An Encode-then-Decompose Approach to Unsupervised Time Series Anomaly Detection on Contaminated Training Data--Extended Version
This addresses the robustness issue in unsupervised time series anomaly detection for applications like system monitoring, though it appears incremental as it builds on autoencoder foundations.
The paper tackles the problem of autoencoder-based time series anomaly detection being sensitive to anomalies in training data, proposing an encode-then-decompose approach that decomposes encoded representations into stable and auxiliary components and uses mutual information instead of reconstruction errors. The method achieves competitive or state-of-the-art performance on eight benchmarks and shows robustness to varying contamination ratios.
Time series anomaly detection is important in modern large-scale systems and is applied in a variety of domains to analyze and monitor the operation of diverse systems. Unsupervised approaches have received widespread interest, as they do not require anomaly labels during training, thus avoiding potentially high costs and having wider applications. Among these, autoencoders have received extensive attention. They use reconstruction errors from compressed representations to define anomaly scores. However, representations learned by autoencoders are sensitive to anomalies in training time series, causing reduced accuracy. We propose a novel encode-then-decompose paradigm, where we decompose the encoded representation into stable and auxiliary representations, thereby enhancing the robustness when training with contaminated time series. In addition, we propose a novel mutual information based metric to replace the reconstruction errors for identifying anomalies. Our proposal demonstrates competitive or state-of-the-art performance on eight commonly used multi- and univariate time series benchmarks and exhibits robustness to time series with different contamination ratios.