Compression as an Adversarial Amplifier Through Decision Space Reduction

TL;DR

This paper reveals that image compression can amplify adversarial attacks on classifiers by reducing decision space, making models more vulnerable in compressed environments.

Contribution

It introduces the concept that compression acts as an adversarial amplifier by shrinking decision margins, and demonstrates this effect through extensive experiments.

Findings

Compression-aware attacks are more effective than pixel-space attacks.

Compression reduces decision margins, increasing sensitivity to perturbations.

A critical vulnerability exists in systems using compression in the inference pipeline.

Abstract

Image compression is a ubiquitous component of modern visual pipelines, routinely applied by social media platforms and resource-constrained systems prior to inference. Despite its prevalence, the impact of compression on adversarial robustness remains poorly understood. We study a previously unexplored adversarial setting in which attacks are applied directly in compressed representations, and show that compression can act as an adversarial amplifier for deep image classifiers. Under identical nominal perturbation budgets, compression-aware attacks are substantially more effective than their pixel-space counterparts. We attribute this effect to decision space reduction, whereby compression induces a non-invertible, information-losing transformation that contracts classification margins and increases sensitivity to perturbations. Extensive experiments across standard benchmarks and architectures support our analysis and reveal a critical vulnerability in compression-in-the-loop deployment settings. Code will be released.