Model Compression with Adversarial Robustness: A Unified Optimization Framework

TL;DR

This paper introduces a unified framework for compressing deep models that preserves adversarial robustness, integrating various compression techniques into a constrained optimization problem and demonstrating improved trade-offs between size, accuracy, and robustness.

Contribution

The paper proposes ATMC, a novel unified optimization framework that combines multiple compression methods with adversarial robustness preservation.

Findings

ATMC achieves better trade-offs among size, accuracy, and robustness.

Experimental results show ATMC outperforms existing methods.

Codes are publicly available for reproducibility.

Abstract

Deep model compression has been extensively studied, and state-of-the-art methods can now achieve high compression ratios with minimal accuracy loss. This paper studies model compression through a different lens: could we compress models without hurting their robustness to adversarial attacks, in addition to maintaining accuracy? Previous literature suggested that the goals of robustness and compactness might sometimes contradict. We propose a novel Adversarially Trained Model Compression (ATMC) framework. ATMC constructs a unified constrained optimization formulation, where existing compression means (pruning, factorization, quantization) are all integrated into the constraints. An efficient algorithm is then developed. An extensive group of experiments are presented, demonstrating that ATMC obtains remarkably more favorable trade-off among model size, accuracy and robustness, over currently available alternatives in various settings. The codes are publicly available at: https://github.com/shupenggui/ATMC.