Adversarial Machine Learning in Latent Representations of Neural Networks

Distributed deep neural networks (DNNs) have been shown to reduce the computational burden of mobile devices and decrease the end-to-end inference latency in edge computing scenarios. While distributed DNNs have been studied, to the best of our knowledge the resilience of distributed DNNs to adversarial action still remains an open problem. In this paper, we fill the existing research gap by rigorously analyzing the robustness of distributed DNNs against adversarial action. We cast this problem in the context of information theory and introduce two new measurements for distortion and robustness. Our theoretical findings indicate that (i) assuming the same level of information distortion, latent features are always more robust than input representations; (ii) the adversarial robustness is jointly determined by the feature dimension and the generalization capability of the DNN. To test our theoretical findings, we perform extensive experimental analysis by considering 6 different DNN architectures, 6 different approaches for distributed DNN and 10 different adversarial attacks to the ImageNet-1K dataset. Our experimental results support our theoretical findings by showing that the compressed latent representations can reduce the success rate of adversarial attacks by 88% in the best case and by 57% on the average compared to attacks to the input space.