New Losses for Generative Adversarial Learning

TL;DR

This paper addresses mathematical issues in training generative adversarial networks by proposing a unifying methodology for defining sound training objectives that account for the discriminator's dependency on generator parameters.

Contribution

It introduces a mathematically rigorous framework for gradient computation in generative models, improving the robustness of training across GAN, VAE, and variants.

Findings

Provides a unified methodology for gradient calculation

Enhances robustness of training objectives

Applicable to GAN, VAE, and variants

Abstract

Generative Adversarial Networks (Goodfellow et al., 2014), a major breakthrough in the field of generative modeling, learn a discriminator to estimate some distance between the target and the candidate distributions. This paper examines mathematical issues regarding the way the gradients for the generative model are computed in this context, and notably how to take into account how the discriminator itself depends on the generator parameters. A unifying methodology is presented to define mathematically sound training objectives for generative models taking this dependency into account in a robust way, covering both GAN, VAE and some GAN variants as particular cases.