No Scale Left Behind: Multi-Scale Autoencoder with Bi-directional Attention for Time Series Anomaly Detection

2026-09-29

AI Summary

This study addresses the limitation of existing time series anomaly detection methods, which are often constrained to single granularity or fixed hierarchical levels and thus struggle to capture interactions among multi-scale anomalies. To this end, we propose MSCAD, a semi-supervised framework that employs parallel multi-branch autoencoders to extract multi-scale features. Furthermore, it introduces a symmetric bidirectional cross-scale attention mechanism that eliminates single-scale privilege, enabling equitable interaction and fusion of information across different scales and overcoming the limitations of isolated multi-scale analysis. Evaluated on the TSB-AD benchmark, MSCAD substantially outperforms 50 baseline methods, achieving an improvement of nearly 10% in the VUS-PR metric. These results demonstrate its superiority in multi-scale time series anomaly detection.

Abstract

Time series anomaly detection (TSAD) plays a crucial role in healthcare, finance, industrial monitoring, and other sectors. Within and between these settings, anomalies span vastly different temporal scales, from sub-second point spikes to multi-hour drift patterns. However, most existing TSAD methods commit to a single temporal granularity, and multi-scale designs either analyze different scales in isolation or are constrained to a predefined coarse-to-fine hierarchy, both failing to sufficiently capture multi-scale interactions. To resolve this limitation, we propose Multi-Scale Autoencoder with Cross-Scale Attention for TSAD (MSCAD), a simple yet powerful semi-supervised TSAD framework founded on parallel autoencoder branches corresponding to different patch sizes. A stack of symmetric bidirectional cross-scale attention blocks enables every pair of scales to exchange information before reconstruction without allowing any single scale to be privileged. On the comprehensive TSB-AD benchmark (40 datasets, 530 series), MSCAD achieves large performance gains against 50 baselines across multiple metrics, with VUS-PR of 0.57(+9.6%) on the univariate split and 0.47(+9.3%) on the multivariate split compared to the state-of-the-art.

Problem

• Time Series Anomaly Detection
• Multi-Scale
• Cross-Scale Interaction
• Temporal Granularity

Innovation

• Time Series Anomaly Detection
• Multi-Scale Autoencoder
• Bidirectional Cross-Scale Attention
• Patch-based Representation
• Semi-supervised Learning