NeurIPS 2023 时间序列相关论文总结

祝大家中秋国庆双节快乐！

NeurIPS 2023将于11月28日到12月9日在美国路易斯安那州新奥尔良举行。

根据官方公布的邮件显示，今年共有12343篇投稿，接受率为26.1%，官网显示一共有3564篇论文。

本文总结了NeurIPS 23 时间序列（不含时空数据，已经另外总结）的相关论文。包括时间序列预测，分类，异常检测，因果发现，交通，医疗等领域时间序列应用和大模型在时间序列问题建模的探索等方向。

1. WITRAN: Water-wave Information Transmission and Recurrent Acceleration Network for Long-range Time Series Forecasting （Spotlight）

**作者：**Yuxin Jia, Youfang Lin, Xinyan Hao, Yan Lin, Shengnan Guo, Huaiyu Wan

链接：https://neurips.cc/virtual/2023/poster/69972

**关键词：**长时预测

**代码：**https://github.com/Water2sea/WITRAN

**论文简介：**捕获语义信息对于长程时间序列的准确预测至关重要，其中包括两大方面：（1）建模全局和局部相关性，（2）挖掘长期和短期的重复模式。以往的研究工作能够一定程度上捕获这些方面中的一部分，但无法完成它们的同时捕获。此外，以往研究工作的时间和空间复杂性仍然很高。

基于此，我们提出了一种新颖的水波信息传递（Water-wave Information Transmission，简称WIT）框架，能够通过双粒度的信息传递捕捉短期和长期的重复模式。在WIT框架中，我们设计了一种新型水平垂直门控选择单元（Horizontal Vertical Gated Selective Unit，简称HVGSU），通过循环地融合和选择信息，来建模全局和局部相关性。此外，在提高计算效率方面，我们提出了一种通用的循环加速网络（Recurrent Acceleration Network，简称RAN），能够在保证空间复杂度为O(L)的同时，将时间复杂度降低到为O(√L)。

综上，我们将提出的方法命名为：水波信息传递和循环加速网络（Water-wave Information Transmission and Recurrent Acceleration Network，简称WITRAN）。通过在能源、交通、天气等领域的四个大型公开数据集上的实验证明，相对于现有方法，WITRAN在长程和超长程时间序列预测任务上成为了最佳方法（SOTA）。

Spotlight信息来源北京交通大学网络科学与智能系统研究所公众号：喜报 | 我所两篇论文被机器学习领域顶会NeurIPS 2023接收

2. Sparse Deep Learning for Time Series Data: Theory and Applications

**作者：**Mingxuan Zhang, Yan Sun, Faming Liang

**链接：**https://neurips.cc/virtual/2023/poster/72629

**关键词：**稀疏学习、感觉像是综述

3. Causal Discovery from Subsampled Time Series with Proxy Variables

**作者：**Mingzhou Liu, Xinwei Sun, Lingjing Hu, Yizhou Wang

**链接：**https://neurips.cc/virtual/2023/poster/70936 arXiv：Causal Discovery from Subsampled Time Series with Proxy Variables

关键词：因果发现

4. Causal Discovery in Semi-Stationary Time Series

**作者：**Shanyun Gao, Raghavendra Addanki, Tong Yu, Ryan Rossi, Murat Kocaoglu

**链接：**https://neurips.cc/virtual/2023/poster/71016

**关键词：**因果发现、半平稳时间序列

5. Frequency-domain MLPs are More Effective Learners in Time Series Forecasting

**作者：**Kun Yi, Qi Zhang, Wei Fan, Hui He, Pengyang Wang, Shoujin Wang, Ning An, Defu Lian, Longbing Cao, Zhendong Niu

**链接：**https://neurips.cc/virtual/2023/poster/70726

**关键词：**频域、MLP(疑似又是化繁为简的工作，类似DLinear)

6. FourierGNN: Rethinking Multivariate Time Series Forecasting from a Pure Graph Perspective

**作者：**Kun Yi, Qi Zhang, Wei Fan, Hui He, Liang Hu, Pengyang Wang, Ning An, Longbing Cao, Zhendong Niu

**链接：**https://neurips.cc/virtual/2023/poster/71159

**关键词：**图神经网络、多元时间序列预测

7. CrossGNN: Confronting Noisy Multivariate Time Series Via Cross Interaction Refinement

**作者：**qihe huang, Lei Shen, Ruixin Zhang, Shouhong Ding, Binwu Wang, Zhengyang Zhou, Yang Wang

**链接：**https://neurips.cc/virtual/2023/poster/70010

**关键词：**图神经网络、多元时间序列预测

8. Scale-teaching: Robust Multi-scale Training for Time Series Classification with Noisy Labels

**作者：**Zhen Liu, ma peitian, Dongliang Chen, Wenbin Pei, Qianli Ma

**链接：**https://neurips.cc/virtual/2023/poster/72608

**关键词：**时间序列分类、鲁棒性

9. Time Series as Images: Vision Transformer for Irregularly Sampled Time Series

**作者：**Zekun Li, Shiyang Li, Xifeng Yan

**链接：**https://neurips.cc/virtual/2023/poster/71219 arXiv：https://arxiv.org/abs/2303.12799 （看格式是ICML改投的）

**代码：**https://github.com/Leezekun/ViTST

**关键词：**VIT

10. Adaptive Normalization for Non-stationary Time Series Forecasting: A Temporal Slice Perspective

**作者：**Zhiding Liu, Mingyue Cheng, Zhi Li, Zhenya Huang, Qi Liu, Yanhu Xie, Enhong Chen

**链接：**https://neurips.cc/virtual/2023/poster/72816

关键词：非平稳时间序列预测

11. Koopa: Learning Non-stationary Time Series Dynamics with Koopman Predictors

**作者：**Yong Liu, Chenyu Li, Jianmin Wang, Mingsheng Long

**链接：**https://neurips.cc/virtual/2023/poster/72562 arXiv：https://arxiv.org/abs/2305.18803

**关键词：**非平稳时间序列预测

12. SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling（Spotlight）

**作者：**Jiaxiang Dong, Haixu Wu, Haoran Zhang, Li Zhang, Jianmin Wang, Mingsheng Long

链接： SimMTM: A Simple Pre-Training Framework for Masked Time-Series Modeling arXiv：https://arxiv.org/abs/2302.00861

**关键词：**预训练、统一框架建模

13. WildfireSpreadTS: A dataset of multi-modal time series for wildfire spread prediction

**作者：**Sebastian Gerard, Yu Zhao, Josephine Sullivan

**链接：**https://neurips.cc/virtual/2023/poster/73593

**关键词：**多模态时间序列、数据集

14. Drift doesn’t Matter: Dynamic Decomposition with Diffusion Reconstruction for Unstable Multivariate Time Series Anomaly Detection

**作者：**Chengsen Wang, Qi Qi, Jingyu Wang, Haifeng Sun, Xingyu Wang, Zirui Zhuang, Jianxin Liao

**链接：**https://neurips.cc/virtual/2023/poster/71195

**关键词：**多元时间序列异常检测

15. Nominality Score Conditioned Time Series Anomaly Detection by Point/Sequential Reconstruction

**作者：**Chih-Yu Lai, Fan-Keng Sun, Zhengqi Gao, Jeffrey H Lang, Duane Boning

**链接：**https://neurips.cc/virtual/2023/poster/70582

**关键词：**时间序列异常检测

16. MEMTO: Memory-guided Transformer for Multivariate Time Series Anomaly Detection

**作者：**Junho Song, Keonwoo Kim, Jeonglyul Oh, Sungzoon Cho

**链接：**https://neurips.cc/virtual/2023/poster/71519

**关键词：**时间序列异常检测

17. Conformal Prediction for Time Series with Modern Hopfield Networks

**作者：**Andreas Auer, Martin Gauch, Daniel Klotz, Sepp Hochreiter（LSTM一作）

链接： https://neurips.cc/virtual/2023/poster/72007 arXiv：https://arxiv.org/abs/2303.12783

**关键词：**共性预测、霍普菲尔德网络

18. Conformal Scorecasting: Anticipatory Uncertainty Quantification for Distribution Shift in Time Series

**作者：**Anastasios Angelopoulos, Ryan Tibshirani, Emmanuel Candes

**链接：**https://neurips.cc/virtual/2023/poster/69896

**关键词：**共性预测、不确定性量化

19. Finding Order in Chaos: A Novel Data Augmentation Method for Time Series in Contrastive Learning

**作者：**Berken Utku Demirel, Christian Holz

**链接：**https://neurips.cc/virtual/2023/poster/71014 arXiv：https://arxiv.org/abs/2309.13439

**关键词：**对比学习、数据增强

20. ContiFormer: Continuous-Time Transformer for Irregular Time Series Modeling

**作者：**Yuqi Chen, Kan Ren, Yansen Wang, Yuchen Fang, Weiwei Sun, Dongsheng Li

**链接：**https://neurips.cc/virtual/2023/poster/71304

关键词：（喜闻乐见的）Former改动、不规则时间序列

21. BasisFormer: Attention-based Time Series Forecasting with Learnable and Interpretable Basis

**链接：**https://neurips.cc/virtual/2023/poster/69976

关键词：（喜闻乐见的）Former改动、时间序列预测

22. Time Series Kernels based on Nonlinear Vector AutoRegressive Delay Embeddings

**作者：**Giovanni De Felice, John Goulermas, Vladimir Gusev

**链接：**https://neurips.cc/virtual/2023/poster/71521

**关键词：**核方法

23. Predict, Refine, Synthesize: Self-Guiding Diffusion Models for Probabilistic Time Series Forecasting

**作者：**Marcel Kollovieh, Abdul Fatir Ansari, Michael Bohlke-Schneider, Jasper Zschiegner, Hao Wang, Yuyang (Bernie) Wang

**链接：**https://neurips.cc/virtual/2023/poster/70377 arXiv：https://arxiv.org/abs/2307.11494

**关键词：**扩散模型、时间序列预测

24. OneNet: Enhancing Time Series Forecasting Models under Concept Drift by Online Ensembling

**作者：**yifan zhang, Qingsong Wen, xue wang, Weiqi Chen, Liang Sun, Zhang Zhang, Liang Wang, Rong Jin, Tieniu Tan

**链接：**https://neurips.cc/virtual/2023/poster/71725 arXiv：OneNet: Enhancing Time Series Forecasting Models under Concept Drift by Online Ensembling

**代码：**https://github.com/yfzhang114/OneNet

关键词：时间序列预测、概念漂移

25. One Fits All: Power General Time Series Analysis by Pretrained LM

这篇热度之前应该就很高，在各大平台应该都有针对的解读

**作者：**Tian Zhou（FEDFormer（ICML 22） FilM(NeurIPS 22)一作）, Peisong Niu, xue wang, Liang Sun, Rong Jin

**链接：**https://neurips.cc/virtual/2023/poster/70856 Researchgate

**关键词：**大模型、时间序列统一任务

26. Large Language Models Are Zero Shot Time Series Forecasters

**作者：**Marc Finzi, Nate Gruver, Shikai Qiu, Andrew Wilson

**链接：**https://neurips.cc/virtual/2023/poster/70543

**关键词：**大模型、零样本、时间预测

27. Contrast Everything: Multi-Granularity Representation Learning for Medical Time-Series

**作者：**Yihe Wang, Yu Han, Haishuai Wang, Xiang Zhang

**链接：**https://neurips.cc/virtual/2023/poster/70272

**关键词：**对比学习、医疗时间序列

28. FOCAL: Contrastive Learning for Multimodal Time-Series Sensing Signals in Factorized Orthogonal Latent Space

**作者：**Shengzhong Liu, Tomoyoshi Kimura, Dongxin Liu, Ruijie Wang, Jinyang Li, Suhas Diggavi, Mani Srivastava, Tarek Abdelzaher

**链接：**https://neurips.cc/virtual/2023/poster/70617

**关键词：**对比学习、多模态时间序列

29. BioMassters: A Benchmark Dataset for Forest Biomass Estimation using Multi-modal Satellite Time-series

**作者：**Andrea Nascetti, Ritu Yadav, Kirill Brodt, Qixun Qu, Hongwei Fan, Yuri Shendryk, Isha Shah, Christine Chung

**链接：**https://neurips.cc/virtual/2023/poster/73499

**关键词：**数据集、多模态

30. Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency

**作者：**Owen Queen, Thomas Hartvigsen, Teddy Koker, Huan He, Theodoros Tsiligkaridis, Marinka Zitnik

**链接：**https://neurips.cc/virtual/2023/poster/69958 arXiv：Encoding Time-Series Explanations through Self-Supervised Model Behavior Consistency

**关键词：**可解释性、自监督

31. On the Constrained Time-Series Generation Problem

**作者：**Andrea Coletta, Sriram Gopalakrishnan, Daniel Borrajo, Svitlana Vyetrenko

链接：On the Constrained Time-Series Generation Problem arXiv：On the Constrained Time-Series Generation Problem

**关键词：**时间序列生成、扩散模型

相关链接：

**NeurIPS 2023全部论文列表：**https://neurips.cc/virtual/2023/papers.html