Model-Agnostic Adversarial Detection by Random Perturbations
Abstract
Adversarial examples induce model classification errors on purpose, which has raised concerns on the security aspect of machine learning techniques. Many existing countermeasures are compromised by adaptive adversaries and transferred examples. We propose a model-agnostic approach to resolve the problem by analysing the model responses to an input under random perturbations, and study the robustness of detecting norm-bounded adversarial distortions in a theoretical framework. Extensive evaluations are performed on the MNIST, CIFAR-10 and ImageNet datasets. The results demonstrate that our detection method is effective and resilient against various attacks including black-box attacks and the powerful CW attack with four adversarial adaptations.
Cite
Text
Huang et al. "Model-Agnostic Adversarial Detection by Random Perturbations." International Joint Conference on Artificial Intelligence, 2019. doi:10.24963/IJCAI.2019/651Markdown
[Huang et al. "Model-Agnostic Adversarial Detection by Random Perturbations." International Joint Conference on Artificial Intelligence, 2019.](https://mlanthology.org/ijcai/2019/huang2019ijcai-model/) doi:10.24963/IJCAI.2019/651BibTeX
@inproceedings{huang2019ijcai-model,
title = {{Model-Agnostic Adversarial Detection by Random Perturbations}},
author = {Huang, Bo and Wang, Yi and Wang, Wei},
booktitle = {International Joint Conference on Artificial Intelligence},
year = {2019},
pages = {4689-4696},
doi = {10.24963/IJCAI.2019/651},
url = {https://mlanthology.org/ijcai/2019/huang2019ijcai-model/}
}