[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"similar-valeman--Transformers_And_LLM_Are_What_You_Dont_Need":3,"tool-valeman--Transformers_And_LLM_Are_What_You_Dont_Need":65},[4,23,32,40,49,57],{"id":5,"name":6,"github_repo":7,"description_zh":8,"stars":9,"difficulty_score":10,"last_commit_at":11,"category_tags":12,"status":22},2268,"ML-For-Beginners","microsoft\u002FML-For-Beginners","ML-For-Beginners 是由微软推出的一套系统化机器学习入门课程，旨在帮助零基础用户轻松掌握经典机器学习知识。这套课程将学习路径规划为 12 周，包含 26 节精炼课程和 52 道配套测验，内容涵盖从基础概念到实际应用的完整流程，有效解决了初学者面对庞大知识体系时无从下手、缺乏结构化指导的痛点。\n\n无论是希望转型的开发者、需要补充算法背景的研究人员，还是对人工智能充满好奇的普通爱好者，都能从中受益。课程不仅提供了清晰的理论讲解，还强调动手实践，让用户在循序渐进中建立扎实的技能基础。其独特的亮点在于强大的多语言支持，通过自动化机制提供了包括简体中文在内的 50 多种语言版本，极大地降低了全球不同背景用户的学习门槛。此外，项目采用开源协作模式，社区活跃且内容持续更新，确保学习者能获取前沿且准确的技术资讯。如果你正寻找一条清晰、友好且专业的机器学习入门之路，ML-For-Beginners 将是理想的起点。",85267,2,"2026-04-18T11:00:28",[13,14,15,16,17,18,19,20,21],"图像","数据工具","视频","插件","Agent","其他","语言模型","开发框架","音频","ready",{"id":24,"name":25,"github_repo":26,"description_zh":27,"stars":28,"difficulty_score":29,"last_commit_at":30,"category_tags":31,"status":22},5784,"funNLP","fighting41love\u002FfunNLP","funNLP 是一个专为中文自然语言处理（NLP）打造的超级资源库，被誉为\"NLP 民工的乐园”。它并非单一的软件工具，而是一个汇集了海量开源项目、数据集、预训练模型和实用代码的综合性平台。\n\n面对中文 NLP 领域资源分散、入门门槛高以及特定场景数据匮乏的痛点，funNLP 提供了“一站式”解决方案。这里不仅涵盖了分词、命名实体识别、情感分析、文本摘要等基础任务的标准工具，还独特地收录了丰富的垂直领域资源，如法律、医疗、金融行业的专用词库与数据集，甚至包含古诗词生成、歌词创作等趣味应用。其核心亮点在于极高的全面性与实用性，从基础的字典词典到前沿的 BERT、GPT-2 模型代码，再到高质量的标注数据和竞赛方案，应有尽有。\n\n无论是刚刚踏入 NLP 领域的学生、需要快速验证想法的算法工程师，还是从事人工智能研究的学者，都能在这里找到急需的“武器弹药”。对于开发者而言，它能大幅减少寻找数据和复现模型的时间；对于研究者，它提供了丰富的基准测试资源和前沿技术参考。funNLP 以开放共享的精神，极大地降低了中文自然语言处理的开发与研究成本，是中文 AI 社区不可或缺的宝藏仓库。",79857,1,"2026-04-08T20:11:31",[19,14,18],{"id":33,"name":34,"github_repo":35,"description_zh":36,"stars":37,"difficulty_score":29,"last_commit_at":38,"category_tags":39,"status":22},5773,"cs-video-courses","Developer-Y\u002Fcs-video-courses","cs-video-courses 是一个精心整理的计算机科学视频课程清单，旨在为自学者提供系统化的学习路径。它汇集了全球知名高校（如加州大学伯克利分校、新南威尔士大学等）的完整课程录像，涵盖从编程基础、数据结构与算法，到操作系统、分布式系统、数据库等核心领域，并深入延伸至人工智能、机器学习、量子计算及区块链等前沿方向。\n\n面对网络上零散且质量参差不齐的教学资源，cs-video-courses 解决了学习者难以找到成体系、高难度大学级别课程的痛点。该项目严格筛选内容，仅收录真正的大学层级课程，排除了碎片化的简短教程或商业广告，确保用户能接触到严谨的学术内容。\n\n这份清单特别适合希望夯实计算机基础的开发者、需要补充特定领域知识的研究人员，以及渴望像在校生一样系统学习计算机科学的自学者。其独特的技术亮点在于分类极其详尽，不仅包含传统的软件工程与网络安全，还细分了生成式 AI、大语言模型、计算生物学等新兴学科，并直接链接至官方视频播放列表，让用户能一站式获取高质量的教育资源，免费享受世界顶尖大学的课堂体验。",79792,"2026-04-08T22:03:59",[18,13,14,20],{"id":41,"name":42,"github_repo":43,"description_zh":44,"stars":45,"difficulty_score":46,"last_commit_at":47,"category_tags":48,"status":22},3128,"ragflow","infiniflow\u002Fragflow","RAGFlow 是一款领先的开源检索增强生成（RAG）引擎，旨在为大语言模型构建更精准、可靠的上下文层。它巧妙地将前沿的 RAG 技术与智能体（Agent）能力相结合，不仅支持从各类文档中高效提取知识，还能让模型基于这些知识进行逻辑推理和任务执行。\n\n在大模型应用中，幻觉问题和知识滞后是常见痛点。RAGFlow 通过深度解析复杂文档结构（如表格、图表及混合排版），显著提升了信息检索的准确度，从而有效减少模型“胡编乱造”的现象，确保回答既有据可依又具备时效性。其内置的智能体机制更进一步，使系统不仅能回答问题，还能自主规划步骤解决复杂问题。\n\n这款工具特别适合开发者、企业技术团队以及 AI 研究人员使用。无论是希望快速搭建私有知识库问答系统，还是致力于探索大模型在垂直领域落地的创新者，都能从中受益。RAGFlow 提供了可视化的工作流编排界面和灵活的 API 接口，既降低了非算法背景用户的上手门槛，也满足了专业开发者对系统深度定制的需求。作为基于 Apache 2.0 协议开源的项目，它正成为连接通用大模型与行业专有知识之间的重要桥梁。",77062,3,"2026-04-04T04:44:48",[17,13,20,19,18],{"id":50,"name":51,"github_repo":52,"description_zh":53,"stars":54,"difficulty_score":46,"last_commit_at":55,"category_tags":56,"status":22},519,"PaddleOCR","PaddlePaddle\u002FPaddleOCR","PaddleOCR 是一款基于百度飞桨框架开发的高性能开源光学字符识别工具包。它的核心能力是将图片、PDF 等文档中的文字提取出来，转换成计算机可读取的结构化数据，让机器真正“看懂”图文内容。\n\n面对海量纸质或电子文档，PaddleOCR 解决了人工录入效率低、数字化成本高的问题。尤其在人工智能领域，它扮演着连接图像与大型语言模型（LLM）的桥梁角色，能将视觉信息直接转化为文本输入，助力智能问答、文档分析等应用场景落地。\n\nPaddleOCR 适合开发者、算法研究人员以及有文档自动化需求的普通用户。其技术优势十分明显：不仅支持全球 100 多种语言的识别，还能在 Windows、Linux、macOS 等多个系统上运行，并灵活适配 CPU、GPU、NPU 等各类硬件。作为一个轻量级且社区活跃的开源项目，PaddleOCR 既能满足快速集成的需求，也能支撑前沿的视觉语言研究，是处理文字识别任务的理想选择。",75872,"2026-04-18T10:54:57",[19,13,20,18],{"id":58,"name":59,"github_repo":60,"description_zh":61,"stars":62,"difficulty_score":29,"last_commit_at":63,"category_tags":64,"status":22},3215,"awesome-machine-learning","josephmisiti\u002Fawesome-machine-learning","awesome-machine-learning 是一份精心整理的机器学习资源清单，汇集了全球优秀的机器学习框架、库和软件工具。面对机器学习领域技术迭代快、资源分散且难以甄选的痛点，这份清单按编程语言（如 Python、C++、Go 等）和应用场景（如计算机视觉、自然语言处理、深度学习等）进行了系统化分类，帮助使用者快速定位高质量项目。\n\n它特别适合开发者、数据科学家及研究人员使用。无论是初学者寻找入门库，还是资深工程师对比不同语言的技术选型，都能从中获得极具价值的参考。此外，清单还延伸提供了免费书籍、在线课程、行业会议、技术博客及线下聚会等丰富资源，构建了从学习到实践的全链路支持体系。\n\n其独特亮点在于严格的维护标准：明确标记已停止维护或长期未更新的项目，确保推荐内容的时效性与可靠性。作为机器学习领域的“导航图”，awesome-machine-learning 以开源协作的方式持续更新，旨在降低技术探索门槛，让每一位从业者都能高效地站在巨人的肩膀上创新。",72149,"2026-04-03T21:50:24",[20,18],{"id":66,"github_repo":67,"name":68,"description_en":69,"description_zh":70,"ai_summary_zh":70,"readme_en":71,"readme_zh":72,"quickstart_zh":73,"use_case_zh":74,"hero_image_url":75,"owner_login":76,"owner_name":76,"owner_avatar_url":77,"owner_bio":78,"owner_company":79,"owner_location":80,"owner_email":81,"owner_twitter":82,"owner_website":83,"owner_url":84,"languages":81,"stars":85,"forks":86,"last_commit_at":87,"license":81,"difficulty_score":29,"env_os":88,"env_gpu":89,"env_ram":89,"env_deps":90,"category_tags":93,"github_topics":81,"view_count":10,"oss_zip_url":81,"oss_zip_packed_at":81,"status":22,"created_at":94,"updated_at":95,"faqs":96,"releases":97},9206,"valeman\u002FTransformers_And_LLM_Are_What_You_Dont_Need","Transformers_And_LLM_Are_What_You_Dont_Need","The best repository showing why transformers might not be the answer for time series forecasting and showcasing the best SOTA non transformer models.","Transformers_And_LLM_Are_What_You_Dont_Need 是一个专注于时间序列预测领域的开源资源库，旨在打破“大模型万能”的迷思。当前业界常盲目将 Transformer 和大语言模型（LLM）应用于各类预测任务，但该仓库通过汇集顶尖学术论文、技术文章及视频演讲，有力论证了在处理时间序列数据时，复杂的 Transformer 架构往往并非最佳选择，甚至表现不如更简单的模型。\n\n它主要解决了研究人员和开发者在选型时的困惑，揭示了过度依赖深度学习的弊端，并系统展示了目前最先进的非 Transformer 替代方案，如线性映射模型（Linear Mapping）、频域 MLP 以及 SCINet 等高效架构。这些方法通常在计算成本更低的情况下，能取得更精准的预测效果。\n\n该资源库特别适合从事时间序列分析的数据科学家、算法工程师以及相关领域的学术研究者使用。对于希望优化模型性能、降低算力消耗或深入理解预测本质的专业人士来说，这里提供的实证研究和代码链接是极具价值的参考指南，帮助大家回归问题本质，选择真正合适的工具而非盲目追逐热点。","# Transformers_And_LLM_Are_What_You_Dont_Need\nBy far the best and only repository showing why transformers don’t work in time series forecasting \n\n## Star History\n![GitHub Repo stars](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fvaleman\u002FTransformers_And_LLM_Are_What_You_Dont_Need?style=social)\n\n\n[![Star History Chart](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fvaleman_Transformers_And_LLM_Are_What_You_Dont_Need_readme_bd8d9d2cb455.png)](https:\u002F\u002Fwww.star-history.com\u002F#valeman\u002FTransformers_And_LLM_Are_What_You_Dont_Need&Date)\n\n\u003Ca href=\"https:\u002F\u002Fwww.buymeacoffee.com\u002Fvaleman\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fcdn.buymeacoffee.com\u002Fbuttons\u002Fdefault-orange.png\" alt=\"Buy Me A Coffee\" height=\"41\" width=\"174\">\u003C\u002Fa>\n\n\n\n## [Table of Contents]() \n\n* [PhD and MSc Theses](#theses)\n\n* [Videos](#videos) \n \n* [Papers](#papers)\n\n* [Articles](#articles)\n\n## Videos\n1. [Problems in the current research on forecasting with transformers, foundational models, etc.](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=vNul_AjRPFw&t=1084s) by Christof Bergmeir\n\n## Theses\n1. [Cotton Price Long-Term Time Series Forecasting: A look at Transformers Suitability](https:\u002F\u002Frepository.eafit.edu.co\u002Fserver\u002Fapi\u002Fcore\u002Fbitstreams\u002F27739bb8-7237-498f-8fba-ddb4ed6ca4fe\u002Fcontent)\n\n## Papers\n1. [Are Transformers Effective for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.13504) by Ailing Zeng, Muxi Chen, Lei Zhang, Qiang Xu (The Chinese University of Hong Kong, International Digital Economy Academy (IDEA), 2022) [code](https:\u002F\u002Fgithub.com\u002Fcure-lab\u002FLTSF-Linear) 🔥🔥🔥🔥🔥\n2.  [LLMs and foundational models for time series forecasting: They are not (yet) as good as you may hope](https:\u002F\u002Fwww.linkedin.com\u002Fpulse\u002Fllms-foundational-models-time-series-forecasting-yet-good-bergmeir-bprwf) by Christoph Bergmeir (2023) 🔥🔥🔥🔥🔥\n3.   [Transformers Are What You Do Not Need](https:\u002F\u002Fmedium.com\u002F@valeman\u002Ftransformers-are-what-you-do-not-need-cf16a4c13ab7) by Valeriy Manokhin (2023) 🔥🔥🔥🔥🔥\n4.   [Time Series Foundational Models: Their Role in Anomaly Detection and Prediction](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.19286v1) (2024) [code](https:\u002F\u002Fgithub.com\u002Fsmtmnfg\u002FTSFM)\n5.   [Deep Learning is What You Do Not Need](https:\u002F\u002Fmedium.com\u002F@valeman\u002F-86655805a676) by Valeriy Manokhin (2022) 🔥🔥🔥🔥🔥\n6. [Why do Transformers suck at Time Series Forecasting](https:\u002F\u002Fmachine-learning-made-simple.medium.com\u002Fwhy-do-transformers-suck-at-time-series-forecasting-46ae3a4d6b11) by Devansh (2023)\n7. [Frequency-domain MLPs are More Effective Learners in Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.06184) by Kun Yi, Qi Zhang, Wei Fan, Shoujin Wang, Pengyang Wang, Hui He, Defu Lian, Ning An, Longbing Cao, Zhendong Niu (Bejing Institute of Technology, Tongji University, University of Oxford, Universuty of Technology Sydney, University of Macau, HeFei University of Technology, Macquarie University) (2023) 🔥🔥🔥🔥🔥\n8. [Forecasting CPI inflation under economic policy and geo-political uncertainties](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.00249) by Shovon Sengupta, Tanujit Chakraborty, Sunny Kumar Singh (Fidelity Investments, Sorbonne University, BITS Pilani Hyderabad). (2024) 🔥🔥🔥🔥🔥\n9. [Revisiting Long-term Time Series Forecasting: An Investigation on Linear Mapping](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.10721) by Zhe Li, Shiyi Qi, Yiduo Li, Zenglin Xu (Harbin Institute of Technology, Shenzhen, 2023) [code](https:\u002F\u002Fgithub.com\u002Fplumprc\u002FRTSF)\n10. [SCINet: Time Series Modeling and Forecasting with Sample Convolution and Interaction](https:\u002F\u002Farxiv.org\u002Fabs\u002F2106.09305) by Minhao Liu, Ailing Zeng, Muxi Chen, Zhijian Xu, Qiuxia Lai, Lingna Ma, Qiang Xu (The Chinese University of Hong Kong,2022) [code](https:\u002F\u002Fgithub.com\u002Fcure-lab\u002FSCINet)\n11. [WINNET:TIME SERIES FORECASTING WITH A WINDOW-ENHANCED PERIOD EXTRACTING AND INTERACTING](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2311.00214.pdf) by Wenjie Ou, Dongyue Guo, Zheng Zhang, Zhishuo Zhao, Yi Lin (Sichuan University, China, 2023)\n12. [A Multi-Scale Decomposition MLP-Mixer for Time Series Analysis](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.11959) by Shuhan Zhong, Sizhe Song, Guanyao Li, Weipeng Zhuo, Yang Liu, S.-H. Gary Chan, The Hong Kong University of Science and Technology\nHong Kong, 2023) [code](https:\u002F\u002Fgithub.com\u002Fzshhans\u002FMSD-Mixer) 🔥🔥🔥🔥🔥\n15. [TimesNet: Temporal 2D-Variation Modeling for General Time Series Analysis](https:\u002F\u002Farxiv.org\u002Fabs\u002F2210.02186) by (Haixu Wu, Tengge Hu, Yong Liu, Hang Zhou, Jianmin Wang, Mingsheng Longj, , Tsinghua University, 2023) [code](https:\u002F\u002Fgithub.com\u002Fthuml\u002FTimesNet) 🔥🔥🔥🔥🔥\n16. [MTS-Mixers: Multivariate Time Series Forecasting via Factorized Temporal and Channel Mixing](https:\u002F\u002Farxiv.org\u002Fabs\u002F2302.04501) [code](https:\u002F\u002Fgithub.com\u002Fplumprc\u002FMTS-Mixers) 🔥🔥🔥🔥🔥\n17. [Reversible Instance Normalization for Accurate Time-Series Forecasting against Distribution Shift](https:\u002F\u002Fopenreview.net\u002Fforum?id=cGDAkQo1C0p) by Taesung Kim, Jinhee Kim, Yunwon Tae, Cheonbok Park, Jang-Ho Choi, Jaegul Choo (Kaist AI, Vuno, Naver Corp, ETRI, ICLR 2022) [code](https:\u002F\u002Fgithub.com\u002Fts-kim\u002FRevIN) [project page](https:\u002F\u002Fseharanul17.github.io\u002FRevIN\u002F) 🔥🔥🔥🔥🔥\n18. [WINNet: Wavelet-inspired Invertible Network for Image Denoising](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.00214) by \nWenjie Ou, Dongyue Guo, Zheng Zhang, Zhishuo Zhao, Yi Lin (College of Computer Science, Sichuan University, China) [code](https:\u002F\u002Fgithub.com\u002Fjjhuangcs\u002FWINNet) 🔥🔥🔥🔥🔥\n19. [Mlinear: Rethink the Linear Model for Time-series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.04800) Wei Li, Xiangxu Meng, Chuhao Chen and Jianing Chen (Harbin Engineering University, 2023) 🔥🔥🔥🔥🔥\n20. [Minimalist Traffic Prediction: Linear Layer Is All You Need](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.10276) by Wenying Duan, Hong Rao, Wei Huang, Xiaoxi He (Nanchang, University, Universify of Macau, 2023)\n21. [Frequency-domain MLPs are More Effective Learners in Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.06184) by Kun Yi, Qi Zhang, Wei Fan, Shoujin Wang, Pengyang Wang, Hui He, Defu Lian, Ning An, Longbing Cao, Zhendong Niu (Beijing Institute of Technology, Tongji University, University of Oxford University of Technology Sydney, University of Macau, USTC, HeFei University of Technology, Macquarie University, 2023) [code](https:\u002F\u002Fgithub.com\u002Faikunyi\u002FFreTS) 🔥🔥🔥🔥🔥\n22. [AN END-TO-END TIME SERIES MODEL FOR SIMULTANEOUS IMPUTATION AND FORECAST](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.00778) by Trang H. Tran, Lam M. Nguyen, Kyongmin Yeo, Nam Nguyen, Dzung Phan, Roman Vaculin Jayant Kalagnanam (School of Operations Research and Information Engineering, Cornell University; IBM Research, Thomas J. Watson Research Center, Yorktown Heights, NY, USA, 2023) 🔥🔥🔥🔥🔥\n23. [Long-term Forecasting with TiDE: Time-series Dense Encoder](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.08424) by Abhimanyu Das, Weihao Kong, Andrew Leach, Shaan Mathur, Rajat Sen, Rose Yu (Google Cloud, University of California, San Diego, 2023)\n24. [TSMixer: Lightweight MLP-Mixer Model for Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.09364) by Vijay Ekambaram, Arindam Jati, Nam Nguyen, Phanwadee Sinthong, Jayant Kalagnanam (IBM Research, 2023) [code](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Ftransformers\u002Fmain\u002Fen\u002Fmodel_doc\u002Fpatchtsmixer) [code](https:\u002F\u002Fgithub.com\u002FIBM\u002Ftsfm\u002Fblob\u002Fmain\u002Fwiki.md)\n25. [Koopa: Learning Non-stationary Time Series Dynamics with Koopman Predictors](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.18803) by Yong Liu, Chenyu Li, Jianmin Wang, Mingsheng Long (Tsinghua University, 2023) [code](https:\u002F\u002Fgithub.com\u002Fthuml\u002FTime-Series-Library\u002Fblob\u002Fmain\u002Fmodels\u002FKoopa.py) 🔥🔥🔥🔥🔥\n26. [Attractor Memory for Long-Term Time Series Forecasting: A Chaos Perspective](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.11463) Jiaxi Hu, Yuehong Hu, Wei Chen, Ming Jin, Shirui Pan, Qingsong Wen, Yuxuan Liang (2024) 🔥🔥🔥🔥🔥\n27. [When and How: Learning Identifiable Latent States for Nonstationary Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.12767) (2024) 🔥🔥🔥🔥🔥\n28. [Deep Coupling Network For Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15134) (2024)\n29. [Linear Dynamics-embedded Neural Network for Long-Sequence Modeling](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15290) by Tongyi Liang and Han-Xiong Li (City University of Hong Kong, 2024).\n30. [PDETime: Rethinking Long-Term Multivariate Time Series Forecasting from the perspective of partial differential equations](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.16913) (2024)\n31. [CATS: Enhancing Multivariate Time Series Forecasting by Constructing Auxiliary Time Series as Exogenous Variables](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.01673) (2024) 🔥🔥🔥🔥🔥\n32. [Is Mamba Effective for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.11144) [code](https:\u002F\u002Fgithub.com\u002Fwzhwzhwzh0921\u002FS-D-Mamba) (2024) 🔥🔥🔥🔥🔥\n33. [STG-Mamba: Spatial-Temporal Graph Learning via Selective State Space Model](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.12418) (2024)\n34. [TimeMachine: A Time Series is Worth 4 Mambas for Long-term Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.09898) [code](https:\u002F\u002Fgithub.com\u002FAtik-Ahamed\u002FTimeMachine) (2024)🔥🔥🔥🔥🔥\n35. [FITS: Modeling Time Series with 10k Parameters](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03756) [code](https:\u002F\u002Fgithub.com\u002FVEWOXIC\u002FFITS) (2023)\n36. [TSLANet: Rethinking Transformers for Time Series Representation Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.08472) [code](https:\u002F\u002Fgithub.com\u002Femadeldeen24\u002FTSLANet) (2024) 🔥🔥🔥🔥🔥\n37. [WFTNet: Exploiting Global and Local Periodicity in Long-term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.11319) [code](https:\u002F\u002Fgithub.com\u002FHank0626\u002FWFTNet) (2024) 🔥🔥🔥🔥🔥\n38. [SiMBA: Simplified Mamba-based Architecture for Vision and Multivariate Time series](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.15360) [code](https:\u002F\u002Fgithub.com\u002Fbadripatro\u002FSimba) (2024) 🔥🔥🔥🔥🔥\n39. [SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14197) [code](https:\u002F\u002Fgithub.com\u002FSecilia-Cxy\u002FSOFTS) (2024) 🔥🔥🔥🔥🔥\n40. [Integrating Mamba and Transformer for Long-Short Range Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14757) [code](https:\u002F\u002Fgithub.com\u002FXiongxiaoXu\u002FMambaformer-in-Time-Series) (2024) 🔥🔥🔥🔥🔥\n41. [SparseTSF: Modeling Long-term Time Series Forecasting with 1k Parameters](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.00946) (2024) 🔥🔥🔥🔥🔥\n42. [Boosting MLPs with a Coarsening Strategy for Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.03199) (2024) 🔥🔥🔥🔥🔥\n43. [Multi-Scale Dilated Convolution Network for Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.05499) (2024)\n44. [ModernTCN: A Modern Pure Convolution Structure for General Time Series Analysis](https:\u002F\u002Fopenreview.net\u002Fforum?id=vpJMJerXHU#) [code](https:\u002F\u002Fgithub.com\u002Fluodhhh\u002FModernTCN) (ICLR 2024 Spotlight)\n45. [Adaptive Extraction Network for Multivariate Long Sequence Time-Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.12038) (2024) 🔥🔥🔥🔥🔥\n46. [Interpretable Multivariate Time Series Forecasting Using Neural Fourier Transform](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.13812) (2024) 🔥🔥🔥🔥🔥\n47. [PERIODICITY DECOUPLING FRAMEWORK FOR LONG- TERM SERIES FORECASTING](https:\u002F\u002Fopenreview.net\u002Fpdf?id=dp27P5HBBt) [code](https:\u002F\u002Fgithub.com\u002FHank0626\u002FPDF)  (2024) 🔥🔥🔥🔥🔥\n48. [Chimera: Effectively Modeling Multivariate Time Series with 2-Dimensional State Space Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.04320) 🔥🔥🔥🔥🔥 (2024)\n49. [Time Evidence Fusion Network: Multi-source View in Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.06419) [code](https:\u002F\u002Fgithub.com\u002Fztxtech\u002FTime-Evidence-Fusion-Network) (2024)\n50. [ATFNet: Adaptive Time-Frequency Ensembled Network for Long-term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.05192) [code](https:\u002F\u002Fgithub.com\u002Fyhyhyhyhyhy\u002Fatfnet) (2024) 🔥🔥🔥🔥\n51. [C-Mamba: Channel Correlation Enhanced State Space Models for Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.05316) (2024) 🔥🔥🔥🔥\n52. [The Power of Minimalism in Long Sequence Time-series Forecasting](https:\u002F\u002Fopenreview.net\u002Fpdf?id=hF8jnnexSB)\n53. [WindowMixer: Intra-Window and Inter-Window Modeling for Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.12921)\n54. [xLSTMTime : Long-term Time Series Forecasting With xLSTM](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.10240) [code](https:\u002F\u002Fgithub.com\u002Fmuslehal\u002FxLSTMTime) (2024)\n55. [Not All Frequencies Are Created Equal:Towards a Dynamic Fusion of Frequencies in Time-Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.12415) (2024) 🔥🔥🔥🔥\n56. [FMamba: Mamba based on Fast-attention for Multivariate Time-series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.14814) (2024)\n57. [Long Input Sequence Network for Long Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.15869) (2024)\n58. [Time-series Forecasting with Tri-Decomposition Linear-based Modelling and Series-wise Metrics](https:\u002F\u002Fpapers.ssrn.com\u002Fsol3\u002Fpapers.cfm?abstract_id=4913290) (2024) 🔥🔥🔥🔥\n59. [An Evaluation of Standard Statistical Models and LLMs on Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.04867) (2024) LLM 🔥🔥🔥🔥\n60. [Macroeconomic Forecasting with Large Language Models](https:\u002F\u002Fpapers.ssrn.com\u002Fsol3\u002Fpapers.cfm?abstract_id=4881094) (2024) LLM 🔥🔥🔥🔥\n61. [Language Models Still Struggle to Zero-shot Reason about Time Series](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.11757) (2024) LLM 🔥🔥🔥🔥\n62. [KAN4TSF: Are KAN and KAN-based models Effective for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.11306) (2024) [code](https:\u002F\u002Fgithub.com\u002F2448845600\u002FEasyTSF)\n63. [Simplified Mamba with Disentangled Dependency Encoding for Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12068) (2024)\n64. [Transformers are Expressive, But Are They Expressive Enough for Regression?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15478) (2024) paper showing transformers cant approximate smooth functions\n65. [MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with 0.1K Parameters](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.02081)\n66. [MMFNet: Multi-Scale Frequency Masking Neural Network for Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.02070)\n67. [Neural Fourier Modelling: A Highly Compact Approach to Time-Series Analysis](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.04703) [code](https:\u002F\u002Fgithub.com\u002Fminkiml\u002FNFM)\n68. [CMMamba: channel mixing Mamba for time series forecasting](https:\u002F\u002Fjournalofbigdata.springeropen.com\u002Farticles\u002F10.1186\u002Fs40537-024-01001-9)\n69. [EffiCANet: Efficient Time Series Forecasting with Convolutional Attention](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.04669)\n70. [Curse of Attention: A Kernel-Based Perspective for Why Transformers Fail to Generalize on Time Series Forecasting and Beyond](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.06061)\n71. [CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet) [code](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet)\n72. [Are Language Models Actually Useful for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.16964)\n73. [SOFTS: Efficient Multivariate Time Series Forecasting with Series-Core Fusion](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14197) [code](https:\u002F\u002Fgithub.com\u002FSecilia-Cxy\u002FSOFTS)\n74. [FTLinear: MLP based on Fourier Transform for Multivariate Time-series Forecasting](https:\u002F\u002Fwww.researchsquare.com\u002Farticle\u002Frs-5654336\u002Fv1)\n75. [WPMixer: Efficient Multi-Resolution Mixing for Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.17176) [code](https:\u002F\u002Fgithub.com\u002FSecure-and-Intelligent-Systems-Lab\u002FWPMixer)\n76. [Zero Shot Time Series Forecasting Using Kolmogorov Arnold Networks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.17853)\n77. [BEAT: Balanced Frequency Adaptive Tuning for Long-Term Time-Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.19065) (2025) 🔥🔥🔥🔥🔥\n78. [A Multi-Task Learning Approach to Linear Multivariate Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.03571) (2025)\n79. [Benchmarking Time Series Forecasting Models: From Statistical Techniques to Foundation Models in Real-World Applications](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.03395) (2025)\n80. [Day-ahead demand response potential prediction in residential buildings with HITSKAN: A fusion of Kolmogorov-Arnold networks and N-HiTS](https:\u002F\u002Fwww.sciencedirect.com\u002Fscience\u002Farticle\u002Fpii\u002FS0378778825001859) (2025)\n81. [Do We Really Need Deep Learning Models for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2101.02118) (2021)\n82. [Two Steps Forward and One Behind: Rethinking Time Series Forecasting with Deep Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.04553) (2023)\n83. [Are Self-Attentions Effective for Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.16877?utm_source=chatgpt.com) (2024)\n84. [What Matters in Transformers? Not All Attention is Needed](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.15786) (2024)\n85. [Time Series Foundational Models: Their Role in Anomaly Detection and Prediction](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.19286v1) (2024)\n86. [Performance of Zero-Shot Time Series Foundation Models on Cloud Data](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12944) (2025) 🔥🔥🔥🔥🔥\n87. [Position: There are no Champions in Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12161) (2025)\n88. [FinTSB: A Comprehensive and Practical Benchmark for Financial Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.18834)\n89. [Cherry-Picking in Time Series Forecasting: How to Select Datasets to Make Your Model Shine](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14435)\n90. [TFB: Towards Comprehensive and Fair Benchmarking of Time Series Forecasting Methods](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.20150) 🔥🔥🔥🔥🔥\n91. [DUET: Dual Clustering Enhanced Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.10859)\n92. [CycleNet: Enhancing Time Series Forecasting through Modeling Periodic Patterns](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.18479) [code](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet)\n93. [Can LLMs Understand Time Series Anomalies?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05440) [code](https:\u002F\u002Fgithub.com\u002Frose-stl-lab\u002Fanomllm)\n94. [EMForecaster: A Deep Learning Framework for Time Series Forecasting in Wireless Networks with Distribution-Free Uncertainty Quantification](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.00120)\n95. [Times2D: Multi-Period Decomposition and Derivative Mapping for General Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.00118) [code](https:\u002F\u002Fgithub.com\u002FTims2D\u002FTimes2D)\n96. [FilterTS: Comprehensive Frequency Filtering for Multivariate Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.04158)\n97. [OLinear: A Linear Model for Time Series Forecasting in Orthogonally Transformed Domain](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08550)\n98. [Does Scaling Law Apply in Time Series Forecasting?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.10172)\n99. [TiRex: Zero-Shot Forecasting Across Long and Short Horizons with Enhanced In-Context Learning](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.23719) [code](https:\u002F\u002Fgithub.com\u002FNX-AI\u002Ftirex) [article](https:\u002F\u002Fhuggingface.co\u002FNX-AI\u002FTiRex) [podcast](https:\u002F\u002Fopen.spotify.com\u002Fepisode\u002F5PmGnhjPf5JMI1SdVqwOu4) 🔥🔥🔥🔥🔥\n100. [FreDF: Learning to Forecast in the Frequency Domain](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.02399) [code](https:\u002F\u002Fgithub.com\u002FMaster-PLC\u002FFreDF)\n101. [KARMA: A Multilevel Decomposition Hybrid Mamba Framework for Multivariate Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.08939) [code](https:\u002F\u002Fgithub.com\u002Fyedadasd\u002FKARMA)\n102. [MS-TVNet:A Long-Term Time Series Prediction Method Based on Multi-Scale Dynamic Convolution](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.17253)\n103. [Variational Mode Decomposition and Linear Embeddings are What You Need For Time-Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.16122) [code](https:\u002F\u002Fgithub.com\u002FEspalemit\u002FVMD-With-LTSF-Linear) 🔥🔥🔥🔥🔥\n104. [TimeSieve: Extracting Temporal Dynamics through Information Bottlenecks](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.05036) [code](https:\u002F\u002Fgithub.com\u002Fxll0328\u002FTimeSieve)\n105. [VisionTS++: Cross-Modal Time Series Foundation Model with Continual Pre-trained Visual Backbones](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.04379) [code](https:\u002F\u002Fgithub.com\u002FHALF111\u002FVisionTSpp) VisionTS 🔥🔥🔥🔥🔥\n106. [VISIONTS: Visual Masked Autoencoders Are Free-Lunch Zero-Shot Time Series Forecasters](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.17253) [code](https:\u002F\u002Fgithub.com\u002FKeytoyze\u002FVisionTS) VisionTS 🔥🔥🔥🔥🔥\n107. [Wavelet Mixture of Experts for Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.08825)\n108. [TFKAN: Time-Frequency KAN for Long-Term Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.12696)\n109. [Why Do Transformers Fail to Forecast Time Series In-Context?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.09776) 🔥🔥🔥🔥🔥 [code](https:\u002F\u002Fgithub.com\u002FMasterZhou1\u002FICL-Time-Series)\n110. [In-Context Learning of Linear Dynamical Systems with Transformers: Error Bounds and Depth-Separation](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.08136)\n111. [Time Series Foundation Models: Benchmarking Challenges and Requirements](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.13654) 🔥🔥🔥🔥🔥\n112. [DoFlow: Causal Generative Flows for Interventional and Counterfactual Time-Series Prediction](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.02137) 🔥🔥🔥🔥🔥\n113. [Hydra: Dual Exponentiated Memory for Multivariate Time Series Analysis](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.00989) 🔥🔥🔥🔥🔥\n114. [Mixture-of-KAN for Multivariate Time Series Forecasting](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fabs\u002F10.1145\u002F3746252.3760836) 🔥🔥🔥🔥🔥\n115. [A Realistic Evaluation of Cross-Frequency Transfer Learning and Foundation Forecasting Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.19465) paper showing foundational models systematically underpeform ARIMA and simple stats ensemble 🔥🔥🔥🔥🔥\n116. [XLinear: A Lightweight and Accurate MLP-Based Model for Long-Term Time Series Forecasting with Exogenous Inputs](https:\u002F\u002Farxiv.org\u002Fabs\u002F2601.09237)\n117. [Position: The Inevitable End of One-Architecture-Fits-All-Domains in Time Series Forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.01736)\n118. [Assessing Electricity Demand Forecasting with Exogenous Data in Time Series Foundation Models](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.05390)\n\n\n\n\n\n\n\n\n    \n## Articles\n1. [TimeGPT vs TiDE: Is Zero-Shot Inference the Future of Forecasting or Just Hype?](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.13504) (https:\u002F\u002Ftowardsdatascience.com\u002Ftimegpt-vs-tide-is-zero-shot-inference-the-future-of-forecasting-or-just-hype-9063bdbe0b76) by Luís Roque\nand Rafael Guedes. (2024)🔥🔥🔥🔥🔥\n2. [TimeGPT-1, discussion on Hacker News](https:\u002F\u002Fnews.ycombinator.com\u002Fitem?id=37874891) (2023) \n3. [TimeGPT : The first Generative Pretrained Transformer for Time-Series Forecasting](https:\u002F\u002Fwww.reddit.com\u002Fr\u002FMachineLearning\u002Fcomments\u002F176wsne\u002Fr_timegpt_the_first_generative_pretrained\u002F)\n\n\n","# 变压器和大语言模型是你不需要的东西\n迄今为止，展示为什么变压器在时间序列预测中不起作用的最佳且唯一的仓库\n\n## 点赞历史\n![GitHub 仓库点赞数](https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fvaleman\u002FTransformers_And_LLM_Are_What_You_Dont_Need?style=social)\n\n\n[![点赞历史图表](https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fvaleman_Transformers_And_LLM_Are_What_You_Dont_Need_readme_bd8d9d2cb455.png)](https:\u002F\u002Fwww.star-history.com\u002F#valeman\u002FTransformers_And_LLM_Are_What_You_Dont_Need&Date)\n\n\u003Ca href=\"https:\u002F\u002Fwww.buymeacoffee.com\u002Fvaleman\" target=\"_blank\">\u003Cimg src=\"https:\u002F\u002Fcdn.buymeacoffee.com\u002Fbuttons\u002Fdefault-orange.png\" alt=\"请我喝杯咖啡\" height=\"41\" width=\"174\">\u003C\u002Fa>\n\n\n\n## [目录]() \n\n* [博士和硕士论文](#theses)\n\n* [视频](#videos) \n \n* [论文](#papers)\n\n* [文章](#articles)\n\n## 视频\n1. [当前使用变压器、基础模型等进行预测研究中的问题](https:\u002F\u002Fwww.youtube.com\u002Fwatch?v=vNul_AjRPFw&t=1084s) 由 Christof Bergmeir 主讲\n\n## 论文\n1. [棉花价格长期时间序列预测：探讨变压器的适用性](https:\u002F\u002Frepository.eafit.edu.co\u002Fserver\u002Fapi\u002Fcore\u002Fbitstreams\u002F27739bb8-7237-498f-8fba-ddb4ed6ca4fe\u002Fcontent)\n\n## 论文\n1. [Transformer 对时间序列预测有效吗？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.13504) 作者：Ailing Zeng、Muxi Chen、Lei Zhang、Qiang Xu（香港中文大学、国际数字经济研究院 (IDEA)，2022 年）[代码](https:\u002F\u002Fgithub.com\u002Fcure-lab\u002FLTSF-Linear) 🔥🔥🔥🔥🔥\n2. [大语言模型与基础模型在时间序列预测中的应用：它们并没有你想象的那么好](https:\u002F\u002Fwww.linkedin.com\u002Fpulse\u002Fllms-foundational-models-time-series-forecasting-yet-good-bergmeir-bprwf) 作者：Christoph Bergmeir（2023 年）🔥🔥🔥🔥🔥\n3. [Transformer 并不是你需要的东西](https:\u002F\u002Fmedium.com\u002F@valeman\u002Ftransformers-are-what-you-do-not-need-cf16a4c13ab7) 作者：Valeriy Manokhin（2023 年）🔥🔥🔥🔥🔥\n4. [时间序列基础模型：它们在异常检测和预测中的作用](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.19286v1)（2024 年）[代码](https:\u002F\u002Fgithub.com\u002Fsmtmnfg\u002FTSFM)\n5. [深度学习并不是你需要的东西](https:\u002F\u002Fmedium.com\u002F@valeman\u002F-86655805a676) 作者：Valeriy Manokhin（2022 年）🔥🔥🔥🔥🔥\n6. [为什么 Transformer 在时间序列预测中表现不佳](https:\u002F\u002Fmachine-learning-made-simple.medium.com\u002Fwhy-do-transformers-suck-at-time-series-forecasting-46ae3a4d6b11) 作者：Devansh（2023 年）\n7. [频域 MLP 在时间序列预测中是更有效的学习者](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.06184) 作者：Kun Yi、Qi Zhang、Wei Fan、Shoujin Wang、Pengyang Wang、Hui He、Defu Lian、Ning An、Longbing Cao、Zhendong Niu（北京理工大学、同济大学、牛津大学、悉尼科技大学、澳门大学、合肥工业大学、麦考瑞大学）（2023 年）🔥🔥🔥🔥🔥\n8. [在经济政策和地缘政治不确定性下的 CPI 通胀预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2401.00249) 作者：Shovon Sengupta、Tanujit Chakraborty、Sunny Kumar Singh（富达投资、索邦大学、BITS Pilani 海得拉巴校区）（2024 年）🔥🔥🔥🔥🔥\n9. [再探长期时间序列预测：线性映射的探究](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.10721) 作者：Zhe Li、Shiyi Qi、Yiduo Li、Zenglin Xu（哈尔滨工业大学深圳校区，2023 年）[代码](https:\u002F\u002Fgithub.com\u002Fplumprc\u002FRTSF)\n10. [SCINet：基于样本卷积与交互的时间序列建模与预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2106.09305) 作者：Minhao Liu、Ailing Zeng、Muxi Chen、Zhijian Xu、Qiuxia Lai、Lingna Ma、Qiang Xu（香港中文大学，2022 年）[代码](https:\u002F\u002Fgithub.com\u002Fcure-lab\u002FSCINet)\n11. [WINNET：基于窗口增强的周期提取与交互的时间序列预测](https:\u002F\u002Farxiv.org\u002Fpdf\u002F2311.00214.pdf) 作者：Wenjie Ou、Dongyue Guo、Zheng Zhang、Zhishuo Zhao、Yi Lin（四川大学，中国，2023 年）\n12. [用于时间序列分析的多尺度分解 MLP-Mixer](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.11959) 作者：Shuhan Zhong、Sizhe Song、Guanyao Li、Weipeng Zhuo、Yang Liu、S.-H. Gary Chan、香港科技大学（香港，2023 年）[代码](https:\u002F\u002Fgithub.com\u002Fzshhans\u002FMSD-Mixer) 🔥🔥🔥🔥🔥\n15. [TimesNet：面向通用时间序列分析的时序二维变体建模](https:\u002F\u002Farxiv.org\u002Fabs\u002F2210.02186) 作者：（Haixu Wu、Tengge Hu、Yong Liu、Hang Zhou、Jianmin Wang、Mingsheng Longj，清华大学，2023 年）[代码](https:\u002F\u002Fgithub.com\u002Fthuml\u002FTimesNet) 🔥🔥🔥🔥🔥\n16. [MTS-Mixers：通过因子化时空混合进行多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2302.04501) [代码](https:\u002F\u002Fgithub.com\u002Fplumprc\u002FMTS-Mixers) 🔥🔥🔥🔥🔥\n17. [可逆实例归一化：应对分布漂移的精准时间序列预测](https:\u002F\u002Fopenreview.net\u002Fforum?id=cGDAkQo1C0p) 作者：Taesung Kim、Jinhee Kim、Yunwon Tae、Cheonbok Park、Jang-Ho Choi、Jaegul Choo（KAIST AI、Vuno、Naver 公司、ETRI、ICLR 2022）[代码](https:\u002F\u002Fgithub.com\u002Fts-kim\u002FRevIN) [项目页面](https:\u002F\u002Fseharanul17.github.io\u002FRevIN\u002F) 🔥🔥🔥🔥🔥\n18. [WINNet：受小波启发的可逆网络用于图像去噪](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.00214) 作者：Wenjie Ou、Dongyue Guo、Zheng Zhang、Zhishuo Zhao、Yi Lin（四川大学计算机学院，中国）[代码](https:\u002F\u002Fgithub.com\u002Fjjhuangcs\u002FWINNet) 🔥🔥🔥🔥🔥\n19. [Mlinear：重新思考时间序列预测中的线性模型](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.04800) 作者：Wei Li、Xiangxu Meng、Chuhao Chen 和 Jianing Chen（哈尔滨工程大学，2023 年）🔥🔥🔥🔥🔥\n20. [极简交通流量预测：线性层就是你需要的一切](https:\u002F\u002Farxiv.org\u002Fabs\u002F2308.10276) 作者：Wenying Duan、Hong Rao、Wei Huang、Xiaoxi He（南昌大学、澳门大学，2023 年）\n21. [频域 MLP 在时间序列预测中是更有效的学习者](https:\u002F\u002Farxiv.org\u002Fabs\u002F2311.06184) 作者：Kun Yi、Qi Zhang、Wei Fan、Shoujin Wang、Pengyang Wang、Hui He、Defu Lian、Ning An、Longbing Cao、Zhendong Niu（北京理工大学、同济大学、牛津大学、悉尼科技大学、澳门大学、中国科学技术大学、合肥工业大学、麦考瑞大学，2023 年）[代码](https:\u002F\u002Fgithub.com\u002Faikunyi\u002FFreTS) 🔥🔥🔥🔥🔥\n22. [端到端时间序列模型：同时进行插补与预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.00778) 作者：Trang H. Tran、Lam M. Nguyen、Kyongmin Yeo、Nam Nguyen、Dzung Phan、Roman Vaculin Jayant Kalagnanam（康奈尔大学运筹学与信息工程学院；IBM 研究所，托马斯·J·沃森研究中心，纽约州约克镇高地，美国，2023 年）🔥🔥🔥🔥🔥\n23. [使用 TiDE：时间序列密集编码器进行长期预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.08424) 作者：Abhimanyu Das、Weihao Kong、Andrew Leach、Shaan Mathur、Rajat Sen、Rose Yu（谷歌云、加州大学圣地亚哥分校，2023 年）\n24. [TSMixer：轻量级 MLP-Mixer 模型用于多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2306.09364) 作者：Vijay Ekambaram、Arindam Jati、Nam Nguyen、Phanwadee Sinthong、Jayant Kalagnanam（IBM 研究所，2023 年）[代码](https:\u002F\u002Fhuggingface.co\u002Fdocs\u002Ftransformers\u002Fmain\u002Fen\u002Fmodel_doc\u002Fpatchtsmixer) [代码](https:\u002F\u002Fgithub.com\u002FIBM\u002Ftsfm\u002Fblob\u002Fmain\u002Fwiki.md)\n25. [Koopa：利用库普曼预测器学习非平稳时间序列动态](https:\u002F\u002Farxiv.org\u002Fabs\u002F2305.18803) 作者：Yong Liu、Chenyu Li、Jianmin Wang、Mingsheng Long（清华大学，2023 年）[代码](https:\u002F\u002Fgithub.com\u002Fthuml\u002FTime-Series-Library\u002Fblob\u002Fmain\u002Fmodels\u002FKoopa.py) 🔥🔥🔥🔥🔥\n26. [吸引子记忆用于长期时间序列预测：混沌视角](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.11463) 作者：Jiaxi Hu、Yuehong Hu、Wei Chen、Ming Jin、Shirui Pan、Qingsong Wen、Yuxuan Liang（2024 年）🔥🔥🔥🔥🔥\n27. [何时以及如何：学习可识别的潜在状态以进行非平稳时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.12767)（2024 年）🔥🔥🔥🔥🔥\n28. [深度耦合网络用于多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15134)（2024 年）\n29. [嵌入线性动力学的神经网络用于长序列建模](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15290) 作者：Tongyi Liang 和 Han-Xiong Li（香港城市大学，2024 年）。\n30. [PDETime：从偏微分方程的角度重新思考长期多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.16913)（2024 年）\n31. [CATS：通过构建辅助时间序列作为外生变量来增强多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.01673)（2024 年）🔥🔥🔥🔥🔥\n32. [Mamba 对时间序列预测有效吗？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.11144) [代码](https:\u002F\u002Fgithub.com\u002Fwzhwzhwzh0921\u002FS-D-Mamba)（2024 年）🔥🔥🔥🔥🔥\n33. [STG-Mamba：通过选择性状态空间模型进行时空图学习](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.12418)（2024 年）\n34. [TimeMachine：对于长期预测而言，一个时间序列的价值相当于四个 Mamba](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.09898) [代码](https:\u002F\u002Fgithub.com\u002FAtik-Ahamed\u002FTimeMachine)（2024 年）🔥🔥🔥🔥🔥\n35. [FITS：用 1 万个参数建模时间序列](https:\u002F\u002Farxiv.org\u002Fabs\u002F2307.03756) [代码](https:\u002F\u002Fgithub.com\u002FVEWOXIC\u002FFITS)（2023 年）\n36. [TSLANet：重新思考 Transformer 在时间序列表征学习中的作用](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.08472) [代码](https:\u002F\u002Fgithub.com\u002Femadeldeen24\u002FTSLANet)（2024 年）🔥🔥🔥🔥🔥\n37. [WFTNet：在长期时间序列预测中利用全局和局部周期性](https:\u002F\u002Farxiv.org\u002Fabs\u002F2309.11319) [代码](https:\u002F\u002Fgithub.com\u002FHank0626\u002FWFTNet)（2024 年）🔥🔥🔥🔥🔥\n38. [SiMBA：基于 Mamba 的简化架构，适用于视觉和多变量时间序列](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.15360) [代码](https:\u002F\u002Fgithub.com\u002Fbadripatro\u002FSimba)（2024 年）🔥🔥🔥🔥🔥\n39. [SOFTS：通过系列核心融合实现高效的多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14197) [代码](https:\u002F\u002Fgithub.com\u002FSecilia-Cxy\u002FSOFTS)（2024 年）🔥🔥🔥🔥🔥\n40. [将 Mamba 和 Transformer 集成用于长短时距时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14757) [代码](https:\u002F\u002Fgithub.com\u002FXiongxiaoXu\u002FMambaformer-in-Time-Series)（2024 年）🔥🔥🔥🔥🔥\n41. [SparseTSF：用 1 千个参数建模长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.00946)（2024 年）🔥🔥🔥🔥🔥\n42. [通过粗粒化策略提升 MLP 在长期时间序列预测中的表现](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.03199)（2024 年）🔥🔥🔥🔥🔥\n43. [多尺度扩张卷积网络用于长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.05499)（2024 年）\n44. [ModernTCN：一种现代纯卷积结构，用于通用时间序列分析](https:\u002F\u002Fopenreview.net\u002Fforum?id=vpJMJerXHU#) [代码](https:\u002F\u002Fgithub.com\u002Fluodhhh\u002FModernTCN)（ICLR 2024 Spotlight）\n45. [自适应提取网络用于多变量长序列时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.12038)（2024 年）🔥🔥🔥🔥🔥\n46. [使用神经傅里叶变换进行可解释的多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.13812)（2024 年）🔥🔥🔥🔥🔥\n47. [长期序列预测的周期解耦框架](https:\u002F\u002Fopenreview.net\u002Fpdf?id=dp27P5HBBt) [代码](https:\u002F\u002Fgithub.com\u002FHank0626\u002FPDF)（2024 年）🔥🔥🔥🔥🔥\n48. [Chimera：用二维状态空间模型有效建模多变量时间序列](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.04320) 🔥🔥🔥🔥🔥（2024 年）\n49. [时间证据融合网络：长期时间序列预测中的多源视角](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.06419) [代码](https:\u002F\u002Fgithub.com\u002Fztxtech\u002FTime-Evidence-Fusion-Network)（2024 年）\n50. [ATFNet：自适应时频集成网络用于长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.05192) [代码](https:\u002F\u002Fgithub.com\u002Fyhyhyhyhyhy\u002Fatfnet)（2024 年）🔥🔥🔥🔥\n51. [C-Mamba：通道相关性增强的状态空间模型用于多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.05316)（2024 年）🔥🔥🔥🔥\n52. [极简主义在长序列时间序列预测中的力量](https:\u002F\u002Fopenreview.net\u002Fpdf?id=hF8jnnexSB)\n53. [WindowMixer：时间序列预测中的窗内与窗间建模](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.12921)\n54. [xLSTMTime：用 xLSTM 进行长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.10240) [代码](https:\u002F\u002Fgithub.com\u002Fmuslehal\u002FxLSTMTime)（2024 年）\n55. [并非所有频率都是一样的：迈向时间序列预测中频率的动态融合](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.12415)（2024 年）🔥🔥🔥🔥\n56. [FMamba：基于快速注意力的 Mamba 用于多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.14814)（2024 年）\n57. [长输入序列网络用于长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2407.15869)（2024 年）\n58. [基于三重分解线性建模和序列指标进行时间序列预测](https:\u002F\u002Fpapers.ssrn.com\u002Fsol3\u002Fpapers.cfm?abstract_id=4913290)（2024 年）🔥🔥🔥🔥\n59. [标准统计模型和大语言模型在时间序列预测中的评估](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.04867)（2024 年）LLM 🔥🔥🔥🔥\n60. [宏观经济预测与大型语言模型](https:\u002F\u002Fpapers.ssrn.com\u002Fsol3\u002Fpapers.cfm?abstract_id=4881094)（2024 年）LLM 🔥🔥🔥🔥\n61. [语言模型在零样本推理时间序列方面仍然存在困难](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.11757)（2024 年）LLM 🔥🔥🔥🔥\n62. [KAN4TSF：KAN 及其基于 KAN 的模型是否对时间序列预测有效？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.11306)（2024 年）[代码](https:\u002F\u002Fgithub.com\u002F2448845600\u002FEasyTSF)\n63. [简化版 Mamba，采用解耦依赖编码进行长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.12068)（2024 年）\n64. [Transformer 表现力强，但它们是否足以进行回归任务？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.15478)（2024 年）一篇论文表明 Transformer 无法近似光滑函数\n65. [MixLinear：超低资源多变量时间序列预测，仅需 0.1 千个参数](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.02081)\n66. [MMFNet：多尺度频率掩码神经网络用于多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.02070)\n67. [神经傅里叶建模：一种高度紧凑的时间序列分析方法](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.04703) [代码](https:\u002F\u002Fgithub.com\u002Fminkiml\u002FNFM)\n68. [CMMamba：用于时间序列预测的通道混合 Mamba](https:\u002F\u002Fjournalofbigdata.springeropen.com\u002Farticles\u002F10.1186\u002Fs40537-024-01001-9)\n69. [EffiCANet：利用卷积注意力进行高效时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2411.04669)\n70. [注意力的诅咒：从核函数视角看为何 Transformer 在时间序列预测及其他领域中无法泛化](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.06061)\n71. [CycleNet：通过建模周期性模式来增强时间序列预测](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet) [代码](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet)\n72. [语言模型真的对时间序列预测有用吗？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.16964)\n73. [SOFTS：通过系列核心融合实现高效的多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2404.14197) [代码](https:\u002F\u002Fgithub.com\u002FSecilia-Cxy\u002FSOFTS)\n74. [FTLinear：基于傅里叶变换的 MLP 用于多变量时间序列预测](https:\u002F\u002Fwww.researchsquare.com\u002Farticle\u002Frs-5654336\u002Fv1)\n75. [WPMixer：用于长期时间序列预测的高效多分辨率混合](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.17176) [代码](https:\u002F\u002Fgithub.com\u002FSecure-and-Intelligent-Systems-Lab\u002FWPMixer)\n76. [使用科尔莫戈洛夫-阿诺德网络进行零样本时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.17853)\n77. [BEAT：平衡频率自适应调谐用于长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2501.19065)（2025 年）🔥🔥🔥🔥🔥\n78. [多任务学习方法用于线性多变量预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.03571)（2025 年）\n79. [时间序列预测模型基准测试：从统计技术到基础模型在真实世界应用中的比较](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.03395)（2025 年）\n80. [使用 HITSKAN 在住宅建筑中预测次日需求响应潜力：科尔莫戈洛夫-阿诺德网络与 N-HiTS 的融合](https:\u002F\u002Fwww.sciencedirect.com\u002Fscience\u002Farticle\u002Fpii\u002FS0378778825001859)（2025 年）\n81. [我们真的需要深度学习模型来进行时间序列预测吗？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2101.02118)（2021 年）\n82. [前进两步，后退一步：重新思考使用深度学习进行时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2304.04553)（2023 年）\n83. [自注意力机制对时间序列预测有效吗？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2405.16877?utm_source=chatgpt.com)（2024 年）\n84. [Transformer 中什么才是关键？并非所有的注意力都是必要的](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.15786)（2024 年）\n85. [时间序列基础模型：它们在异常检测和预测中的作用](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.19286v1)（2024 年）\n86. [零样本时间序列基础模型在云端数据上的表现](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12944)（2025 年）🔥🔥🔥🔥🔥\n87. [立场：长期时间序列预测中不存在绝对的冠军](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.12161)（2025 年）\n88. [FinTSB：一个全面且实用的金融时间序列预测基准](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.18834)\n89. [时间序列预测中的挑拣现象：如何选择数据集让你的模型脱颖而出](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.14435)\n90. [TFB：迈向全面而公平的时间序列预测方法基准测试](https:\u002F\u002Farxiv.org\u002Fabs\u002F2403.20150) 🔥🔥🔥🔥🔥\n91. [DUET：双聚类增强的多变量时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2412.10859)\n92. [CycleNet：通过建模周期性模式来增强时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2409.18479) [代码](https:\u002F\u002Fgithub.com\u002FACAT-SCUT\u002FCycleNet)\n93. [LLM 能否理解时间序列异常？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2410.05440) [代码](https:\u002F\u002Fgithub.com\u002Frose-stl-lab\u002Fanomllm)\n94. [EMForecaster：一个用于无线网络时间序列预测的深度学习框架，具备无分布不确定性量化功能](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.00120)\n95. [Times2D：多周期分解和导数映射用于通用时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2504.00118) [代码](https:\u002F\u002Fgithub.com\u002FTims2D\u002FTimes2D)\n96. [FilterTS：针对多变量时间序列的全面频率过滤](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.04158)\n97. [OLinear：正交变换域中的线性模型用于时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.08550)\n98. [规模法则是否适用于时间序列预测？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.10172)\n99. [TiRex：通过增强上下文学习，在长短时距范围内进行零样本预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2505.23719) [代码](https:\u002F\u002Fgithub.com\u002FNX-AI\u002Ftirex) [文章](https:\u002F\u002Fhuggingface.co\u002FNX-AI\u002FTiRex) [播客](https:\u002F\u002Fopen.spotify.com\u002Fepisode\u002F5PmGnhjPf5JMI1SdVqwOu4) 🔥🔥🔥🔥🔥\n100. [FreDF：在频域中学习预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2402.02399) [代码](https:\u002F\u002Fgithub.com\u002FMaster-PLC\u002FFreDF)\n101. [KARMA：一个多层级分解的混合 Mamba 框架，用于多变量长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.08939) [代码](https:\u002F\u002Fgithub.com\u002Fyedadasd\u002FKARMA)\n102. [MS-TVNet：一种基于多尺度动态卷积的长期时间序列预测方法](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.17253)\n103. [变分模态分解和线性嵌入是你进行时间序列预测所需要的](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.16122) [代码](https:\u002F\u002Fgithub.com\u002FEspalemit\u002FVMD-With-LTSF-Linear) 🔥🔥🔥🔥🔥\n104. [TimeSieve：通过信息瓶颈提取时间动态](https:\u002F\u002Farxiv.org\u002Fabs\u002F2406.05036) [代码](https:\u002F\u002Fgithub.com\u002Fxll0328\u002FTimeSieve)\n105. [VisionTS++：跨模态时间序列基础模型，具有持续预训练的视觉骨干网络](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.04379) [代码](https:\u002F\u002Fgithub.com\u002FHALF111\u002FVisionTSpp) VisionTS 🔥🔥🔥🔥🔥\n106. [VISIONTS：视觉掩码自编码器是免费午餐式的零样本时间序列预测者](https:\u002F\u002Farxiv.org\u002Fabs\u002F2408.17253) [代码](https:\u002F\u002Fgithub.com\u002FKeytoyze\u002FVisionTS) VisionTS 🔥🔥🔥🔥🔥\n107. [小波专家混合用于时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2508.08825)\n108. [TFKAN：用于长期时间序列预测的时频 KAN](https:\u002F\u002Farxiv.org\u002Fabs\u002F2506.12696)\n109. [为什么 Transformer 在上下文中无法预测时间序列？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.09776) 🔥🔥🔥🔥🔥 [代码](https:\u002F\u002Fgithub.com\u002FMasterZhou1\u002FICL-Time-Series)\n110. [使用 Transformer 学习线性动力学系统：误差范围与深度分离](https:\u002F\u002Farxiv.org\u002Fabs\u002F2502.08136)\n111. [时间序列基础模型：基准测试的挑战与要求](https:\u002F\u002Farxiv.org\u002Fabs\u002F2510.13654) 🔥🔥🔥🔥🔥\n112. [DoFlow：用于干预性和反事实时间序列预测的因果生成流](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.02137) 🔥🔥🔥🔥🔥\n113. [Hydra：用于多变量时间序列分析的双重指数记忆](https:\u002F\u002Farxiv.org\u002Fabs\u002F2511.00989) 🔥🔥🔥🔥🔥\n114. [KAN 混合用于多变量时间序列预测](https:\u002F\u002Fdl.acm.org\u002Fdoi\u002Fabs\u002F10.1145\u002F3746252.3760836) 🔥🔥🔥🔥🔥\n115. [跨频率迁移学习和基础预测模型的真实评估](https:\u002F\u002Farxiv.org\u002Fabs\u002F2509.19465) 一篇论文显示，基础模型在性能上系统性地逊于 ARIMA 和简单的统计集成模型 🔥🔥🔥🔥🔥\n116. [XLinear：一种轻量级且准确的基于 MLP 的模型，用于带有外生变量的长期时间序列预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2601.09237)\n117. [立场：时间序列预测中“一刀切”架构的时代终将结束](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.01736)\n118. [在时间序列基础模型中使用外生数据评估电力需求预测](https:\u002F\u002Farxiv.org\u002Fabs\u002F2602.05390)\n\n## 文章\n1. [TimeGPT 与 TiDE：零样本推理是预测的未来，还是只是炒作？](https:\u002F\u002Farxiv.org\u002Fabs\u002F2205.13504) (https:\u002F\u002Ftowardsdatascience.com\u002Ftimegpt-vs-tide-is-zero-shot-inference-the-future-of-forecasting-or-just-hype-9063bdbe0b76) 作者：Luís Roque 和 Rafael Guedes。（2024年）🔥🔥🔥🔥🔥\n2. [TimeGPT-1，在 Hacker News 上的讨论](https:\u002F\u002Fnews.ycombinator.com\u002Fitem?id=37874891)（2023年）\n3. [TimeGPT：首个用于时间序列预测的生成式预训练 Transformer](https:\u002F\u002Fwww.reddit.com\u002Fr\u002FMachineLearning\u002Fcomments\u002F176wsne\u002Fr_timegpt_the_first_generative_pretrained\u002F)","# Transformers_And_LLM_Are_What_You_Dont_Need 快速上手指南\n\n**项目说明**：\n本项目并非一个单一的 Python 库或可安装的工具包，而是一个**学术资源汇总仓库**。它收集并整理了大量证明\"Transformer 和大语言模型（LLM）在时间序列预测中并非最优解”的论文、视频、学位论文和技术文章。核心理念是：对于时间序列预测任务，简单的线性模型（Linear）、多层感知机（MLP）或基于频域的方法往往比复杂的 Transformer 架构更有效且高效。\n\n本指南将指导你如何获取这些资源，并运行其中提到的代表性替代模型代码。\n\n## 环境准备\n\n由于本项目包含多篇不同论文的参考实现链接，你需要准备一个通用的深度学习开发环境。大多数列出的模型（如 DLinear, FreTS, TimesNet 等）均基于 PyTorch。\n\n*   **操作系统**: Linux (推荐 Ubuntu 20.04+), macOS, 或 Windows (WSL2 推荐)\n*   **Python 版本**: 3.8 - 3.10\n*   **核心依赖**:\n    *   PyTorch >= 1.10\n    *   NumPy\n    *   Pandas\n    *   Scikit-learn\n    *   Matplotlib\n\n**前置依赖安装命令**：\n```bash\npip install torch numpy pandas scikit-learn matplotlib\n```\n\n> **国内加速建议**：\n> 推荐使用清华源或阿里源加速 PyTorch 及相关库的安装：\n> ```bash\n> pip install torch numpy pandas scikit-learn matplotlib -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n> ```\n\n## 安装步骤\n\n本项目本身**无需安装**。它是一个索引列表。要使用其中的技术，你需要根据需求选择具体的论文代码仓库进行克隆。\n\n以下是获取该领域最具代表性的“反 Transformer\"模型（例如 **DLinear**，出自论文 *Are Transformers Effective for Time Series Forecasting?*）的步骤：\n\n1.  **克隆代表性模型仓库** (以 LTSF-Linear 为例)：\n    ```bash\n    git clone https:\u002F\u002Fgithub.com\u002Fcure-lab\u002FLTSF-Linear.git\n    cd LTSF-Linear\n    ```\n    *(注：若国内访问 GitHub 缓慢，可使用 `git clone https:\u002F\u002Fghproxy.com\u002Fhttps:\u002F\u002Fgithub.com\u002Fcure-lab\u002FLTSF-Linear.git` 等加速服务)*\n\n2.  **安装该具体模型的依赖**：\n    大多数此类项目都包含 `requirements.txt`。\n    ```bash\n    pip install -r requirements.txt -i https:\u002F\u002Fpypi.tuna.tsinghua.edu.cn\u002Fsimple\n    ```\n\n## 基本使用\n\n以下以 **DLinear** (Decomposition Linear) 模型为例，展示如何使用这类轻量级模型进行时间序列预测。这是 README 中排名前列且被广泛验证有效的模型。\n\n### 1. 数据准备\n确保你拥有格式为 `.csv` 的时间序列数据（通常包含时间戳和数值列）。该项目通常自带示例数据（如 `dataset\u002FETT-small\u002F`）。\n\n### 2. 运行预测脚本\n大多数仓库提供了统一的运行脚本。以下命令展示了如何训练并评估 DLinear 模型：\n\n```bash\npython run.py --model DLinear --data ETTm1 --features M --seq_len 96 --pred_len 96 --e_layers 2 --d_model 512 --batch_size 32 --learning_rate 0.001\n```\n\n**参数简析**：\n*   `--model DLinear`: 指定使用线性分解模型。\n*   `--data ETTm1`: 指定数据集（电力变压器温度数据）。\n*   `--seq_len 96`: 输入序列长度。\n*   `--pred_len 96`: 预测未来序列长度。\n*   `--features M`: 多变量预测模式。\n\n### 3. 尝试其他架构\n根据 README 列表，你可以轻松切换到其他非 Transformer 架构，只需更改 `--model` 参数（需确保对应代码已集成或克隆）：\n\n*   **频域 MLP (FreTS)**:\n    ```bash\n    python run.py --model FreTS --data ETTm1 ...\n    ```\n*   **纯卷积网络 (ModernTCN)**:\n    ```bash\n    python run.py --model ModernTCN --data ETTm1 ...\n    ```\n*   **状态空间模型 (Mamba 变体)**:\n    ```bash\n    python run.py --model Mamba --data ETTm1 ...\n    ```\n\n### 4. 查阅更多资源\n若要深入研究 README 中列出的其他 50+ 篇论文，请直接访问原仓库中的 [Papers](https:\u002F\u002Fgithub.com\u002Fvaleman\u002FTransformers_And_LLM_Are_What_You_Dont_Need#papers) 章节，点击对应链接获取特定模型的官方代码实现。\n\n**核心理念总结**：在处理时间序列预测时，优先尝试线性模型、MLP 或频域方法，仅在必要时再考虑复杂的 Transformer 架构。","某零售连锁企业的数据科学团队正试图利用深度学习预测未来三个月的区域销量，以优化库存管理。\n\n### 没有 Transformers_And_LLM_Are_What_You_Dont_Need 时\n- 团队盲目跟风，花费数周复现复杂的 Transformer 架构，却因时间序列数据缺乏语义关联而导致模型收敛困难。\n- 计算资源被大量浪费在训练参数量巨大的大语言模型（LLM）上，推理延迟高，无法满足每日快速更新预测的需求。\n- 由于缺乏对领域内 SOTA（最先进）非 Transformer 模型的了解，团队误以为“模型越复杂效果越好”，导致预测准确率反而低于简单的线性基准。\n- 在遇到性能瓶颈时，团队难以找到针对性的学术依据来说服管理层调整技术路线，陷入无效优化的死循环。\n\n### 使用 Transformers_And_LLM_Are_What_You_Dont_Need 后\n- 团队通过仓库中收录的《Are Transformers Effective for Time Series Forecasting?》等关键论文，迅速意识到 Transformer 在此场景的局限性，及时止损。\n- 转而采用仓库推荐的频域 MLP 或线性映射模型（如 LTSF-Linear），在大幅降低算力成本的同时，将预测准确率提升了 15%。\n- 借助仓库整理的视频讲解和硕士论文案例，团队快速构建了坚实的理论防线，向管理层清晰阐述了为何“简单模型”更适合当前的时间序列任务。\n- 直接复用仓库提供的 SCINet 或 WinNet 等高效模型代码库，将原本需要一个月的研发周期缩短至三天，迅速上线了高精度的库存预测系统。\n\nTransformers_And_LLM_Are_What_You_Dont_Need 的核心价值在于帮助开发者打破“大模型迷信”，用实证研究引导团队回归数学本质，选择真正适合时间序列预测的高效方案。","https:\u002F\u002Foss.gittoolsai.com\u002Fimages\u002Fvaleman_Transformers_And_LLM_Are_What_You_Dont_Need_a4cc446a.png","valeman","https:\u002F\u002Foss.gittoolsai.com\u002Favatars\u002Fvaleman_d9fe8df6.jpg","Machine Learning | Probabilistic Prediction | Time-series | Systematic Trading | Forecasting Innovation  | Creator of Awesome Conformal Prediction","Big Business","London, UK",null,"predict_addict","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fvaleriy-manokhin-phd-mba-cqf-704731236\u002F","https:\u002F\u002Fgithub.com\u002Fvaleman",851,60,"2026-04-18T10:14:42","","未说明",{"notes":91,"python":89,"dependencies":92},"该仓库并非一个可运行的 AI 工具或代码库，而是一个资源汇总列表（Awesome List）。它主要收集了论证'Transformers 和大语言模型在时间序列预测中效果不佳’的论文、文章、视频和学位论文链接。因此，不存在具体的运行环境、依赖库或硬件需求。用户只需通过浏览器访问列表中提供的链接即可阅读相关内容。",[],[18],"2026-03-27T02:49:30.150509","2026-04-19T03:03:48.729081",[],[]]