An Empirical Evaluation of Adversarial Robustness under Transfer Learning

TL;DR

This paper empirically evaluates how adversarial robustness transfers from a source to a target network in transfer learning, focusing on the effects of robust training strategies and attack methods.

Contribution

It investigates the transferability of adversarial robustness using PGD-based training and attack methods across different transfer learning strategies.

Findings

PGD training on source leads to more transferable robust features

Transfer robustness improves white-box PGD attack accuracy by 5.2%

Insights into robustness generalization in transfer learning scenarios

Abstract

In this work, we evaluate adversarial robustness in the context of transfer learning from a source trained on CIFAR 100 to a target network trained on CIFAR 10. Specifically, we study the effects of using robust optimisation in the source and target networks. This allows us to identify transfer learning strategies under which adversarial defences are successfully retained, in addition to revealing potential vulnerabilities. We study the extent to which features learnt by a fast gradient sign method (FGSM) and its iterative alternative (PGD) can preserve their defence properties against black and white-box attacks under three different transfer learning strategies. We find that using PGD examples during training on the source task leads to more general robust features that are easier to transfer. Furthermore, under successful transfer, it achieves 5.2% more accuracy against white-box PGD attacks than suitable baselines. Overall, our empirical evaluations give insights on how well adversarial robustness under transfer learning can generalise.