Bootstrap

ICLR 2025 | 时间序列(Time Series)论文总结

ICLR 2025将在2025年4月24日到28日于新加坡举行。ICLR 2025共有11,565篇投稿，录取率32.08%。本文总结了2025 ICLR上有关时间序列（time series）相关论文。如有疏漏，欢迎大家补充。

时间序列Topic：预测，插补，分类，生成，因果分析，异常检测，LLM以及基础模型等内容。总计42篇。其中1-4为Oral，5，6为Spotlight

  1. [Oral] TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis

  2. [Oral] Root Cause Analysis of Anomalies in Multivariate Time Series through Granger Causal Discovery

  3. [Oral] Amortized Control of Continuous State Space Feynman-Kac Model for Irregular Time Series

  4. [Oral] On the Identification of Temporal Causal Representation with Instantaneous Dependence

  5. [Spotlight] CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series

  6. [Spotlight] Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts

  7. Can LLMs Understand Time Series Anomalies?

  8. In-context Time Series Predictor

  9. Optimal Transport for Time Series Imputation

  10. TimeInf: Time Series Data Contribution via Influence Functionn

  11. Compositional simulation-based inference for time series

  12. DyCAST: Learning Dynamic Causal Structure from Time Series

  13. Timer-XL: Long-Context Transformers for Unified Time Series Forecasting

  14. PPT: Patch Order Do Matters In Time Series Pretext Task

  15. Diffusion Transformers for Tabular Data Time Series Generation

  16. Going Beyond Static: Understanding Shifts with Time-Series Attribution

  17. Kernel-based Optimally Weighted Conformal Time-Series Prediction

  18. Learn hybrid prototypes for multivariate time series anomaly detection

  19. Investigating Pattern Neurons in Urban Time Series Forecasting

  20. Diffusion-based Decoupled Deterministic and Uncertain Framework for Probabilistic Multivariate Time Series Forecasting

  21. Label Correlation Biases Direct Time Series Forecast

  22. Fast and Slow Streams for Online Time Series Forecasting Without Information Leakage

  23. Locally Connected Echo State Networks for Time Series Forecasting

  24. Shifting the Paradigm: A Diffeomorphism Between Time Series Data Manifolds for Achieving Shift-Invariancy in Deep Learning

  25. Efficient Source-Free Time-Series Adaptation via Parameter Subspace Disentanglement

  26. Physiome-ODE: A Benchmark for Irregularly Sampled Multivariate Time-Series Forecasting Based on Biological ODEs

  27. Towards Neural Scaling Laws for Time Series Foundation Models

  28. CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patching

  29. CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution

  30. Context-Alignment: Activating and Enhancing LLMs Capabilities in Time Series

  31. TVNet: A Novel Time Series Analysis Method Based on Dynamic Convolution and 3D-Variation

  32. Multi-Resolution Decomposable Diffusion Model for Non-Stationary Time Series Anomaly Detection

  33. KooNPro: A Variance-Aware Koopman Probabilistic Model Enhanced by Neural Processes for Time Series Forecasting

  34. Towards a General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders

  35. TimeKAN: KAN-based Frequency Decomposition Learning Architecture for Long-term Time Series Forecasting

  36. S4M: S4 for multivariate time series forecasting with Missing values

  37. A Simple Baseline for Multivariate Time Series Forecasting

  38. Shedding Light on Time Series Classification using Interpretability Gated Networks

  39. Flow Matching with Gaussian Process Priors for Probabilistic Time Series Forecasting

  40. Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data

  41. TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting

  42. Quantifying Past Error Matters: Conformal Inference for Time Series

🌟【紧跟前沿】“时空探索之旅”与你一起探索时空奥秘！🚀
欢迎大家关注时空探索之旅时空探索之旅在这里插入图片描述

[Oral] 1 TimeMixer++: A General Time Series Pattern Machine for Universal Predictive Analysis

链接：https://openreview.net/forum?id=1CLzLXSFNn

作者：Shiyu Wang, Jiawei LI, Xiaoming Shi, Zhou Ye, Baichuan Mo, Wenze Lin, Ju Shengtong, Zhixuan Chu, Ming Jin

关键词：多任务（预测，分类，插补，异常检测），基础模型

TL; DR：TimeMixer++ 是一种时间序列模式机器，它采用多尺度和多分辨率模式提取，在 8 种不同的分析任务中提供 SOTA 性能，包括预测、分类、异常检测和插补。

分数：6810

[Oral]2 Root Cause Analysis of Anomalies in Multivariate Time Series through Granger Causal Discovery

链接：https://openreview.net/forum?id=k38Th3x4d9

作者：Xiao Han, Saima Absar, Lu Zhang, Shuhan Yuan

关键词：因果发现, 格兰杰因果

分数：88888

[Oral] 3 Amortized Control of Continuous State Space Feynman-Kac Model for Irregular Time Series

链接：https://openreview.net/forum?id=8zJRon6k5v

作者：Byoungwoo Park, Hyungi Lee, Juho Lee

关键词：随机最优控制，变分推断，不规则时间序列，状态空间模型, 不规则时间序列

TL; DR：提出了一个多边际 Doob 的 -transform 用于不规则的时间序列和变分推理，并使用随机最优控制来近似它。

分数：8888

[Oral] 4 On the Identification of Temporal Causal Representation with Instantaneous Dependence

链接：https://openreview.net/forum?id=2efNHgYRvM

作者：Zijian Li, Yifan Shen, Kaitao Zheng, Ruichu Cai, Xiangchen Song, Mingming Gong, Guangyi Chen, Kun Zhang

关键词：因果表示学习, Instantaneous Dependency, Identification

分数：888

[Spotlight] 5 CausalRivers - Scaling up benchmarking of causal discovery for real-world time-series

链接：https://openreview.net/forum?id=wmV4cIbgl6

作者：Gideon Stein, Maha Shadaydeh, Jan Blunk, Niklas Penzel, Joachim Denzler

关键词：因果发现， Benchmarking

TL; DR：迄今为止最大的真实世界因果发现基准测试，包括高分辨率 ts 和具有 1000 多个节点的真实因果图。

分数：888

[Spotlight] 6 Time-MoE: Billion-Scale Time Series Foundation Models with Mixture of Experts

链接：https://openreview.net/forum?id=e1wDDFmlVu

作者：Xiaoming Shi, Shiyu Wang, Yuqi Nie, Dianqi Li, Zhou Ye, Qingsong Wen, Ming Jin

关键词：预测，基础模型，混合专家系统

TL; DR：Time-MoE 是一系列具有 mixture-of-experts 架构的时间序列基础模型。Time-MoE 首次扩展到 24 亿个参数，从而显著提高了零镜头/全镜头预测性能。

分数：688

7 Can LLMs Understand Time Series Anomalies?

链接：https://openreview.net/forum?id=LGafQ1g2D2

作者：Zihao Zhou, Rose Yu

关键词：大语言模型，异常检测，多模态学习

TL; DR：本研究挑战了关于大型语言模型在时间序列异常检测方面能力的常见假设。

分数：63665

8 In-context Time Series Predictor

链接：https://openreview.net/forum?id=dCcY2pyNIO

作者：Jiecheng Lu, Yan Sun, Shihao Yang

关键词：预测，上下文学习

分数：3668

9 Optimal Transport for Time Series Imputation

链接：https://openreview.net/forum?id=xPTzjpIQNp

作者：Hao Wang, zhengnan li, Haoxuan Li, Xu Chen, Mingming Gong, BinChen, Zhichao Chen

关键词：插补，最优传输

分数：588

10 TimeInf: Time Series Data Contribution via Influence Functions

链接：https://openreview.net/forum?id=Vz0CWFMPUe

作者：Yizi Zhang, Jingyan Shen, Xiaoxue Xiong, Yongchan Kwon

关键词：异常检测

分数：56666

11 Compositional simulation-based inference for time series

链接：https://openreview.net/forum?id=uClUUJk05H

作者：Manuel Gloeckler, Shoji Toyota, Kenji Fukumizu, Jakob H. Macke

关键词：Simulation-based inference, Bayesian inference, time series, markovian simulators, Amortized Bayesian inference

分数：566668

12 DyCAST: Learning Dynamic Causal Structure from Time Series

链接：https://openreview.net/forum?id=WjDjem8mWE

作者：Yue Cheng, Bochen Lyu, Weiwei Xing, Zhanxing Zhu

关键词：动态因果发现;时间序列

分数：3668

13 Timer-XL: Long-Context Transformers for Unified Time Series Forecasting

链接：https://openreview.net/forum?id=KMCJXjlDDr

作者：Yong Liu, Guo Qin, Xiangdong Huang, Jianmin Wang, Mingsheng Long

关键词：预测，Transformer

TL; DR：将 next token prediction 扩展到多变量时间序列，为各种预测场景提供了一个生成式 Transformer，它在各种基准中实现了 SOTA，并展示了作为 one-for-all 预测器的强大实力。

分数：656

14 PPT: Patch Order Do Matters In Time Series Pretext Task

链接：https://openreview.net/forum?id=7zwIEbSTDy

作者：Jaeho Kim, Kwangryeol Park, Sukmin Yun, Seulki Lee

关键词：分类，自监督

TL; DR：用于时间序列分类的补丁顺序感知自监督学习方法

分数：6683

15 Diffusion Transformers for Tabular Data Time Series Generation

链接：https://openreview.net/forum?id=bhOysNJvWm

作者：Fabrizio Garuti, Enver Sangineto, Simone Luetto, Lorenzo Forni, Rita Cucchiara

关键词：表格数据生成，扩散模型

分数：665

16 Going Beyond Static: Understanding Shifts with Time-Series Attribution

链接：https://openreview.net/forum?id=XQlccqJpCC

作者：Jiashuo Liu, Nabeel Seedat, Peng Cui, Mihaela van der Schaar

关键词：分布变化、性能下降、归因

TL; DR：提出了一个时间序列偏移归因框架，该框架以理论分析和实证结果的支持，将各种类型偏移的性能下降详细归因于每个时间数据属性。

分数：5886

17 Kernel-based Optimally Weighted Conformal Time-Series Prediction

链接：https://openreview.net/forum?id=oP7arLOWix

作者：Jonghyeok Lee, Chen Xu, Yao Xie

关键词：共形预测，非参数核回归

TL; DR：一种通过非参数核回归对时间序列进行顺序共形预测方法，具有严格的理论分析和强大的实证性能。

分数：6666

18 Learn hybrid prototypes for multivariate time series anomaly detection

链接：https://openreview.net/forum?id=8TBGdH3t6a

作者：Ke-Yuan Shen

关键词：原型，异常检测

TL; DR：本文提出了一种基于重建的 MTSAD 混合原型学习模型，以对抗过度泛化。

分数：66556

19 Investigating Pattern Neurons in Urban Time Series Forecasting

链接：https://openreview.net/forum?id=a9vey6B54y

作者：Chengxin Wang, Yiran Zhao, Shaofeng Cai, Gary Tan

关键词：时空预测（更像是），neuron detection

分数：6666

20 Diffusion-based Decoupled Deterministic and Uncertain Framework for Probabilistic Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=HdUkF1Qk7g

作者：Qi Li, Zhenyu Zhang, Lei Yao, Zhaoxia Li, Tianyi Zhong, Yong Zhang

关键词：长时预测，扩散模型

分数：6666

21 Label Correlation Biases Direct Time Series Forecast

链接：https://openreview.net/forum?id=4A9IdSa1ul

作者：Hao Wang, Lichen Pan, Yuan Shen, Zhichao Chen, Degui Yang, Yifei Yang, Sen Zhang, Xinggao Liu, Haoxuan Li, Dacheng Tao

关键词：长时预测

TL; DR：学习在频域中进行预测可显著提高预测性能。

分数：8686

22 Fast and Slow Streams for Online Time Series Forecasting Without Information Leakage

链接：https://openreview.net/forum?id=I0n3EyogMi

作者：Ying-yee Ava Lau, Zhiwen Shao, Dit-Yan Yeung

关键词：online time series forecasting, concept drift, online learning

TL; DR：重新定义了在线时间序列预测的设置以防止信息泄露，并为此设置提出了一个与模型无关的框架。

分数：6688

23 Locally Connected Echo State Networks for Time Series Forecasting

链接：https://openreview.net/forum?id=KeRwLLwZaw

作者：Filip Matzner, František Mráz

关键词：Time Series Analysis, Time Series Forecasting, Recurrent Networks, Regression, Echo State Networks

TL; DR：改进的本地连接 ESN 方法可与真实世界时间序列数据集上的最新方法相媲美。

分数：6666

24 Shifting the Paradigm: A Diffeomorphism Between Time Series Data Manifolds for Achieving Shift-Invariancy in Deep Learning

链接：https://openreview.net/forum?id=nibeaHUEJx

作者：Berken Utku Demirel, Christian Holz

关键词：Time series analysis, invariance in neural networks

分数：6688

25 Efficient Source-Free Time-Series Adaptation via Parameter Subspace Disentanglement

链接：https://openreview.net/forum?id=Q5Sawm0nqo

作者：Gaurav Patel, Christopher Michael Sandino, Behrooz Mahasseni, Ellen L. Zippi, Erdrin Azemi, Ali Moin, Juri Minxha

关键词： Time Series, Source-Free Domain Adaptation, Efficiency

TL; DR：用于时间序列的高效无源域适应框架，可提高适应过程的参数和样本效率。

分数：6105565

26 Physiome-ODE: A Benchmark for Irregularly Sampled Multivariate Time-Series Forecasting Based on Biological ODEs

链接：https://openreview.net/forum?id=6ouZaBzeNO

作者：Christian Klötergens, Vijaya Krishna Yalavarthi, Randolf Scholz, Maximilian Stubbemann, Stefan Born, Lars Schmidt-Thieme

关键词：Irregular Time Series, ODE

TL; DR：基于存储在 Physiome 模型存储库中的生物 ODE 模型创建的不规则时间序列预测基准。

分数：6333

27 Towards Neural Scaling Laws for Time Series Foundation Models

链接：https://openreview.net/forum?id=uCqxDfLYrB

作者：Qingren Yao, Chao-Han Huck Yang, Renhe Jiang, Yuxuan Liang, Ming Jin, Shirui Pan

关键词：Time series, scaling law, foundation model, transformer, forecasting

分数：5668

28 CATCH: Channel-Aware Multivariate Time Series Anomaly Detection via Frequency Patching

链接：https://openreview.net/forum?id=m08aK3xxdJ

作者：Xingjian Wu, Xiangfei Qiu, Zhengyu Li, Yihang Wang, Jilin Hu, Chenjuan Guo, Hui Xiong, Bin Yang

关键词：多元时间序列异常检测

分数：5668

29 CoMRes: Semi-Supervised Time Series Forecasting Utilizing Consensus Promotion of Multi-Resolution

链接：https://openreview.net/forum?id=bRa4JLPzii

作者：Yunju Cho, Jay-Yoon Lee

关键词：预测，多尺度，半监督

TL; DR：提出了一种利用 CON 的新型半监督时间序列预测

分数：8665

30 Context-Alignment: Activating and Enhancing LLMs Capabilities in Time Series

链接：https://openreview.net/forum?id=syC2764fPc

作者：Yuxiao Hu, Qian Li, Jinyue Yan, Dongxiao Zhang, Yuntian Chen

关键词：大模型，上下文对齐

TL; DR：用于时间序列任务的 LLM

分数：6666

31 TVNet: A Novel Time Series Analysis Method Based on Dynamic Convolution and 3D-Variation

链接：https://openreview.net/forum?id=MZDdTzN6Cy

作者：Chenghan Li, Mingchen Li, Ruisheng Diao

关键词：时序分析，动态卷积

TL; DR：新的基于时间序列建模视角的 3D 变化和新的分析框架

分数：5668

32 Multi-Resolution Decomposable Diffusion Model for Non-Stationary Time Series Anomaly Detection

链接：https://openreview.net/forum?id=eWocmTQn7H

作者：Guojin Zhong, pan wang, Jin Yuan, Zhiyong Li, Long Chen

关键词：Diffusion Model, Non-Stationary Time Series, Anomaly Detection, Multi-Resolution

TL; DR：本文深入探讨了多分辨率技术和扩散模型在非平稳时间序列异常检测方面的潜力，并得到了严格的数学证明的支持。

分数：6668

33 KooNPro: A Variance-Aware Koopman Probabilistic Model Enhanced by Neural Processes for Time Series Forecasting

链接：https://openreview.net/forum?id=5oSUgTzs8Y

作者：Ronghua Zheng, Hanru Bai, Weiyang Ding

关键词：概率预测，库普曼模型，神经过程

分数：66666

34 Towards a General Time Series Anomaly Detector with Adaptive Bottlenecks and Dual Adversarial Decoders

链接：https://openreview.net/forum?id=aKcd7ImG5e

作者：Qichao Shentu, Beibu Li, Kai Zhao, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, Chenjuan Guo

关键词：异常检测

TL; DR：提出了一种通用的时间序列异常检测模型，该模型在多域数据集上进行了预训练，随后可以应用于许多下游场景

分数：6666

35 TimeKAN: KAN-based Frequency Decomposition Learning Architecture for Long-term Time Series Forecasting

链接：https://openreview.net/forum?id=wTLc79YNbh

作者：Songtao Huang, Zhen Zhao, Can Li, LEI BAI

关键词：kolmogorov-Arnold Network;预测

分数：3588

36 S4M: S4 for multivariate time series forecasting with Missing values

链接：https://openreview.net/forum?id=BkftcwIVmR

作者：Peng Jing, Meiqi Yang, Qiong Zhang, Xiaoxiao Li

关键词：S4 Models, Multivariate Time Series Forecasting, Missing Value, Prototype Bank

分数：5665

37 A Simple Baseline for Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=oANkBaVci5

作者：Hui Chen, Viet Luong, Lopamudra Mukherjee, Vikas Singh

关键词：预测，小波

分数：5688

38 Shedding Light on Time Series Classification using Interpretability Gated Networks

链接：https://openreview.net/forum?id=n34taxF0TC

作者：Yunshi Wen, Tengfei Ma, Ronny Luss, Debarun Bhattacharjya, Achille Fokoue, Anak Agung Julius

关键词：可解释性，Shapelet

TL;DR: 将可解释模型与深度神经网络相结合，实现可解释的时间序列分类的框架。

分数：65688

39 Flow Matching with Gaussian Process Priors for Probabilistic Time Series Forecasting

链接：https://openreview.net/forum?id=uxVBbSlKQ4

作者：Marcel Kollovieh, Marten Lienen, David Lüdke, Leo Schwinn, Stephan Günnemann

关键词： 流匹配，预测，生成模型

TL; DR：提出了 TSFlow，这是一种用于时间序列预测的条件流匹配模型，它利用特定于领域的先验

分数：8865

40 Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data

链接：https://openreview.net/forum?id=Vp2OAxMs2s

作者：Manuel Brenner, Elias Weber, Georgia Koppe, Daniel Durstewitz

关键词：动力系统、循环神经网络、分层建模、时间序列、可解释性、非线性动力学、基础模型

TL; DR：跨多个领域/主题推断动态系统模型的通用框架。

分数：6818

41 TS-LIF: A Temporal Segment Spiking Neuron Network for Time Series Forecasting

链接：https://openreview.net/forum?id=rDe9yQQYKt

作者：FENG SHIBO, Wanjin Feng, Xingyu Gao, Peilin Zhao, Zhiqi Shen

关键词：spiking neural network, time series forecasting, Application

TL; DR：提出了一种用于多变量时间序列预测的 Temporal Segment Spiking Neuron Network （TS-LIF），并辅以稳定性分析和频率响应分析，以证明其有效性和效率。

分数：6666

42 Quantifying Past Error Matters: Conformal Inference for Time Series

链接：https://openreview.net/forum?id=RD9q5vEe1Q

作者：Junxi Wu, Dongjian Hu, Yajie Bao, Shu-Tao Xia, Changliang Zou

关键词：不确定性量化，共形预测，分布偏移

TL; DR：通过量化覆盖不足/过度的程度提出了一种新的在线共形推理方法 ECI，该方法可以对时间序列中的分布变化做出快速反应。

分数：5668

🌟【紧跟前沿】“时空探索之旅”与你一起探索时空奥秘！🚀
欢迎大家关注时空探索之旅时空探索之旅在这里插入图片描述

悦读

道可道，非常道；名可名，非常名。 无名，天地之始，有名，万物之母。 故常无欲，以观其妙，常有欲，以观其徼。 此两者，同出而异名，同谓之玄，玄之又玄，众妙之门。

最新收录
20、《Spring AOP：日志记录与权限控制实战》
老男孩-Python之路,Day1 - Python基础1
c++动态绑定和静态绑定
RiskSense Spotlight：全球知名开源软件漏洞分析报告
一文彻底搞懂Nginx的.conf文件路径配置
java~字节码操作Javassist
TensorRT&Sample&Python[network_api_pytorch_mnist]
Python+chatGPT编程5分钟快速上手，强烈推荐！！！
JVM定位问题之jstack使用
[源码解析] PyTorch 流水线并行实现 (4)--前向计算
;