PaAno+: Multiscale Encoding and Cross-Variable Attention for Time Series Anomaly Detection
Researchers have developed PaAno, a novel lightweight model for time-series anomaly detection designed for efficiency and accuracy. The model utilizes a multiscale feature-extraction backbone with convolutional kernels and cross-scale attention to capture hierarchical temporal characteristics. It also incorporates a cross-variable fusion attention module to model inter-variable correlations and a pretext task based on temporal patch-window sorting to enhance feature discrimination. Experiments on the TSB-AD benchmark show PaAno achieving state-of-the-art detection accuracy on both univariate and multivariate tasks with favorable computational efficiency for real-time inference. AI
IMPACT Offers a more efficient and accurate solution for real-time anomaly detection in critical domains.