Adversarial Neural Pruning with Latent Vulnerability Suppression

TL;DR

This paper introduces a novel method combining vulnerability suppression and neural pruning to enhance adversarial robustness in deep neural networks, achieving state-of-the-art results while reducing model complexity.

Contribution

It proposes a new vulnerability suppression loss and a Bayesian pruning framework to improve adversarial robustness and efficiency of neural networks.

Findings

Achieves state-of-the-art adversarial robustness on benchmarks.

Improves clean data performance with fewer parameters.

Suppresses feature-level vulnerability effectively.

Abstract

Despite the remarkable performance of deep neural networks on various computer vision tasks, they are known to be susceptible to adversarial perturbations, which makes it challenging to deploy them in real-world safety-critical applications. In this paper, we conjecture that the leading cause of adversarial vulnerability is the distortion in the latent feature space, and provide methods to suppress them effectively. Explicitly, we define \emph{vulnerability} for each latent feature and then propose a new loss for adversarial learning, \emph{Vulnerability Suppression (VS)} loss, that aims to minimize the feature-level vulnerability during training. We further propose a Bayesian framework to prune features with high vulnerability to reduce both vulnerability and loss on adversarial samples. We validate our \emph{Adversarial Neural Pruning with Vulnerability Suppression (ANP-VS)} method on multiple benchmark datasets, on which it not only obtains state-of-the-art adversarial robustness but also improves the performance on clean examples, using only a fraction of the parameters used by the full network. Further qualitative analysis suggests that the improvements come from the suppression of feature-level vulnerability.