Abstract: The growth of machine learning (ML) has revealed model vulnerabilities to adversarial attacks, where small data perturbations degrade performance. Classical defenses often struggle, ...