New adversarial purification method enhances DNN robustness

Researchers have developed a new method called Consistency Model-based Adversarial Purification (CMAP) to defend deep neural networks against adversarial attacks. CMAP optimizes vectors within the latent space of a pre-trained consistency model to restore clean data from perturbed samples. The approach incorporates a perceptual consistency restoration mechanism, a latent distribution consistency constraint, and a latent vector consistency prediction scheme to enhance robustness and preserve data integrity. Experiments on CIFAR-10 and ImageNet-100 datasets demonstrate CMAP's effectiveness in improving robustness against strong adversarial attacks while maintaining high natural accuracy. AI

IMPACT This method could improve the reliability and security of AI models in real-world applications by defending against adversarial attacks.

RANK_REASON The cluster contains a research paper detailing a new method for improving AI model robustness. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• CIFAR-10
• Consistency Model-based Adversarial Purification
• Deep Neural Networks
• ImageNet-100
• Mingkui Tan

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →