AnomalyTCN: Dual-branch Convolution with Contrastive Representation for Efficient Time Series Anomaly Detection
Keywords: Time series anomaly detection, Deep learning, Contrastive-based time series anomaly detection
TL;DR: We combine the pure convolution structure with contrastive discrepancy learning and propose a novel solution for efficient time series anomaly detection, which achieves the consistent state-of-the-art on various datasets with much better efficiency.
Abstract: This paper focuses on the rising contrastive-based method for time series anomaly detection, which works on the idea of contrastive discrepancy learning and breaks through the performance bottleneck of previous reconstruction-based methods. But we also find that, existing contrastive-based methods only work with the complicated attention mechanisms, which brings heavier computational costs. To address this efficiency issue, we propose AnomalyTCN as a more efficient and effective contrastive-based solution. In detail, we design a dual-branch convolution structure to produce different representations of the same input under two different views for contrastive learning. Then we adopt the representation discrepancy between these two branches as a more distinguishable criterion to detect the anomalies, leading to better detection performance. Meanwhile, since we adopt a simple and light-weight pure convolution structure to avoid the complicated attention computation, our method can enjoy much more advantages in efficiency. Experimentally, our AnomalyTCN achieves the consistent state-of-the-art performance on various time series anomaly detection tasks while saving 83.6\% running time and 20.1\% memory usage. These results validate that our AnomalyTCN is a novel solution for time series anomaly detection with a better balance of performance and efficiency.
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: 2675
Loading