Improving GANs Using Optimal Transport

Abstract

We present Optimal Transport GAN (OT-GAN), a variant of generative adversarial nets minimizing a new metric measuring the distance between the generator distribution and the data distribution. This metric, which we call mini-batch energy distance, combines optimal transport in primal form with an energy distance defined in an adversarially learned feature space, resulting in a highly discriminative distance function with unbiased mini-batch gradients. Experimentally we show OT-GAN to be highly stable when trained with large mini-batches, and we present state-of-the-art results on several popular benchmark problems for image generation.

References in corpus (10)

• Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks
• Generative Adversarial Text to Image Synthesis
• Improved Techniques for Training GANs
• Equivalence of distance-based and RKHS-based statistics in hypothesis testing
• Unrolled Generative Adversarial Networks
• The Cramer Distance as a Solution to Biased Wasserstein Gradients
• Learning What and Where to Draw
• The Reversible Residual Network: Backpropagation Without Storing Activations
• From optimal transport to generative modeling: the VEGAN cookbook
• GAN and VAE from an Optimal Transport Point of View

Cited by in corpus (20)

• A Probabilistic U-Net for Segmentation of Ambiguous Images
• Road Segmentation for Remote Sensing Images using Adversarial Spatial Pyramid Networks
• Missing Data Imputation using Optimal Transport
• UPHDR-GAN: Generative Adversarial Network for High Dynamic Range Imaging with Unpaired Data
• COT-GAN: Generating Sequential Data via Causal Optimal Transport
• Small-GAN: Speeding Up GAN Training Using Core-sets
• Autoregressive Quantile Networks for Generative Modeling
• Adversarial Distillation of Bayesian Neural Network Posteriors
• Rates of Estimation of Optimal Transport Maps using Plug-in Estimators via Barycentric Projections
• GANs May Have No Nash Equilibria
• 2-Wasserstein Approximation via Restricted Convex Potentials with Application to Improved Training for GANs
• A Survey on Optimal Transport for Machine Learning: Theory and Applications
• Adversarial Computation of Optimal Transport Maps
• Differentiable Deep Clustering with Cluster Size Constraints
• Learning Spatial Pyramid Attentive Pooling in Image Synthesis and Image-to-Image Translation
• Training Generative Reversible Networks
• Artist Style Transfer Via Quadratic Potential
• Mode Penalty Generative Adversarial Network with adapted Auto-encoder
• Regularized Optimal Transport for Dynamic Semi-supervised Learning
• MMCGAN: Generative Adversarial Network with Explicit Manifold Prior