Sat Ya: Defending Against Adversarial Attacks Using Statistical Hypothesis Testing
Abstract
The paper presents a new defense against adversarial attacks for deep neural networks. We demonstrate the effectiveness of our approach against the popular adversarial image generation method DeepFool. Our approach uses Wald’s Sequential Probability Ratio Test to sufficiently sample a carefully chosen neighborhood around an input image to determine the correct label of the image. On a benchmark of 50,000 randomly chosen adversarial images generated by DeepFool we demonstrate that our method SAT YA is able to recover the correct labels for 95.76% of the images for CaffeNet and 97.43% of the correct label for GoogLeNet.
Publication Date
1-1-2018
Publication Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume
10723 LNCS
Number of Pages
277-292
Document Type
Article; Proceedings Paper
Personal Identifier
scopus
DOI Link
https://doi.org/10.1007/978-3-319-75650-9_18
Copyright Status
Unknown
Socpus ID
85042553265 (Scopus)
Source API URL
https://api.elsevier.com/content/abstract/scopus_id/85042553265
STARS Citation
Raj, Sunny; Pullum, Laura; Ramanathan, Arvind; and Jha, Sumit Kumar, "Sat Ya: Defending Against Adversarial Attacks Using Statistical Hypothesis Testing" (2018). Scopus Export 2015-2019. 9547.
https://stars.library.ucf.edu/scopus2015/9547