E2USD: Efficient-yet-effective Unsupervised State Detection for Multivariate Time Series
WWW '24 Proceedings of the ACM on Web Conference 2024(2024)
Abstract
We propose E2USD that enables efficient-yet-accurate unsupervised MTS statedetection. E2USD exploits a Fast Fourier Transform-based Time Series Compressor(FFTCompress) and a Decomposed Dual-view Embedding Module (DDEM) that togetherencode input MTSs at low computational overhead. Additionally, we propose aFalse Negative Cancellation Contrastive Learning method (FNCCLearning) tocounteract the effects of false negatives and to achieve more cluster-friendlyembedding spaces. To reduce computational overhead further in streamingsettings, we introduce Adaptive Threshold Detection (ADATD). Comprehensiveexperiments with six baselines and six datasets offer evidence that E2USD iscapable of SOTA accuracy at significantly reduced computational overhead. Ourcode is available at https://github.com/AI4CTS/E2Usd.
MoreTranslated text
AI Read Science
Must-Reading Tree
Example
![](https://originalfileserver.aminer.cn/sys/aminer/pubs/mrt_preview.jpeg)
Generate MRT to find the research sequence of this paper
Chat Paper
Summary is being generated by the instructions you defined