Adversarial Lipschitz Regularization

Abstract

Generative adversarial networks (GANs) are one of the most popular approaches when it comes to training generative models, among which variants of Wasserstein GANs are considered superior to the standard GAN formulation in terms of learning stability and sample quality. However, Wasserstein GANs require the critic to be 1-Lipschitz, which is often enforced implicitly by penalizing the norm of its gradient, or by globally restricting its Lipschitz constant via weight normalization techniques. Training with a regularization term penalizing the violation of the Lipschitz constraint explicitly, instead of through the norm of the gradient, was found to be practically infeasible in most situations. Inspired by Virtual Adversarial Training, we propose a method called Adversarial Lipschitz Regularization, and show that using an explicit Lipschitz penalty is indeed viable and leads to competitive performance when applied to Wasserstein GANs, highlighting an important connection between Lipschitz regularization and adversarial training.

arXiv commentICLR 2020

References in corpus (11)

• GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Spectral Normalization for Generative Adversarial Networks
• An empirical study on evaluation metrics of generative adversarial networks
• On the regularization of Wasserstein GANs
• Banach Wasserstein GAN
• Wasserstein Adversarial Examples via Projected Sinkhorn Iterations
• Batch Normalization is a Cause of Adversarial Vulnerability
• Lipschitz Generative Adversarial Nets
• Primal-Dual Wasserstein GAN
• Towards Efficient and Unbiased Implementation of Lipschitz Continuity in GANs

Cited by in corpus (11)

• Adversarial Attack and Defense Strategies for Deep Speaker Recognition Systems
• CLIP: Cheap Lipschitz Training of Neural Networks
• Training Generative Adversarial Networks by Solving Ordinary Differential Equations
• Improving GAN Training with Probability Ratio Clipping and Sample Reweighting
• Self-Supervised GANs with Label Augmentation
• Robust Learning Meets Generative Models: Can Proxy Distributions Improve Adversarial Robustness?
• Gradient Shaping: Enhancing Backdoor Attack Against Reverse Engineering
• Variational Autoencoders: A Harmonic Perspective
• Analytical bounds on the local Lipschitz constants of affine-ReLU functions
• Moreau-Yosida $f$-divergences
• Stein Latent Optimization for Generative Adversarial Networks