Proper Network Interpretability Helps Adversarial Robustness in Classification (2020-06-26T00:00:00.000000Z)

TL;DR

This paper theoretically shows that with a proper measurement of interpretation, it is actually difficult to prevent prediction-evasion adversarial attacks from causing interpretation discrepancy, and develops an interpretability-aware defensive scheme built only on promoting robust interpretation (without the need for resorting to adversarial loss minimization).

Abstract

Recent works have empirically shown that there exist adversarial examples that can be hidden from neural network interpretability (namely, making network interpretation maps visually similar), or interpretability is itself susceptible to adversarial attacks. In this paper, we theoretically show that with a proper measurement of interpretation, it is actually difficult to prevent prediction-evasion adversarial attacks from causing interpretation discrepancy, as confirmed by experiments on MNIST, CIFAR-10 and Restricted ImageNet. Spurred by that, we develop an interpretability-aware defensive scheme built only on promoting robust interpretation (without the need for resorting to adversarial loss minimization). We show that our defense achieves both robust classification and robust interpretation, outperforming state-of-the-art adversarial training methods against attacks of large perturbation in particular.

Authors

Pin-Yu Chen

17 papers

Shiyu Chang

12 papers

Sijia Liu

4 papers

TL;DR

Abstract

Authors

References40 items

Adversarial Robustness as a Prior for Learned Representations

Explanations can be manipulated and geometry is to blame

Robust Attribution Regularization

Interpreting Adversarial Examples by Activation Promotion and Suppression

Is AmI (Attacks Meet Interpretability) Robust to Adversarial Examples?

On the (In)fidelity and Sensitivity for Explanations.

Theoretically Principled Trade-off between Robustness and Accuracy

Fooling Network Interpretation in Image Classification

Towards Hiding Adversarial Examples from Network Interpretation

Interpretable Deep Learning under Fire

Robustness via Curvature Regularization, and Vice Versa

Structured Adversarial Attack: Towards General Implementation and Better Interpretability

Is Robustness the Cost of Accuracy? - A Comprehensive Study on the Robustness of 18 Deep Image Classification Models

RISE: Randomized Input Sampling for Explanation of Black-box Models

Robust Physical-World Attacks on Deep Learning Visual Classification

Robustness May Be at Odds with Accuracy

Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples

Adversarial Patch

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

Grad-CAM++: Generalized Gradient-Based Visual Explanations for Deep Convolutional Networks

Interpretation of Neural Networks is Fragile

EAD: Elastic-Net Attacks to Deep Neural Networks via Adversarial Examples

Synthesizing Robust Adversarial Examples

Towards Deep Learning Models Resistant to Adversarial Attacks

SmoothGrad: removing noise by adding noise

Network Dissection: Quantifying Interpretability of Deep Visual Representations

Axiomatic Attribution for Deep Networks

Grad-CAM: Visual Explanations from Deep Networks via Gradient-Based Localization

Towards Evaluating the Robustness of Neural Networks

Wide Residual Networks

Learning Deep Features for Discriminative Localization

The Limitations of Deep Learning in Adversarial Settings

Distillation as a Defense to Adversarial Perturbations Against Deep Neural Networks

Striving for Simplicity: The All Convolutional Net

Explaining and Harnessing Adversarial Examples

Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

Intriguing properties of neural networks

Visualizing and Understanding Convolutional Networks

Testing Robustness Against Unforeseen Adversaries

Proper Network Interpretability Helps Adversarial

Field of Study

Venue Information

Name

Type

URL

Alternate Names