影、严格线性和ADF解释不成立。", "evidence_pages": [ 4, 5, 10, 11 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "B3", "dimension_name": "数据与评测流程完整性", "score": 65, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "划分、三种子及来源披露，归一化配置映射和完整选择程序未清。", "evidence_pages": [ 10, 16, 17 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "B4", "dimension_name": "比较对象与比较公平性", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "公平表含PDF/FITS且消融跨模型，转载与重跑仍未完全同预算。", "evidence_pages": [ 12, 14, 17, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "C1", "dimension_name": "核心主张的证据充分度", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "支持多数据强经验表现，复杂度和平稳机制不支持。", "evidence_pages": [ 7, 11, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "C2", "dimension_name": "机制归因与竞争解释区分度", "score": 60, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "较多组件和可迁移消融，但容量、训练目标交互未全隔离。", "evidence_pages": [ 12, 13, 14, 16 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "C3", "dimension_name": "结果稳定性与不确定性处理", "score": 55, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "明确三种子均值，缺标准差/配对区间，不能沿用无重复旧判断。", "evidence_pages": [ 16, 17 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "C4", "dimension_name": "适用范围与边界验证", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "多领域和长度、失败设置充分，噪声压力和跨域迁移不足。", "evidence_pages": [ 18, 19, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "D1", "dimension_name": "过程透明度", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "v3来源解释详细，部分消融配置与主表对应仍不清。", "evidence_pages": [ 16, 17 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "D2", "dimension_name": "研究资源可获得性", "score": 80, "status": "tentative", "confidence": "中", "judgment_range": [ 70, 85 ], "rationale": "官方环境、脚本、消融入口可访问，未执行。", "evidence_pages": [ 1 ], "missing_information": "见理由与材料边界", "source_ids": [ "S01" ] }, { "dimension_id": "D3", "dimension_name": "主张克制与报告一致性", "score": 50, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "主文广泛概括，O(1)、平稳保证、动态融合等需收窄。", "evidence_pages": [ 4, 5, 7, 11 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "D4", "dimension_name": "表达清晰度", "score": 65, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "流程和附录丰富，维度、符号及术语不精确。", "evidence_pages": [ 4, 5, 11, 13 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "E1", "dimension_name": "复杂度合理性", "score": 70, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "分量分别建模合理，双流额外成本有收益但非最简。", "evidence_pages": [ 4, 7, 13 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "E2", "dimension_name": "资源效率与成本收益", "score": 65, "status": "tentative", "confidence": "中", "judgment_range": [ 50, 75 ], "rationale": "提供训练推理单步时间，比DLinear慢；完整测量边界欠缺。", "evidence_pages": [ 7 ], "missing_information": "见理由与材料边界", "source_ids": [ "S01" ] }, { "dimension_id": "E3", "dimension_name": "可迁移与可复用程度", "score": 80, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "EMA、加权损失与调度器可独立移植且有多模型对照。", "evidence_pages": [ 12, 14, 16 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "E4", "dimension_name": "后续研究启发性", "score": 80, "status": "rated", "confidence": "高", "judgment_range": null, "rationale": "促使区分预处理、目标与结构贡献及公平评测协议。", "evidence_pages": [ 17, 20 ], "missing_information": "", "source_ids": [ "S01" ] }, { "dimension_id": "F1", "dimension_name": "与当前问题的相关性", "score": null, "status": "unknown", "confidence": "低", "judgment_range": null, "rationale": "未给出用户具体研究问题、已有知识、资源与当前决策，不能从论文的一般价值推断。", "evidence_pages": [], "missing_information": "个人背景未提供", "source_ids": [] }, { "dimension_id": "F2", "dimension_name": "相对于已有知识的学习收益", "score": null, "status": "unknown", "confidence": "低", "judgment_range": null, "rationale": "未给出用户具体研究问题、已有知识、资源与当前决策，不能从论文的一般价值推断。", "evidence_pages": [], "missing_information": "个人背景未提供", "source_ids": [] }, { "dimension_id": "F3", "dimension_name": "当前使用可行性", "score": null, "status": "unknown", "confidence": "低", "judgment_range": null, "rationale": "未给出用户具体研究问题、已有知识、资源与当前决策，不能从论文的一般价值推断。", "evidence_pages": [], "missing_information": "个人背景未提供", "source_ids": [] }, { "dimension_id": "F4", "dimension_name": "当前决策信息价值", "score": null, "status": "unknown", "confidence": "低", "judgment_range": null, "rationale": "未给出用户具体研究问题、已有知识、资源与当前决策，不能从论文的一般价值推断。", "evidence_pages": [], "missing_information": "个人背景未提供", "source_ids": [] } ], "risks": [ { "severity": "major", "scope": "CL02", "issue": "长度n点积至少O(n)工作，向量化不构成O(1)证明。", "pages": [ 11 ], "action": "暂停常数复杂度结论。" }, { "severity": "major", "scope": "CL03", "issue": "平均p值未校准，单位根检验不能承担一般平稳证明。", "pages": [ 10, 11 ], "action": "将机制证据降为描述。" }, { "severity": "moderate", "scope": "reproduction", "issue": "式8漏投影、部分消融配置未对齐，未核代码。", "pages": [ 5, 16, 18 ], "action": "保留实现与配置未知。" } ], "aggregation": { "weights": null, "weighted_score_observed": null, "weighted_coverage": null, "tentative_weight_share": null, "notes": "保留24维原始分，不擅自合成总分；未知不作0。" }, "sources": [ { "source_id": "S01", "source_type": "arxiv PDF", "location": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0100/ATT-0100_paper.pdf", "version": "2412.17323v3" }, { "source_id": "S02", "source_type": "待核历史记录", "location": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0100/legacy_reference_only.json" } ], "limits": { "no_model_execution": true, "no_training_or_inference": true, "no_full_code_audit": true, "external_closest_prior_full_text_check": "本次未独立取得并阅读全文，原创性为暂定判断。", "author_result_not_independent_reproduction": true }, "resource_read": { "url": "https://github.com/stitsyuk/xPatch/blob/main/README.md", "blob_sha": "d76efe851768d60084316340112eaa2a0d5db316", "notes": "已读三套协议与drop-last说明、环境/脚本入口；未执行。" }, "legacy_reconciliation": "v3第16—17页明确三种子平均并详细区分转载与重跑，纠正历史摘要未定位重复以及来源边界未澄清的范围。" }