Adversarial defense via the data-dependent activation, total variation minimization, and adversarial training
- 1 January 2021
- journal article
- research article
- Published by American Institute of Mathematical Sciences (AIMS) in Inverse Problems and Imaging
- Vol. 15 (1) , 129-145
Abstract
We improve the robustness of Deep Neural Net (DNN) to adversarial attacks by using an interpolating function as the output activation. This data-dependent activation remarkably improves both the generalization and robustness of DNN. In the CIFAR10 benchmark, we raise the robust accuracy of the adversarially trained ResNet20 from $ \sim 46\% $ to $ \sim 69\% $ under the state-of-the-art Iterative Fast Gradient Sign Method (IFGSM) based adversarial attack. When we combine this data-dependent activation with total variation minimization on adversarial images and training data augmentation, we achieve an improvement in robust accuracy by 38.9$ \% $ for ResNet56 under the strongest IFGSM attack. Furthermore, We provide an intuitive explanation of our defense by analyzing the geometry of the feature space.
Keywords
This publication has 12 references indexed in Scilit:
- Boosting Adversarial Attacks with Momentum2018
- Defense Against Universal Adversarial Perturbations2018
- Defense Against Adversarial Attacks Using High-Level Representation Guided Denoiser2018
- Deflecting Adversarial Attacks with Pixel Deflection2018
- Threat of Adversarial Attacks on Deep Learning in Computer Vision: A SurveyIEEE Access, 2018
- Universal Adversarial Perturbations2017
- Deep Residual Learning for Image Recognition2016
- The Limitations of Deep Learning in Adversarial Settings2016
- A Machine Learning Approach to Visual Perception of Forest Trails for Mobile RobotsIEEE Robotics and Automation Letters, 2015
- Nonlinear total variation based noise removal algorithmsPhysica D: Nonlinear Phenomena, 1992