Here are all HW submissions of mine & Zvika Deutsch made during the "Adversarial Learning Course" Of Y-DATA, Yandex between May 2021 to July 2021.
HW descriptions:
- Ex1-Intro to Adversarial-Learning - A task which introduces the basic terms in adversarial attacks. Results are available here.
- Ex2-Adversarial Re-training - A task which focuses on the basic "Fast-Gradient descent method" (FGSM) scheme and the defense machanism of re-training. Results are available here.
- Ex3-Understanding Convolutional Networks - A task which analyzes the performance of a specific CNN.
- Ex4-Defensive Distillation - A basic implementation of the defensive distillation scheme of [1].
- Ex5-Robust Manifold Defense - A basic implementation of the manifold defense-scheme of [2].
- Ex6-Hop, Skip & Jump Attack - A basic implementation of the "Hop, Skip & Jump" of [3]. Results are available here.
[1] Nicolas Papernot, Patrick McDaniel, Xi Wu, Somesh Jha, Ananthram Swami, Distillation as a Defense to Adversarial Perturbations against Deep Neural Networks, 2016.
[2] Ajil Jalal, Andrew Ilyas, Constantinos Daskalakis, Alexandros G. Dimakis, The Robust Manifold Defense: Adversarial Training using Generative Models, 2019.
[3] Jianbo Chen, Michael I. Jordan, Martin J. Wainwright, HopSkipJumpAttack: A Query-Efficient Decision-Based Attack, 2020.