Adversarial Robustness of Bottleneck Injected Deep Neural Networks for Task-Oriented Communication

TL;DR

This paper evaluates the adversarial robustness of information bottleneck-based deep neural networks in task-oriented communication, revealing vulnerabilities and robustness patterns related to bottleneck depth and task complexity.

Contribution

It provides empirical analysis of how IB-based models' robustness varies with architecture and task complexity, highlighting security implications for neural network-based communication systems.

Findings

Shallow IB models are less robust than deep IB models.

Robustness varies with task complexity and bottleneck depth.

IB-based systems are more resilient to high-intensity pixel attacks.

Abstract

This paper investigates the adversarial robustness of Deep Neural Networks (DNNs) using Information Bottleneck (IB) objectives for task-oriented communication systems. We empirically demonstrate that while IB-based approaches provide baseline resilience against attacks targeting downstream tasks, the reliance on generative models for task-oriented communication introduces new vulnerabilities. Through extensive experiments on several datasets, we analyze how bottleneck depth and task complexity influence adversarial robustness. Our key findings show that Shallow Variational Bottleneck Injection (SVBI) provides less adversarial robustness compared to Deep Variational Information Bottleneck (DVIB) approaches, with the gap widening for more complex tasks. Additionally, we reveal that IB-based objectives exhibit stronger robustness against attacks focusing on salient pixels with high intensity compared to those perturbing many pixels with lower intensity. Lastly, we demonstrate that task-oriented communication systems that rely on generative models to extract and recover salient information have an increased attack surface. The results highlight important security considerations for next-generation communication systems that leverage neural networks for goal-oriented compression.