Improving Adversarial Robustness of Ensembles with Diversity Training

Abstract

Deep Neural Networks are vulnerable to adversarial attacks even in settings where the attacker has no direct access to the model being attacked. Such attacks usually rely on the principle of transferability, whereby an attack crafted on a surrogate model tends to transfer to the target model. We show that an ensemble of models with misaligned loss gradients can provide an effective defense against transfer-based attacks. Our key insight is that an adversarial example is less likely to fool multiple models in the ensemble if their loss functions do not increase in a correlated fashion. To this end, we propose Diversity Training, a novel method to train an ensemble of models with uncorrelated loss functions. We show that our method significantly improves the adversarial robustness of ensembles and can also be combined with existing methods to create a stronger defense.

References in corpus (9)

• Transferability in Machine Learning: from Phenomena to Black-Box Attacks using Adversarial Samples
• Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples
• Delving into Transferable Adversarial Examples and Black-box Attacks
• Technical Report on the CleverHans v2.1.0 Adversarial Examples Library
• Adapting Auxiliary Losses Using Gradient Similarity
• Ensemble Methods as a Defense to Adversarial Perturbations Against Deep Neural Networks
• Cascade Adversarial Machine Learning Regularized with a Unified Embedding
• Towards Robust Neural Networks via Random Self-ensemble
• Training Ensembles to Detect Adversarial Examples

Cited by in corpus (28)

• The Pitfalls of Simplicity Bias in Neural Networks
• DVERGE: Diversifying Vulnerabilities for Enhanced Robust Generation of Ensembles
• Recent Advances in Adversarial Training for Adversarial Robustness
• Towards a Robust Deep Neural Network in Texts: A Survey
• Mutual Adversarial Training: Learning together is better than going alone
• Post-breach Recovery: Protection against White-box Adversarial Examples for Leaked DNN Models
• Enhancing Certifiable Robustness via a Deep Model Ensemble
• TRS: Transferability Reduced Ensemble via Encouraging Gradient Diversity and Model Smoothness
• On the Certified Robustness for Ensemble Models and Beyond
• DICE: Diversity in Deep Ensembles via Conditional Redundancy Adversarial Estimation
• $n$-ML: Mitigating Adversarial Examples via Ensembles of Topologically Manipulated Classifiers
• Learning Diverse-Structured Networks for Adversarial Robustness
• Ensemble Defense with Data Diversity: Weak Correlation Implies Strong Robustness
• Omni: Automated Ensemble with Unexpected Models against Adversarial Evasion Attack
• Push Stricter to Decide Better: A Class-Conditional Feature Adaptive Framework for Improving Adversarial Robustness
• Ensemble-in-One: Learning Ensemble within Random Gated Networks for Enhanced Adversarial Robustness
• Mixture of Robust Experts (MoRE):A Robust Denoising Method towards multiple perturbations
• Robustness via Cross-Domain Ensembles
• Uncovering the Connections Between Adversarial Transferability and Knowledge Transferability
• Deep Neural Network Perception Models and Robust Autonomous Driving Systems
• Evaluating Ensemble Robustness Against Adversarial Attacks
• "What's in the box?!": Deflecting Adversarial Attacks by Randomly Deploying Adversarially-Disjoint Models
• Robustness from Simple Classifiers
• Greedy Bayesian Posterior Approximation with Deep Ensembles
• Adversarial Training with Stochastic Weight Average
• Towards Comprehensive Monocular Depth Estimation: Multiple Heads Are Better Than One
• CoG: a Two-View Co-training Framework for Defending Adversarial Attacks on Graph
• Orthogonal Deep Models As Defense Against Black-Box Attacks