Training Ensembles to Detect Adversarial Examples

Abstract

We propose a new ensemble method for detecting and classifying adversarial examples generated by state-of-the-art attacks, including DeepFool and C&W. Our method works by training the members of an ensemble to have low classification error on random benign examples while simultaneously minimizing agreement on examples outside the training distribution. We evaluate on both MNIST and CIFAR-10, against oblivious and both white- and black-box adversaries.

References in corpus (1)

• Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong

Cited by in corpus (8)

• Improving Adversarial Robustness of Ensembles with Diversity Training
• DVERGE: Diversifying Vulnerabilities for Enhanced Robust Generation of Ensembles
• Mutual Adversarial Training: Learning together is better than going alone
• PAC-learning in the presence of evasion adversaries
• Exploiting the Inherent Limitation of L0 Adversarial Examples
• Ensemble-in-One: Learning Ensemble within Random Gated Networks for Enhanced Adversarial Robustness
• Evaluating Ensemble Robustness Against Adversarial Attacks
• Measuring the Contribution of Multiple Model Representations in Detecting Adversarial Instances