COMET: Codebook-based Online-adaptive Multi-scale Embedding for Time-series Anomaly Detection

TL;DR

COMET introduces a multi-scale, codebook-based approach for online adaptive time-series anomaly detection, effectively capturing complex dependencies and handling distribution shifts, leading to superior performance across multiple benchmarks.

Contribution

The paper presents a novel multi-scale patch encoding and online codebook adaptation method that improves anomaly detection in multivariate time series.

Findings

Achieves state-of-the-art results on five benchmark datasets.

Outperforms existing methods in 36 out of 45 evaluation metrics.

Effectively adapts to distribution shifts during inference.

Abstract

Time series anomaly detection is a critical task across various industrial domains. However, capturing temporal dependencies and multivariate correlations within patch-level representation learning remains underexplored, and reliance on single-scale patterns limits the detection of anomalies across different temporal ranges. Furthermore, focusing on normal data representations makes models vulnerable to distribution shifts at inference time. To address these limitations, we propose Codebook-based Online-adaptive Multi-scale Embedding for Time-series anomaly detection (COMET), which consists of three key components: (1) Multi-scale Patch Encoding captures temporal dependencies and inter-variable correlations across multiple patch scales. (2) Vector-Quantized Coreset learns representative normal patterns via codebook and detects anomalies with a dual-score combining quantization error and memory distance. (3) Online Codebook Adaptation generates pseudo-labels based on codebook entries and dynamically adapts the model at inference through contrastive learning. Experiments on five benchmark datasets demonstrate that COMET achieves the best performance in 36 out of 45 evaluation metrics, validating its effectiveness across diverse environments.