Visible to the public Towards Understanding Limitations of Pixel Discretization Against Adversarial Attacks

TitleTowards Understanding Limitations of Pixel Discretization Against Adversarial Attacks
Publication TypeConference Paper
Year of Publication2019
AuthorsChen, Jiefeng, Wu, Xi, Rastogi, Vaibhav, Liang, Yingyu, Jha, Somesh
Conference Name2019 IEEE European Symposium on Security and Privacy (EuroS P)
Date PublishedJune 2019
PublisherIEEE
ISBN Number978-1-7281-1148-3
Keywordsadversarial attacks, Artificial neural networks, composability, cryptography, Data models, Deep Learning, image processing, ImageNet, low computational overhead, machine learning, Measurement, Metrics, MNIST, neural nets, Neural networks, Perturbation methods, pixel discretization, pixel discretization defense method, preprocessing defense, preprocessing defense methods, pubcrawl, resilience, Resiliency, Robustness, Training, white box cryptography, white-box attacks
Abstract

Wide adoption of artificial neural networks in various domains has led to an increasing interest in defending adversarial attacks against them. Preprocessing defense methods such as pixel discretization are particularly attractive in practice due to their simplicity, low computational overhead, and applicability to various systems. It is observed that such methods work well on simple datasets like MNIST, but break on more complicated ones like ImageNet under recently proposed strong white-box attacks. To understand the conditions for success and potentials for improvement, we study the pixel discretization defense method, including more sophisticated variants that take into account the properties of the dataset being discretized. Our results again show poor resistance against the strong attacks. We analyze our results in a theoretical framework and offer strong evidence that pixel discretization is unlikely to work on all but the simplest of the datasets. Furthermore, our arguments present insights why some other preprocessing defenses may be insecure.

URLhttps://ieeexplore.ieee.org/document/8806764
DOI10.1109/EuroSP.2019.00042
Citation Keychen_towards_2019