Machine vs Machine: Minimax-Optimal Defense Against Adversarial Examples

Abstract

Recently, researchers have discovered that the state-of-the-art object classifiers can be fooled easily by small perturbations in the input unnoticeable to human eyes. It is also known that an attacker can generate strong adversarial examples if she knows the classifier parameters. Conversely, a defender can robustify the classifier by retraining if she has access to the adversarial examples. We explain and formulate this adversarial example problem as a two-player continuous zero-sum game, and demonstrate the fallacy of evaluating a defense or an attack as a static problem. To find the best worst-case defense against whitebox attacks, we propose a continuous minimax optimization algorithm. We demonstrate the minimax defense with two types of attack classes -- gradient-based and neural network-based attacks. Experiments with the MNIST and the CIFAR-10 datasets demonstrate that the defense found by numerical minimax optimization is indeed more robust than non-minimax defenses. We discuss directions for improving the result toward achieving robustness against multiple types of attack classes.

References in corpus (8)

• Ensemble Adversarial Training: Attacks and Defenses
• Towards Deep Neural Network Architectures Robust to Adversarial Examples
• Adversarial Transformation Networks: Learning to Generate Adversarial Examples
• MagNet: a Two-Pronged Defense against Adversarial Examples
• Universal adversarial perturbations
• Gradient descent GAN optimization is locally stable
• Robustness of classifiers to universal perturbations: a geometric perspective
• SafetyNet: Detecting and Rejecting Adversarial Examples Robustly

Cited by in corpus (2)

• Motivating the Rules of the Game for Adversarial Example Research
• Robust Machine Learning via Privacy/Rate-Distortion Theory