Keep It Real: Challenges in Attacking Compression-Based Adversarial Purification

A recent study published on arXiv explores the effectiveness of using lossy compression as a defense against adversarial attacks on images. While previous research hinted at its potential, this paper rigorously evaluates various compression models and highlights a significant challenge for attackers: achieving high realism in reconstructed images makes it much harder to execute successful attacks. This research is important as it sheds light on the complexities of defending against adversarial perturbations, which is crucial for enhancing the security of machine learning systems.