ICLR 2026 | 时间序列（Time Series）论文总结（上）【预测，多模态，预测×LLM，基础模型】

时空探索之旅

发布于 2026-03-10 16:01:19

2580

ICLR 2026将在2025年4月24日到28日于巴西里约热内卢（Rio de Janeiro, Brazil）举行。ICLR 2025共有19,000多篇投稿，录用5,359篇，录取率28.18%。

根据笔者不完全统计，ICLR 2026时间序列（Time Series）的论文，总计87篇，本文涉及39篇，如有疏漏，欢迎补充。

观察：39篇时序预测相关文章统计值

最大均分	均值	最小均分
6.5	5.21	4

其中均分≥6的有9篇，根据OpenReview显示，应该均为Poster，无Oral和Spotlight（如有错误，还请大家评论区更正）。

注：由于论文数目较多，分为上下篇，此为上篇，主要涵盖时间序列预测，包括但不限于不规则时序，多模态，可解释性，概率预测，不确定性，在线预测，benchmark，预测和LLM以及基础模型结合。

下篇主要涵盖基础模型，异常检测，分类，插补，表示学习等

祝大家小年快乐！

1. MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with Parameters2. CoRA: Boosting Time Series Foundation Models for Multivariate Forecasting through Correlation-aware Adapter3. PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting4. Reliable Probabilistic Forecasting of Irregular Time Series through Marginalization-Consistent Flows5. Numerion: A Multi-Hypercomplex Model for Time Series Forecasting6. Conditionally Whitened Generative Models for Probabilistic Time Series Forecasting7. Delta-XAI: A Unified Framework for Explaining Prediction Changes in Online Time Series Monitoring8. COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting9. CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting10. Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval11. Bridging Past and Future: Distribution-Aware Alignment for Time Series Forecasting12. Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling13. ResCP: Reservoir Conformal Prediction for Time Series Forecasting14. GCGNet: Graph-Consistent Generative Network for Time Series Forecasting with Exogenous Variables15. Test-Time Efficient Pretrained Model Portfolios for Time Series Forecasting16. MMPD: Diverse Time Series Forecasting via Multi-Mode Patch Diffusion Loss17. Local Geometry Attention for Time Series Forecasting under Realistic Corruptions18. FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting19. Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting20. Aurora: Towards Universal Generative Multimodal Time Series Forecasting21. TEDM: Time Series Forecasting with Elucidated Diffusion Models22. The Forecast After the Forecast: A Post-Processing Shift in Time Series23. TAMMs: Change Understanding and Forecasting in Satellite Image Time Series with Temporal-Aware Multimodal Models24. Online time series prediction using feature adjustment25. DoFlow: Flow-based Generative Models for Interventional and Counterfactual Forecasting on Time Series26. Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness27. Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition28. Flow-based Conformal Prediction for Multi-dimensional Time Series29. ASTGI: Adaptive Spatio-Temporal Graph Interactions for Irregular Multivariate Time Series Forecasting30. ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting31. PhaseFormer: From Patches to Phases for Efficient and Effective Time Series Forecasting32. Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting33. Characteristic Root Analysis and Regularization for Linear Time Series Forecasting34. Semantic-Enhanced Time-Series Forecasting via Large Language Models35. TimeSeg: An Information-Theoretic Segment-Wise Explainer for Time-Series Predictions36. TimeRecipe: A Time-Series Forecasting Recipe via Benchmarking Module Level Effectiveness37. Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift38. Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models39. DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment

1 MixLinear: Extreme Low Resource Multivariate Time Series Forecasting with Parameters

链接：https://openreview.net/forum?id=QUj0KuCumD

关键词：Long-term Time Series Forecasting, Segmentation, Adaptive Low-Rank Spectral Filtering

作者：Aitian Ma, Dongsheng Luo, Mo Sha

分数：4, 8, 6, 4

信心：5, 2, 3, 4

均分：5.5

2 CoRA: Boosting Time Series Foundation Models for Multivariate Forecasting through Correlation-aware Adapter

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

关键词：Time Series Forecasting

作者：Hanyin Cheng, Xingjian Wu, Yang Shu, Zhongwen Rao, Lujia Pan, Bin Yang, Chengjuan Guo

分数：8, 2, 8, 6

信心：5, 4, 4, 3

均分：6.0

3 PHAT: Modeling Period Heterogeneity for Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=lr4RlISR6x

关键词：Time series forecasting, time series data, deep learning

作者：Jiaming Ma, Guanjun Wang, Qihe Huang, Sheng Huang, Haofeng Ma, Zhengyang Zhou, Pengkun Wang, Xu Wang, Binwu Wang, Yang Wang

分数：6, 6, 4, 6

信心：3, 4, 4, 4

均分：5.5

4 Reliable Probabilistic Forecasting of Irregular Time Series through Marginalization-Consistent Flows

链接：https://openreview.net/forum?id=awWi4hJI7O

关键词：Irregular Time Series, Probabilistic Forecasting, Normalizing Flows

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

分数：6, 4, 6, 6

信心：3, 3, 4, 3

均分：5.5

TL; DR: Propose MOSES—a mixture of separable flows over Gaussian processes—that guarantees marginalization consistency while achieving strong predictive performance for probabilistic irregular time series forecasting.

5 Numerion: A Multi-Hypercomplex Model for Time Series Forecasting

链接：https://openreview.net/forum?id=2hCwLqx5gm

关键词：Time Series Forecasting, Hypercomplex Numbers, Hypercomplex Time Series Models, Multi-Hypercomplex Space

作者：Hanzhong Cao, WenBo Yan, Ying Tan

分数：4, 6, 4, 8

信心：5, 4, 4, 4

均分：5.5

TL; DR：We propose Numerion, a hypercomplex space-based model, decomposes and forecasts time series using multi-dimensional RHR-MLPs, achieving state-of-the-art results.

6 Conditionally Whitened Generative Models for Probabilistic Time Series Forecasting

链接：https://openreview.net/forum?id=GG01lCopSK

关键词：Diffusion Model, Probabilistic Time Series Forecasting, Conditional Generation

作者：Yanfeng Yang, Siwei Chen, Pingping Hu, Zhaotong Shen, Yingjie Zhang, Zhuoran Sun, Shuai Li, Ziqi Chen, Kenji Fukumizu

分数：6, 6, 6, 6

信心：2, 4, 4, 3

均分：6.0

TL; DR：A novel class of generative models for probabilistic time series forecasting.

7 Delta-XAI: A Unified Framework for Explaining Prediction Changes in Online Time Series Monitoring

链接：https://openreview.net/forum?id=ZHW5pp5nE5

关键词：Time Series, Online Time Series Monitoring, Explainable Artificial Intelligence, XAI

作者：Changhun Kim, Yechan Mun, Hyeongwon Jang, Eunseo Lee, Sangchul Hahn, Eunho Yang

分数：6, 6, 4, 6

信心：3, 3, 4, 4

均分：5.5

8 COSA: Context-aware Output-Space Adapter for Test-Time Adaptation in Time Series Forecasting

链接：https://openreview.net/forum?id=L7Z5wBMPrW

关键词：Test-Time Adaptation, Time-Series Forecasting, Simple Adapter

作者：Jeonghwan Im, Hyuk-Yoon Kwon

分数：6, 8, 4, 6

信心：4, 4, 4, 3

均分：6.0

9 CPiRi: Channel Permutation-Invariant Relational Interaction for Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=tgnXCCjKE3

关键词：Multivariate Time Series Forecasting, Channel Permutation Invariance, Spatio-temporal Decoupling, Meta-Learning, Foundation Models

作者：Jiyuan Xu, Wenyu Zhang, Xin Jing, Jiahao Nie, Shuai Chen, Shuai Zhang

分数：6, 6, 4, 6

信心：4, 4, 4, 5

均分：5.5

TL; DR：CPiRi enables channel-permutation-invariant MTSF by combining frozen temporal encoding with lightweight spatial attention trained via channel shuffling, achieving SOTA accuracy with zero performance drop under dynamic sensor changes.

10 Enhancing Multivariate Time Series Forecasting with Global Temporal Retrieval

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

关键词：Time-series forecasting, model plugins

作者：Fanpu Cao, Lu Dai, Jindong Han, Hui Xiong

分数：8, 4, 4, 6

信心：3, 3, 4, 4

均分：5.5

TL; DR：A lightweight, model-agnostic plug-and-play module for time-series forecasting models.

11 Bridging Past and Future: Distribution-Aware Alignment for Time Series Forecasting

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

关键词：Time Series Forecasting, Representation Learning, Alignment

作者：Yifan Hu, Jie Yang, Tian Zhou, Peiyuan Liu, Yujin Tang, Rong Jin, Liang Sun

分数：6, 2, 6, 8

信心：3, 4, 4, 5

均分：5.5

12 Are Global Dependencies Necessary? Scalable Time Series Forecasting via Local Cross-Variate Modeling

链接：https://openreview.net/forum?id=CNVL194fO5

关键词：Time Series Forecasting, Time Series Analysis, Deep Learning

作者：Kun Liu, Renjun Jia, Ruifeng Yang, Xirui Zeng, Yuqi Liang, Cen Chen

分数：4, 6, 6

信心：4, 3, 4

均分：5.333333333

TL; DR： This work shows that local cross-variate dependency capturing is effective for dense time series and introduces VPNet, which reinterprets patch embeddings as a variate–patch 2D field to enable accurate, scalable forecasting with linear complexity.

13 ResCP: Reservoir Conformal Prediction for Time Series Forecasting

链接：https://openreview.net/forum?id=WGqibe5H3W

关键词：Conformal prediction, Time series, Uncertainty quantification

作者：Roberto Neglia, Andrea Cini, Michael Bronstein, Filippo Maria Bianchi

分数：2, 6, 4, 8

信心：4, 3, 3, 4

均分：5.0

14 GCGNet: Graph-Consistent Generative Network for Time Series Forecasting with Exogenous Variables

链接：https://openreview.net/forum?id=EO5jwQ5NCw

关键词：time series forecasting

作者：Zhengyu Li, Xiangfei Qiu, Yuhan Zhu, Xingjian Wu, Jilin Hu, Guo, Bin Yang

分数：2, 6, 6, 8

信心：4, 5, 4, 5

均分：5.5

15 Test-Time Efficient Pretrained Model Portfolios for Time Series Forecasting

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

关键词：pretrained time series models, time series forecasting, foundation model combination

作者：Mert Kayaalp, Ali Caner Turkmen, Oleksandr Shchur, Pedro Mercado, Abdul Fatir Ansari, Michael Bohlke-Schneider, Bernie Wang

分数：8, 2, 4, 6

信心：4, 4, 4, 3

均分：5.0

16 MMPD: Diverse Time Series Forecasting via Multi-Mode Patch Diffusion Loss

链接：https://openreview.net/forum?id=NEUgHT8dvH

关键词：time series forecasting, loss function

作者：Yunhao Zhang, Wenyao Hu, Jiale Zheng, Lujia Pan, Junchi Yan

分数：6, 6, 8, 6

信心：3, 3, 2, 4

均分：6.5

TL; DR：We propose the MMPD loss for patch-based time series forecasting backbones to model complex future distributions, enabling them to generate multiple diverse predictions with corresponding probabilities.

17 Local Geometry Attention for Time Series Forecasting under Realistic Corruptions

链接：https://openreview.net/forum?id=NCQPCxN7ds

关键词：Local Geometry, Local Gaussian Process, Transformer Architecture, Time Series Analysis, Corruption Benchmark

作者：Dongbin Kim, Youngjoo Park, Woojin Jeong, Jaewook Lee

分数：4, 6, 6, 4

信心：2, 4, 4, 4

均分：5.0

TL; DR：We propose LGA, the local geometry-aware attention mechanism based on the local Gaussian Process, and introduce TSRBench, the first benchmark for evaluating time series forecasting model robustness under realistic corruptions.

18 FACT: Fine-grained Across-variable Convolution for Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=j3gNYqrHtl

关键词：Multivariate time series forecasting, Fine-grained dynamic variable interactions, Multi-dilated depth-wise convolution.

作者：Huiqiang Wang, Jieming Shi, Qing Li

分数：4, 6, 6, 4

信心：3, 4, 4, 4

均分：5.0

TL; DR：Effective variable interactions modeling from both time and frequency domains; Efficient multi-dilated depth-wise convolution architecture.

19 Unlocking the Value of Text: Event-Driven Reasoning and Multi-Level Alignment for Time Series Forecasting

链接：https://openreview.net/forum?id=0TAFiyHgEl

关键词：time series forecasting, multimodal

作者：Siyuan Wang, Peng Chen, Yihang Wang, Wanghui Qiu, Yang Shu, Guo, Bin Yang

分数：6, 6, 4, 4

信心：4, 3, 4, 3

均分：5.0

20 Aurora: Towards Universal Generative Multimodal Time Series Forecasting

链接：https://openreview.net/forum?id=VVJ6Ck9JBl

关键词：Time Series Forecasting, Multimodality

作者：Xingjian Wu, Jianxin Jin, Wanghui Qiu, Peng Chen, Yang Shu, Bin Yang, Guo

分数：4, 6, 8, 6

信心：4, 4, 5, 4

均分：6.0

21 TEDM: Time Series Forecasting with Elucidated Diffusion Models

链接：https://openreview.net/forum?id=kQee8MObMc

关键词：Score-based generative models, Diffusion models, Stochastic Differential Equations, Time-series forecasting

作者：Edgardo Solano Carrillo, Sreerag Vadakkemeppully Naveenachandran, Julia Niebling

分数：4, 4, 6, 6

信心：2, 4, 3, 4

均分：5.0

22 The Forecast After the Forecast: A Post-Processing Shift in Time Series

链接：https://openreview.net/forum?id=syfWdclGE1

关键词：Time Series Forecasting, Post-Processing, Fine-Tuning

作者：Daojun Liang, Qi Li, Yinglong Wang, Jing Chen, Hu Zhang, Xiaoxiao Cui, Qizheng Wang, Shuo Li

分数：4, 6, 8, 6

信心：4, 3, 4, 4

均分：6.0

TL; DR：We propose post-hoc, a lightweight, architecture-agnostic way to boost deployed time series forecasters without retraining.

23 TAMMs: Change Understanding and Forecasting in Satellite Image Time Series with Temporal-Aware Multimodal Models

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

关键词：Remote Sensing, Satellite Image Time Series, Temporal Reasoning, Generative models, Change-aware Generation, Multimodal Large Language Models

作者：Zhongbin Guo · Yuhao Wang · Ping Jian · Chengzhi Li · Xinyue Chen · Zhen Yang · Ertai E

分数：4, 6, 4, 6

信心：2, 3, 4, 4

均分：5.0

TL; DR：We introduce TAMMs, a unified framework which provides a single solution to both describe historical changes and forecast future scenes, significantly outperforming prior methods on both tasks.

24 Online time series prediction using feature adjustment

链接：https://openreview.net/forum?id=s4U2FWEMTU

关键词：time series, neural network, online adaption

作者：Xiannan Huang, Shuhan Qiu, Jiayuan Du, Chao Yang

分数：4, 4, 6, 4

信心：3, 4, 5, 4

均分：4.5

TL; DR：We propose ADAPT-Z, a new online learning method for time series forecasting that updates feature representations to handle distribution shifts.

25 DoFlow: Flow-based Generative Models for Interventional and Counterfactual Forecasting on Time Series

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

关键词：Time Series, Causal Inference, Generative Models, Flow Matching

作者：Dongze Wu, Feng Qiu, Yao Xie

分数：6, 2, 6, 4

信心：3, 5, 2, 2

均分：4.5

26 Towards Robust Real-World Multivariate Time Series Forecasting: A Unified Framework for Dependency, Asynchrony, and Missingness

链接：https://openreview.net/forum?id=r4ZamwBE8P

关键词：Time Series Forecasting, Channel Dependence, Asynchronous Sampling, Missing Blocks

作者：Jinkwan Jang, Hyungjin Park, Jinmyeong Choi, Taesup Kim

分数：4, 2, 6, 4

信心：4, 5, 4, 3

均分：4.0

TL; DR：To address a trio of real-world challenges involving inter-channel dependencies, asynchronous sampling, and test-time missing blocks, we introduce ChannelTokenFormer that achieves robust performance without requiring explicit temporal alignment.

27 Routing Channel-Patch Dependencies in Time Series Forecasting with Graph Spectral Decomposition

链接：https://openreview.net/forum?id=uIPAuyno4Z

关键词：Time Series Forecasting, Channel Dependency, Graph Learning

作者：Dongyuan Li, Shun Zheng, Chang XU, Jiang Bian, Renhe Jiang

分数：4, 4, 4, 4

信心：4, 5, 3, 5

均分：4.0

TL; DR：We propose xCPD, a lightweight plugin that models channel-patch dependencies via spectral decomposition and routing, enabling accurate and generalizable forecasting with minimal overhead.

28 Flow-based Conformal Prediction for Multi-dimensional Time Series

链接：https://openreview.net/forum?id=Uv3efQiPBZ

关键词：Conformal Prediction, Time Series Prediction

作者：Junghwan Lee, Chen Xu, Yao Xie

分数：0, 4, 6, 6

信心：4, 3, 3, 4

均分：4.0

TL; DR：We propose a novel conformal prediction method for time series using flow with classifier-free guidance.

29 ASTGI: Adaptive Spatio-Temporal Graph Interactions for Irregular Multivariate Time Series Forecasting

链接：https://openreview.net/forum?id=Wg9Rx5rjgo

关键词：Irregular Multivariate Time Series, Time Series Forecasting, Dynamic Graph Neural Networks, Spatio-Temporal Modeling, Data-Driven Interaction

作者：Xvyuan Liu, Xiangfei Qiu, Hanyin Cheng, Xingjian Wu, Guo, Bin Yang, Jilin Hu

分数：2, 4, 6

信心：3, 4, 4

均分：4.0

30 ProtoTS: Learning Hierarchical Prototypes for Explainable Time Series Forecasting

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

关键词：Time series forecasting; Interpretability

作者：Ziheng Peng, Shijie Ren, Xinyue Gu, Linxiao Yang, Xiting Wang, Liang Sun

分数：4, 4, 4, 4

信心：3, 3, 4, 3

均分：4.0

31 PhaseFormer: From Patches to Phases for Efficient and Effective Time Series Forecasting

链接：https://openreview.net/forum?id=Lk9SqMQzhX

关键词：time series forecasting, nonstationary, efficiency

作者：Yiming Niu, Jinliang Deng, Yongxin Tong

分数：2, 6, 6, 4

信心：4, 4, 4, 4

均分：4.5

TL; DR：PhaseFormer replaces patch-based inefficiency with a phase-driven approach, achieving efficient and robust time series forecasting.

32 Learning Recursive Multi-Scale Representations for Irregular Multivariate Time Series Forecasting

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

关键词：Irregular Multivariate Time Series, Time Series Forecasting, Multi-Scale Learning

作者：Boyuan Li, Zhen Liu, Yicheng Luo, Qianli Ma

分数：6, 6, 4, 6

信心：4, 4, 3, 4

均分：5.5

TL; DR：We propose a recursive multi-scale modeling approach that preserves sampling patterns for irregular multivariate time series forecasting task, which boosts performance of existing models while maintaining good efficiency.

33 Characteristic Root Analysis and Regularization for Linear Time Series Forecasting

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

关键词：long term time series forecasting, linear model, characteristic roots, modes, noise robustness, rank reduction, root purge

作者：Zheng Wang, Kaixuan Zhang, Wanfang Chen, Xiaonan Lu, Longyuan Li, Tobias Schlagenhauf

分数：4, 8, 6

信心：4, 2, 4

均分：6.0

34 Semantic-Enhanced Time-Series Forecasting via Large Language Models

链接：https://openreview.net/forum?id=GZ9uSxY3Yn

关键词：Large Language Models; Time Series Forecasting; Semantic Ehanced; Time-Adapter

作者：Hao Liu, Zhang xiaoxing, Chun Yang, Xiaobin Zhu

分数：4, 6, 4, 4

信心：4, 4, 5, 4

均分：4.5

35 TimeSeg: An Information-Theoretic Segment-Wise Explainer for Time-Series Predictions

链接：https://openreview.net/forum?id=alt9mSWULk

关键词：Explainability AI, Interpretability, Time Series Explanations, Segment-wise Explanations, Conditional Mutual Information

作者：Hwijin Kim, Jaeho Kim, Changhee Lee

分数：4, 6, 6

信心：4, 5, 2

均分：5.3

36 TimeRecipe: A Time-Series Forecasting Recipe via Benchmarking Module Level Effectiveness

链接：https://openreview.net/forum?id=CsoR8ztROC

关键词：Time-Series Forecasting, Module Effectiveness, Benchmark

作者：Zhiyuan Zhao, Juntong Ni, Shangqing Xu, Haoxin Liu, Wei Jin, B. Aditya Prakash

分数：4, 2, 6, 8

信心：5, 4, 4, 4

均分：5.0

37 Tackling Time-Series Forecasting Generalization via Mitigating Concept Drift

链接：https://openreview.net/forum?id=emkvZ7NanK

关键词：Time-Series Forecasting, Distribution Shift, Concept Drift

作者：Zhiyuan Zhao, Haoxin Liu, B. Aditya Prakash

分数：6, 6, 6

信心：4, 3, 4

均分：6.0

38 Quadratic Direct Forecast for Training Multi-Step Time-Series Forecast Models

链接：https://openreview.net/forum?id=vpO8n9AqEG

关键词：Time-series, time-series forecast

作者：Eric Wang, Licheng Pan,Yuan Lu, Zi Chan, Tianqiao Liu, Shuting He, Zhixuan Chu, Qingsong Wen, Haoxuan Li, Zhouchen Lin

分数：6, 6, 6

信心：3, 4, 2

均分：6.0

39 DistDF: Time-series Forecasting Needs Joint-distribution Wasserstein Alignment

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

关键词：Large Language Models; Time Series Forecasting; Semantic Ehanced; Time-Adapter

作者：Eric Wang, Licheng Pan, Yuan Lu, Zhixuan Chu, Xiaoxi Li, Shuting He, Zi Chan, Haoxuan Li, Qingsong Wen, Zhouchen Lin

分数：6, 8, 4, 6

信心：4, 3, 4, 3

均分：6.0