Browse Teach Curate

GAN-Based Generative Replay for Continual Malware Classification

A continual-learning approach for malware classification that combats catastrophic forgetting by using a generative adversarial network with a feature-matching loss to synthesize samples of previously seen malware classes, and by selecting which synthetic samples to replay through the classifier's hidden-representation logits (per-class mean logits compared with L1 distance).

Data Text

MalCL: Leveraging GAN-Based Generative Replay to Combat Catastrophic Forgetting in Malware

MalCL is a generative-replay continual-learning system for malware classification that combats catastrophic forgetting. A generative adversarial network trained with a feature-matching loss synthesiz…