Understanding Fault Tolerance of Adversarially Robust Pruned Models

2026-08-04

AI Summary

This study addresses the coupled challenges of model pruning, adversarial perturbations, and stuck-at-zero hardware weight faults in deep neural networks deployed on resource-constrained neuromorphic hardware, where their joint impact remains poorly understood. Focusing on a three-layer CNN trained on MNIST, the work presents a systematic multidimensional evaluation integrating pruning, adversarial training, and hardware fault injection. It reveals, for the first time, that while adversarial training enhances robustness against attacks such as FGSM, it substantially increases sensitivity to stuck-at-zero faults. In contrast, pruning exhibits limited influence on fault tolerance and demonstrates consistent performance across varying fault rates and attack intensities, challenging prevailing assumptions in the field.

Abstract

Deep neural networks (DNNs) deployed on resource-constrained neuromorphic hardware face three concurrent challenges: the need for model compression through pruning, vulnerability to adversarial input perturbations, and susceptibility to hardware-induced weight faults such as stuck-at-zero errors. While each of these factors has been studied in isolation, their combined effects on model reliability have received little attention. This paper presents an empirical investigation of how pruning, adversarial training, and hardware fault injection interact to affect the robustness of convolutional neural networks. Using a compact three-layer CNN trained on MNIST, we conduct three experiments: (1) comparing the fault tolerance of naturally and adversarially trained models under simultaneous hardware faults and adversarial attacks, (2) evaluating how pruning affects adversarial robustness, and (3) characterizing the joint accuracy surface across fault rates, adversarial perturbation magnitudes, and pruning levels. Our results show that adversarial training improves robustness against input perturbations but increases sensitivity to stuck-at-zero weight faults. Contrary to intuition, pruning did not significantly increase fault sensitivity, and varying the pruning level had little effect across fault rates and attack strengths. These results highlight the need to jointly consider adversarial robustness and hardware reliability.

Problem

• fault tolerance
• adversarial robustness
• model pruning
• hardware faults
• neuromorphic hardware

Innovation

• adversarial robustness
• model pruning
• hardware fault tolerance
• stuck-at-zero errors
• joint robustness analysis