TIMESFORMER:A MULTI-REPRESENTATION FRAMEWORK COMBINING CONVOLUTION AND TRANSFORMER FOR MULTIVARIATE TIME SERIES CLASSIFICATION TASKS
Keywords: Time series classification
Abstract: Multivariate time series classification is an important research area in machine learning, which is challenged by leveraging the inherent complexity of temporal, periodic, and multivariate dependencies.Classification models need to effectively extract features hidden within these dependencies, Such as the CNN-based model TimesNet, which effectively encodes periodic dependencies, performs well in classification tasks. However, its lack of effective encoding for multivariate dependencies limits its performance in classifying high-dimensional time series data. Additionally, due to the constraints of receptive fields, CNNs struggle to capture long-term dependencies in excessively long sequences. To address these issues, we propose a multi-representation framework that applies both point-wise and patch-wise temporal representations to multivariate time series data. This allows the model to leverage CNNs for capturing periodic dependencies while using transformer to capture global features across time and variable dimensions. In this way, we overcome the CNN's limitations in capturing long-term dependencies and integrate cross-variable dependencies into the model. Experimental results show that our approach achieves higher average accuracy than 17 of the latest time series classification models across 10 EUA classification datasets.
Primary Area: learning on time series and dynamical systems
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 8529
Loading