Improving the Adversarial Robustness and Interpretability of Deep Neural Networks by Regularizing their Input Gradients

Abstract

Deep neural networks have proven remarkably effective at solving many classification problems, but have been criticized recently for two major weaknesses: the reasons behind their predictions are uninterpretable, and the predictions themselves can often be fooled by small adversarial perturbations. These problems pose major obstacles for the adoption of neural networks in domains that require security or transparency. In this work, we evaluate the effectiveness of defenses that differentiably penalize the degree to which small changes in inputs can alter model predictions. Across multiple attacks, architectures, defenses, and datasets, we find that neural networks trained with this input gradient regularization exhibit robustness to transferred adversarial examples generated to fool all of the other models. We also find that adversarial examples generated to fool gradient-regularized models fool all other models equally well, and actually lead to more "legitimate," interpretable misclassifications as rated by people (which we confirm in a human subject experiment). Finally, we demonstrate that regularizing input gradients makes them more naturally interpretable as rationales for model predictions. We conclude by discussing this relationship between interpretability and robustness in deep neural networks.

To appear in AAAI 2018

Cited by in corpus (27)

• Assuring the Machine Learning Lifecycle: Desiderata, Methods, and Challenges
• On the Effectiveness of Mitigating Data Poisoning Attacks with Gradient Shaping
• Opportunities and Challenges in Deep Learning Adversarial Robustness: A Survey
• Bridging Adversarial Robustness and Gradient Interpretability
• Mitigating Advanced Adversarial Attacks with More Advanced Gradient Obfuscation Techniques
• EMPIR: Ensembles of Mixed Precision Deep Networks for Increased Robustness against Adversarial Attacks
• Interpreting and Evaluating Neural Network Robustness
• Jacobian Adversarially Regularized Networks for Robustness
• AIR5: Five Pillars of Artificial Intelligence Research
• Second Order Optimization for Adversarial Robustness and Interpretability
• Blind Adversarial Network Perturbations
• From Shallow to Deep Interactions Between Knowledge Representation, Reasoning and Machine Learning (Kay R. Amel group)
• Input Hessian Regularization of Neural Networks
• Double Backpropagation for Training Autoencoders against Adversarial Attack
• Disentangled Deep Autoencoding Regularization for Robust Image Classification
• Why Blocking Targeted Adversarial Perturbations Impairs the Ability to Learn
• Robustness from Simple Classifiers
• Rethinking Empirical Evaluation of Adversarial Robustness Using First-Order Attack Methods
• Deep Adaptive Wavelet Network
• Graph Interpolating Activation Improves Both Natural and Robust Accuracies in Data-Efficient Deep Learning
• Deep Minimax Probability Machine
• Constraining Logits by Bounded Function for Adversarial Robustness
• Semantics-Preserving Adversarial Training
• Likelihood Landscapes: A Unifying Principle Behind Many Adversarial Defenses
• Improving Resistance to Adversarial Deformations by Regularizing Gradients
• On Connections between Regularizations for Improving DNN Robustness
• Orthogonal Deep Models As Defense Against Black-Box Attacks