Heat-RL: Online Model Selection for Streaming Time-Series Anomaly Detection

Yujing Wang, Luoxin Xiong, Mingliang Zhang, Hui Xue, Qi Chen, Yaming Yang, Yunhai Tong, Congrui Huang, Bixiong Xu
Proceedings of The 1st Conference on Lifelong Learning Agents, PMLR 199:767-777, 2022.

Abstract

Time-series anomaly detection plays an important role in various applications. In a commercial system, anomaly detection models are either unsupervised or pre-trained in a self-supervised manner offline; while in the online serving stage, an appropriate model should be selected to fulfill each customer’s requirement with only a few human interactions. Existing online model selection methods do not have good data efficiency, failing to achieve good performance with limited number of manual feedbacks. In this paper, we propose Heat-RL, a novel reinforcement learning algorithm tailored to online model selection for streaming time-series data. Specifically, we design a new state based on metric-oriented heatmaps and apply ResNet for policy and value networks to capture the correlations among similar model configurations. Experiments demonstrated the effectiveness of Heat-RL on both academic and industrial datasets. On all datasets, the average F1 and last F1 scores have been improved by 5.5% and 14.6% respectively compared to the best state-of-the-art solution.

Cite this Paper

BibTeX
@InProceedings{pmlr-v199-wang22a, title = {Heat-RL: Online Model Selection for Streaming Time-Series Anomaly Detection}, author = {Wang, Yujing and Xiong, Luoxin and Zhang, Mingliang and Xue, Hui and Chen, Qi and Yang, Yaming and Tong, Yunhai and Huang, Congrui and Xu, Bixiong}, booktitle = {Proceedings of The 1st Conference on Lifelong Learning Agents}, pages = {767--777}, year = {2022}, editor = {Chandar, Sarath and Pascanu, Razvan and Precup, Doina}, volume = {199}, series = {Proceedings of Machine Learning Research}, month = {22--24 Aug}, publisher = {PMLR}, pdf = {https://proceedings.mlr.press/v199/wang22a/wang22a.pdf}, url = {https://proceedings.mlr.press/v199/wang22a.html}, abstract = {Time-series anomaly detection plays an important role in various applications. In a commercial system, anomaly detection models are either unsupervised or pre-trained in a self-supervised manner offline; while in the online serving stage, an appropriate model should be selected to fulfill each customer’s requirement with only a few human interactions. Existing online model selection methods do not have good data efficiency, failing to achieve good performance with limited number of manual feedbacks. In this paper, we propose Heat-RL, a novel reinforcement learning algorithm tailored to online model selection for streaming time-series data. Specifically, we design a new state based on metric-oriented heatmaps and apply ResNet for policy and value networks to capture the correlations among similar model configurations. Experiments demonstrated the effectiveness of Heat-RL on both academic and industrial datasets. On all datasets, the average F1 and last F1 scores have been improved by 5.5% and 14.6% respectively compared to the best state-of-the-art solution.} }
Endnote
%0 Conference Paper %T Heat-RL: Online Model Selection for Streaming Time-Series Anomaly Detection %A Yujing Wang %A Luoxin Xiong %A Mingliang Zhang %A Hui Xue %A Qi Chen %A Yaming Yang %A Yunhai Tong %A Congrui Huang %A Bixiong Xu %B Proceedings of The 1st Conference on Lifelong Learning Agents %C Proceedings of Machine Learning Research %D 2022 %E Sarath Chandar %E Razvan Pascanu %E Doina Precup %F pmlr-v199-wang22a %I PMLR %P 767--777 %U https://proceedings.mlr.press/v199/wang22a.html %V 199 %X Time-series anomaly detection plays an important role in various applications. In a commercial system, anomaly detection models are either unsupervised or pre-trained in a self-supervised manner offline; while in the online serving stage, an appropriate model should be selected to fulfill each customer’s requirement with only a few human interactions. Existing online model selection methods do not have good data efficiency, failing to achieve good performance with limited number of manual feedbacks. In this paper, we propose Heat-RL, a novel reinforcement learning algorithm tailored to online model selection for streaming time-series data. Specifically, we design a new state based on metric-oriented heatmaps and apply ResNet for policy and value networks to capture the correlations among similar model configurations. Experiments demonstrated the effectiveness of Heat-RL on both academic and industrial datasets. On all datasets, the average F1 and last F1 scores have been improved by 5.5% and 14.6% respectively compared to the best state-of-the-art solution.
APA
Wang, Y., Xiong, L., Zhang, M., Xue, H., Chen, Q., Yang, Y., Tong, Y., Huang, C. & Xu, B.. (2022). Heat-RL: Online Model Selection for Streaming Time-Series Anomaly Detection. Proceedings of The 1st Conference on Lifelong Learning Agents, in Proceedings of Machine Learning Research 199:767-777 Available from https://proceedings.mlr.press/v199/wang22a.html.

Related Material