Assistive-GAN Based Adversarial Learning and Defence for Black-box And White-box Attacks

Abstract

This research paper addresses the ongoing challenge of adversarial attacks in machine learning security by introducing an Assistive-GAN framework tailored to enhance adversarial learning and defence mechanisms against black-box and white-box attacks. The framework is designed to integrate seamlessly with existing defence strategies, augmenting model resilience while maintaining performance metrics. Utilizing a dual-phase training process, the Assistive-GAN generates assistive samples strategically to reinforce the model's ability to identify and withstand adversarial perturbations. Through comprehensive experiments evaluating diverse datasets and attack scenarios, including black-box and white-box attacks, the framework demonstrates significant improvements in model robustness and accuracy compared to state-of-the-art techniques. This research highlights the potential of the Assistive-GAN framework as an effective proactive defence mechanism in bolstering machine learning security against adversarial threats, contributing valuable insights to the cybersecurity domain.