CLASSIFICATION ATTACK DETECTION VIA CLASS VISUAL CONTEXT DISCREPANCY MEASURE
Abstract
Despite the excellent performance of deep learning models, it has been recently discovered that these models are vulnerable to adversarial examples that can easily fool the model predictions. While many methods have been proposed to detect the adversarial attacks in progress, these methods have been demonstrated to be susceptible to gradient-based adaptive attacks. We propose a novel optimization based detection method which is resistant to the gradient-based adaptive attacks by construction. Our preliminary results on ImageNet dataset show that the proposed measure exhibits a promising detection discriminative performance.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Aug 11, 2021
- Source ID
- FA23862014043
Entities
People
- Hyun Oh Song
Organizations
- Air Force Office of Scientific Research
- Seoul National University
- United States Air Force