Robust GANs against Dishonest Adversaries

Abstract

Robustness of deep learning models is a property that has recently gained increasing attention. We explore a notion of robustness for generative adversarial models that is pertinent to their internal interactive structure, and show that, perhaps surprisingly, the GAN in its original form is not robust. Our notion of robustness relies on a perturbed discriminator, or noisy, adversarial interference with its feedback. We explore, theoretically and empirically, the effect of model and training properties on this robustness. In particular, we show theoretical conditions for robustness that are supported by empirical evidence. We also test the effect of regularization. Our results suggest variations of GANs that are indeed more robust to noisy attacks and have more stable training behavior, requiring less regularization in general. Inspired by our theoretical results, we further extend our framework to obtain a class of models related to WGAN, with good empirical performance. Overall, our results suggest a new perspective on understanding and designing GAN models from the viewpoint of their internal robustness.

References in corpus (22)

• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Spectral Normalization for Generative Adversarial Networks
• Ensemble Adversarial Training: Attacks and Defenses
• Generative Adversarial Text to Image Synthesis
• Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling
• NIPS 2016 Tutorial: Generative Adversarial Networks
• Obfuscated Gradients Give a False Sense of Security: Circumventing Defenses to Adversarial Examples
• Theoretically Principled Trade-off between Robustness and Accuracy
• Energy-based Generative Adversarial Network
• BEGAN: Boundary Equilibrium Generative Adversarial Networks
• Adversarially Learned Inference
• Measuring the tendency of CNNs to Learn Surface Statistical Regularities
• Distributionally Robust Logistic Regression
• AdaGAN: Boosting Generative Models
• Loss-Sensitive Generative Adversarial Networks on Lipschitz Densities
• Adversarial Example Defenses: Ensembles of Weak Defenses are not Strong
• Robustness of Conditional GANs to Noisy Labels
• ME-Net: Towards Effective Adversarial Robustness with Matrix Estimation
• Generative Multi-Adversarial Networks
• Generative Adversarial Nets from a Density Ratio Estimation Perspective
• Robust Conditional Generative Adversarial Networks
• Calibrating Energy-based Generative Adversarial Networks