Publishing Efficient On-device Models Increases Adversarial Vulnerability
Authors:
- 1 February 2023
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 271-290
Abstract
Recent increases in the computational demands of deep neural networks (DNNs) have sparked interest in efficient deep learning mechanisms, e.g., quantization or pruning. These mechanisms enable the construction of a small, efficient version of commercial-scale models with comparable accuracy, accelerating their deployment to resource-constrained devices. In this paper, we study the security considerations of publishing on-device variants of large-scale models. We first show that an adversary can exploit on-device models to make attacking the large models easier. In evaluations across 19 DNNs, by exploiting the published on-device models as a transfer prior, the adversarial vulnerability of the original commercial-scale models increases by up to 100x. We then show that the vulnerability increases as the similarity between a full-scale and its efficient model increase. Based on the insights, we propose a defense, similarity-unpairing, that fine-tunes on-device models with the objective of reducing the similarity. We evaluated our defense on all the 19 DNNs and found that it reduces the transferability up to 90% and the number of queries required by a factor of 10-100x. Our results suggest that further research is needed on the security (or even privacy) threats caused by publishing those efficient siblings.Keywords
All Related Versions
This publication has 30 references indexed in Scilit:
- LQ-Nets: Learned Quantization for Highly Accurate and Compact Deep Neural Networks2018
- Simple Black-Box Adversarial Attacks on Deep Neural Networks2017
- Densely Connected Convolutional Networks2017
- Towards Evaluating the Robustness of Neural Networks2017
- Practical Black-Box Attacks against Machine Learning2017
- DeepFool: A Simple and Accurate Method to Fool Deep Neural Networks2016
- Deep Residual Learning for Image Recognition2016
- Model Inversion Attacks that Exploit Confidence Information and Basic Countermeasures2015
- Evasion Attacks against Machine Learning at Test Time2013
- ImageNet: A large-scale hierarchical image database2009