Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
While temporal anomalies range from sub-second spikes to multi-hour drifts, conventional time-series anomaly detection architectures typically rely on single granularities or restrictive, coarse-to-fine multi-scale hierarchies. To eliminate these representational bottlenecks, MSCAD processes inputs across parallel autoencoder streams of varying patch sizes and enables exhaustive, non-hierarchical scale interaction via symmetric bidirectional cross-scale attention. On the comprehensive 40-dataset TSB-AD benchmark against 50 baselines, MSCAD establishes a new state of the art, boosting VUS-PR by over 9% across both univariate (0.57) and multivariate (0.47) time series.
Rigid coarse-to-fine hierarchies cripple multi-scale anomaly detection: letting arbitrary patch granularities interact symmetrically via bidirectional cross-scale attention yields a 9%+ VUS-PR leap over 50 baselines across 530 time series benchmarks.
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.