Search papers, labs, and topics across Lattice.
The paper introduces CGSTA, a novel framework for multivariate time-series anomaly detection that addresses the challenges of evolving inter-variable dependencies and noise by constructing dynamic layered graphs representing local, regional, and global views of variable relationships. CGSTA employs Contrastive Discrimination across Scales (CDS) to contrast graph representations within each view and align the same window across views, enhancing structure-aware learning. A Stability-Aware Alignment (SAA) mechanism further suppresses noise by guiding fast-changing graphs toward a stable reference learned from normal data, leading to improved anomaly detection performance.
Stop chasing noisy dynamic graphs: CGSTA anchors time-series anomaly detection with a stability-aware alignment to suppress noise and improve performance.
Multivariate time-series anomaly detection is essential for reliable industrial control, telemetry, and service monitoring. However, the evolving inter-variable dependencies and inevitable noise render it challenging. Existing methods often use single-scale graphs or instance-level contrast. Moreover, learned dynamic graphs can overfit noise without a stable anchor, causing false alarms or misses. To address these challenges, we propose the CGSTA framework with two key innovations. First, Dynamic Layered Graph Construction (DLGC) forms local, regional, and global views of variable relations for each sliding window; rather than contrasting whole windows, Contrastive Discrimination across Scales (CDS) contrasts graph representations within each view and aligns the same window across views to make learning structure-aware. Second, Stability-Aware Alignment (SAA) maintains a per-scale stable reference learned from normal data and guides the current window's fast-changing graphs toward it to suppress noise. We fuse the multi-scale and temporal features and use a conditional density estimator to produce per-time-step anomaly scores. Across four benchmarks, CGSTA delivers optimal performance on PSM and WADI, and is comparable to the baseline methods on SWaT and SMAP.
