Anomaly Detection
Overview
Anomaly detection studies how to identify observations or events that depart from expected behavior, often when examples of real anomalies are scarce or unknown. Common approaches learn normality through density estimation, isolation methods, reconstruction or forecasting, while newer methods use graph structure, multimodal features, synthetic anomalies, and pretrained vision or language models. Work spans settings such as industrial images, sensor time series, network traffic, and transaction graphs, with benchmarks including MVTec AD, SWaT, and DCASE used to assess detection and localization. A central challenge is balancing sensitivity against false alarms as data shift, training sets become contaminated, and models move from curated benchmarks to constrained real-world deployments.
Latest papers
657 papers61 papers in the last four weeks, level with the four weeks before. 0.4% of all new papers.