{ "candidate_id": "ATT-0272", "reviewer": "D", "rubric": "PAPER-RUBRIC-1.0", "review_status": "frozen", "identity": { "candidate_id": "ATT-0272", "title": "MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with $0.1K$ Parameters", "authors": [ "Ma, Aitian", "Luo, Dongsheng", "Sha, Mo" ], "arxiv_id": "2410.02081", "arxiv_version": "v2", "first_arxiv_date": "2024/10/02", "latest_arxiv_date": "2026/02/16", "source_urls": [ "https://proceedings.iclr.cc/paper_files/paper/2026/hash/7db2348b5bfeca620aa7327df815adcc-Abstract-Conference.html" ], "publication_status": "ICLR 2026 official proceedings verified", "full_text_path": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0272/ATT-0272_paper.pdf", "metadata_source_path": "/workspace/scratch/ac01e91af073/candidate_recovery/priority_packet_20261108/papers/ATT-0272/ATT-0272_arxiv.html", "full_text_sha256": "bccc6a5a9d54f871fedb758c93424d09c02b6dfeb3dae3644e81bac3df622792", "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，先读历史候选行，再逐页阅读26页全文（正文1–9、参考10–13、附录A–E 14–26）；视觉复核第4、14页公式、第19页参数图及24–25页预测图，逐项核对主附表关键对照与复杂度。未读其他新评，未执行模型。", "freeze_time_utc": "2026-11-08T02:39:12.711275+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": "aaf3fb2b1d7c6a36247d88103dd4a3208965300764f5bf847ad8fe9f60fd1db5", "records": [ { "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 } ] }, "claims": [ "通过分段时间线性映射和低秩频谱映射，实现极少参数的共享通道预测。", "论文报告常用任务上的精度成本折中，但统一最优及极限计数仍需澄清。" ], "review_body": "旧材料只有“尚未评审、引用低于门槛”的候选行，没有完整旧评。本次不沿用引用门槛，也不以会议录收录代替方法判断。MixLinear 的实质构思是先聚合并降采样，再把共享通道的线性预测拆成时间分段映射与低秩频域映射，最后相加恢复预测。面向极小参数模型的方向有意义，尤其将 SparseTSF、FITS 这两个直接近邻纳入比较，优于只对比大型 Transformer 来显示轻量。分段因子化与频谱压缩的组合可以带来有用的精度和参数折中，但这些操作分别是成熟工具，论文所谓“谱时分解原理”没有给出新的定理或可检验的普适条件。\n\n主要数值支持的是轻量竞争力，而非统一最优。表五中 ETTh1 四个预测长度均较好，Exchange 的中长预测也超过 SparseTSF；不过 Electricity 和 Traffic 的多数预测长度不及 SparseTSF，ETTh2 多项落后 FITS，Weather 的四项也没有领先。因此相对某一个对手最高百分之十六点二的改进不能当作总体优势。主文说八个数据集，附表又多出 Weather，应补齐其数据配置。主表主要用相对百分比展示其他模型，附表给原始误差是必要补充，但两处并非全部一致：例如 ETTh1 的九十六步 DLinear，表一给相对 SparseTSF 负百分之十一，按零点三六二反推约零点四零二，而表五是零点三八四。这不是普通四舍五入能解释的差别，需要标明来源或配置变化。\n\n基线公平性不能由“统一七百二十输入长度”一项确认。附录明确说基线结果来自 FITS 第一版，但未逐项分清哪些精度数字直接引用、哪些重新训练，SparseTSF 等更新方法的来源如何衔接；自身又有分段长度、谱秩及降采样率的搜索。没有统一调参预算、数据切分边界和测试末批处理的核验记录，比较应保留折扣。本文正文与附录未定位随机种子次数、标准差或重复时延的分布。很多改进只有千分之一量级，单点均值不足以区分稳定优势与训练波动。\n\n技术书写存在妨碍复现的实质问题，不只是措辞。正文把频谱变换矩阵定义为两个涉及完整降采样长度的复数矩阵，那么实参数数目应随完整长度乘谱秩增长，却在同页写成四倍分段长度乘谱秩，两个长度未建立等价关系。附录又把编码和解码矩阵写成实数，引入正文未说明的低通截断，模型是否同一参数化不清楚。第十四页式六经视觉核对，第一矩阵写成根号补齐长度乘隐维，却直接左乘方形输入，通常维度不相容，应转置或重新定义。伪代码把原序列重排为降采样长度乘周期矩阵后，又直接重排成只含降采样长度元素的方阵，没有明确取样或聚合掉周期轴；非平方长度的填充与任意输出长度的裁剪也未交代。可修正的形状错误不证明实际代码运行失败，但按现有论文无法唯一还原声称的极小模型。\n\n相关理论解释也应降级。两个线性层组合仍是线性映射，引用通用逼近定理不能证明其保留任意表达能力；文中反复称“正交”，却没有正交约束或相应性质。对于不额外限制的完整频谱低秩线性算子，傅里叶换基本身不创造更低的算子秩。要证明频域而非单纯低秩正则化是收益来源，需要同参数的时域低秩对照、不同截频策略和真实谱能量分析。把通道独立共享模型的稳定性解释为“跨通道信息补偿”也过强：共享训练样本确能改变估计，但这里没有显式通道交互的证据。二次复杂度降为线性或近线性应称多项式阶数降低，不是正文声称的指数级改进。\n\n现有消融确实有价值。六个数据集七百二十步比较去除时间支路、去除频谱支路和完整模型，完整模型误差均较小；频谱支路尤其帮助 Traffic，时间支路在 ETTh1 较强。分段长度、谱秩和降采样率的附录扫描也不是缺失。问题是删除支路同时减少参数与计算，未排除容量差异；仅凭谱秩增加后误差接近，就断言自然序列必然集中在少数频率，并不充分。降采样表还显示二十四的周期不是所有情况最好，压缩过强有代价。更值得保留的是这些可操作的超参经验，而非普遍机制结论。\n\n低资源主张必须拆开看。四十五至一百七十六个参数的报告很吸引人，但参数表、图和正文有明显数量口径冲突：同样七百二十设置的 FITS 在正文约一万零五百，表五约四万一千；图七纵轴标为千参数，MixLinear 柱却接近百万分之一数量级，不能对应一百七十六个参数。短预测长度时，表一 MixLinear 的乘加数还高于 SparseTSF，因此“所有场景最低”不成立。给出的 A100 推理时间支持该机器上的速度优势，但未交代批大小、预热、同步及重复统计，也没有边缘设备测量；使用八十吉字节显存显卡不表示模型最低需要它，同样也不能替代实际边端部署证明。频谱复数算子、卷积聚合和上下采样的统一计数尤其应给脚本。\n\n整体判断是：想法值得作为极简基线候选，消融和附录展示了真实工作量；然而数学形状、参数口径、跨表数据和过强论断的累积，明显降低了现版证据的可审核性。优先级应是先核对官方实现与每张表的配置映射，再谈极限参数的新纪录。本文未执行仓库或模型，不把可修复问题擅自升级为造假，也不把尚未验证的实现当作已复现。", "review_chinese_character_count": 1678, "dimensions": [ { "dimension": "A1", "name": "问题重要性", "score": 80, "status": "rated", "confidence": "high", "reason": "极低资源预测有实际价值", "evidence_pages": "1–2", "judgment_range": null }, { "dimension": "A2", "name": "原创性", "score": 60, "status": "tentative", "confidence": "mid", "reason": "分段因子化与谱低秩组合，近邻已有成熟构件", "evidence_pages": "2–4,9", "judgment_range": [ 50, 70 ] }, { "dimension": "A3", "name": "实质推进程度", "score": 60, "status": "rated", "confidence": "mid", "reason": "参数折中值得考察，记录口径问题限制推进可信度", "evidence_pages": "5–8,18–19", "judgment_range": null }, { "dimension": "A4", "name": "知识与解释增量", "score": 45, "status": "rated", "confidence": "high", "reason": "经验消融可用，谱稀疏、正交和跨通道解释过度", "evidence_pages": "4–5,19–22", "judgment_range": null }, { "dimension": "B1", "name": "问题定义与评价有效性", "score": 70, "status": "rated", "confidence": "high", "reason": "任务MSE与计算指标明确，缺实际边缘约束及多误差指标", "evidence_pages": "2,5–7", "judgment_range": null }, { "dimension": "B2", "name": "技术与论证正确性", "score": 40, "status": "rated", "confidence": "high", "reason": "式六维度、伪码重排、频域参数计数不一致，已视觉核对", "evidence_pages": "4,14–15", "judgment_range": null }, { "dimension": "B3", "name": "数据与评测流程完整性", "score": 55, "status": "tentative", "confidence": "mid", "reason": "切分和归一化有说明，天气追加配置及末批等未齐", "evidence_pages": "15,17–18", "judgment_range": [ 45, 70 ] }, { "dimension": "B4", "name": "比较对象与比较公平性", "score": 50, "status": "rated", "confidence": "high", "reason": "有直接近邻，但引用/重跑来源、搜索及预算未统一", "evidence_pages": "5,15–18", "judgment_range": null }, { "dimension": "C1", "name": "核心主张的证据充分度", "score": 50, "status": "rated", "confidence": "high", "reason": "轻量竞争力有表支持，跨表精度与资源口径冲突", "evidence_pages": "6–8,18–19", "judgment_range": null }, { "dimension": "C2", "name": "机制归因与竞争解释区分度", "score": 55, "status": "rated", "confidence": "high", "reason": "双支路与超参消融真实存在，缺同参时域低秩和容量对照", "evidence_pages": "8,19–24", "judgment_range": null }, { "dimension": "C3", "name": "结果稳定性与不确定性处理", "score": 30, "status": "rated", "confidence": "high", "reason": "全文未定位多种子、误差条或时延重复统计", "evidence_pages": "5–8,15–24", "judgment_range": null }, { "dimension": "C4", "name": "适用范围与边界验证", "score": 55, "status": "rated", "confidence": "high", "reason": "常用多变量集及长度覆盖，缺缺失漂移和真实边缘验证", "evidence_pages": "5–9,17–24", "judgment_range": null }, { "dimension": "D1", "name": "过程透明度", "score": 40, "status": "rated", "confidence": "high", "reason": "伪码存在但形状、采样、参数与测试配置无法唯一还原", "evidence_pages": "4,14–16", "judgment_range": null }, { "dimension": "D2", "name": "研究资源可获得性", "score": 60, "status": "tentative", "confidence": "mid", "reason": "官方仓库链接与数据来源提供，未核验代码和环境", "evidence_pages": "11,16–17", "judgment_range": [ 45, 75 ] }, { "dimension": "D3", "name": "主张克制与报告一致性", "score": 30, "status": "rated", "confidence": "high", "reason": "指数改进、普遍最低、图表数量和跨通道机制表述不匹配", "evidence_pages": "5–9,18–23", "judgment_range": null }, { "dimension": "D4", "name": "表达清晰度", "score": 45, "status": "rated", "confidence": "high", "reason": "框架能理解，但符号复用、形状和夸大文字妨碍核对", "evidence_pages": "3–4,14–15", "judgment_range": null }, { "dimension": "E1", "name": "复杂度合理性", "score": 70, "status": "rated", "confidence": "mid", "reason": "极简组合有吸引力，必要性仍待等容量对照", "evidence_pages": "3–4,8", "judgment_range": null }, { "dimension": "E2", "name": "资源效率与成本收益", "score": 50, "status": "tentative", "confidence": "mid", "reason": "参数和A100时延提示收益，统计口径冲突且无边缘测量", "evidence_pages": "5–8,18–23", "judgment_range": [ 35, 65 ] }, { "dimension": "E3", "name": "可迁移与可复用程度", "score": 65, "status": "rated", "confidence": "mid", "reason": "共享通道结构和代码有复用潜力，文档错误加大成本", "evidence_pages": "3–4,11,14–17", "judgment_range": null }, { "dimension": "E4", "name": "后续研究启发性", "score": 65, "status": "rated", "confidence": "high", "reason": "提示极低参数预测与统一低秩对照研究", "evidence_pages": "8–9,19–24", "judgment_range": null }, { "dimension": "F1", "name": "与当前问题的相关性", "score": null, "status": "unknown", "confidence": "low", "reason": "当前研究问题未知", "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": "可用硬件及部署需求未知", "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": "当前决策目标未知", "evidence_pages": "用户背景未提供", "judgment_range": null } ], "missing_and_limits": [ "旧材料只有候选行，没有完整旧评。", "全部判断针对arXiv v2；未逐字比对ICLR终稿。", "未执行代码，矩阵维度和伪代码问题不能证明实现具有同一错误。", "未定位多种子、测试末批、时延同步及重复统计。", "公开代码链接 https://github.com/aitianma/MixLinear ，未验证当前提交。", "F组背景不足全部未知。" ], "experiments_executed": false, "aggregate_score": null }