CADIC: Continual Anomaly Detection Based on Incremental Coreset

The introduction of the Continual Anomaly Detection (CAD) framework represents a significant leap in machine learning methodologies aimed at identifying anomalies in dynamic data environments. Unlike traditional methods that rely on class-specific memory banks, this innovative approach utilizes a shared memory bank, allowing for more flexible and scalable anomaly detection. By incrementally updating embeddings within a fixed-size coreset, the framework not only mitigates catastrophic forgetting but also enhances the model's ability to learn from sequential tasks. Comprehensive experiments conducted on the MVTec AD and Visa datasets validate the effectiveness of this method, achieving impressive average image-level AUROC scores of 0.972 and 0.891, respectively. Furthermore, the framework's ability to maintain 100% accuracy in detecting anomaly samples underscores its robustness in practical applications, making it a valuable contribution to the field of artificial intelligence.