Multi Class Anomaly Detection

Multi-class anomaly detection aims to identify and classify different types of anomalies within a dataset containing multiple classes of normal data, without prior knowledge of anomaly types. Current research focuses on developing unified models, avoiding the need for separate models per class, often employing techniques like normalizing flows, isolation forests, diffusion models, and transformers to achieve this. These advancements are significant for various applications, including industrial quality control, astronomical transient detection, and network intrusion detection, by enabling more efficient and scalable anomaly detection systems.