DBLP: Noise Bridge Consistency Distillation For Efficient And Reliable Adversarial Purification
This addresses the critical vulnerability of DNNs to adversarial attacks, offering an efficient solution for real-time deployment, though it appears incremental as it builds on existing diffusion methods.
The paper tackles the problem of adversarial purification in deep neural networks by proposing DBLP, a diffusion-based framework that achieves state-of-the-art robust accuracy and image quality with around 0.2s inference time.
Recent advances in deep neural networks (DNNs) have led to remarkable success across a wide range of tasks. However, their susceptibility to adversarial perturbations remains a critical vulnerability. Existing diffusion-based adversarial purification methods often require intensive iterative denoising, severely limiting their practical deployment. In this paper, we propose Diffusion Bridge Distillation for Purification (DBLP), a novel and efficient diffusion-based framework for adversarial purification. Central to our approach is a new objective, noise bridge distillation, which constructs a principled alignment between the adversarial noise distribution and the clean data distribution within a latent consistency model (LCM). To further enhance semantic fidelity, we introduce adaptive semantic enhancement, which fuses multi-scale pyramid edge maps as conditioning input to guide the purification process. Extensive experiments across multiple datasets demonstrate that DBLP achieves state-of-the-art (SOTA) robust accuracy, superior image quality, and around 0.2s inference time, marking a significant step toward real-time adversarial purification.