Abstract: We propose a method for learning a classifier that accurately predicts both instance and bag classes in multiple instance learning (MIL) for anomaly detection, achieving significant ...