Jiyue Jiang, Yitao Xu, Zikang Wang, Yihan Ye, Yanruisheng Shao, Yuheng Shan, Jiuming Wang, Xiaodan Fan, Jiao Yuan, Yu Li. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

Time series anomaly detection (TSAD) is critical for ensuring the reliability of equipment in complex industrial environments. However, existing methods face significant challenges, including imbalanced data, lack of labeled samples, reliance on prior knowledge, poor generalization across diverse industrial scenarios, and low sensitivity to subtle anomalies. To address these limitations, we propose KARMAD, a novel framework that integrates Kolmogorov-Arnold Networks (KANs) for bidirectional function learning, adversarial training to enhance sensitivity to minor anomalies, and an adaptive thresholding strategy for improved precision and transferability. Evaluated on five public datasets against 14 state-of-the-art methods, KARMAD achieves state-of-the-art performance on all datasets, with an average F1 score improvement of 13.14%. Further experiments showcase its robustness to noise and adaptability in real-world industrial applications. KARMAD represents a significant advancement in developing scalable and accurate TSAD models suitable for diverse and high-stakes environments.