ConGANomaly: A Contrastive Learning Approach Of Anomaly Detection Using Generative Adversarial Networks
Shikhar Asthana, Anurag Goel · 2024
Anomaly detection (AD) is essential for numerous machine learning applications across a wide range of fields. Semi-supervised AD involves identifying anomalies in data with limited labeled examples, using labeled and unlabeled data for training the model. This research endeavors to enhance the effectiveness of semi-supervised AD models, namely GANomaly and Skip-GANomaly, by adding novel changes to the model so as to incorporate contrastive learning into their architecture. In contrastive learning model learns to distinguish between positive and negative sample pairs to develop robust representations. The performances of the modified architecture were measured across-different datasets and yielded positive improvements in model performance.