Calibrating Energy-based Generative Adversarial Networks

TL;DR

This paper introduces a novel training framework for energy-based GANs that enables direct energy estimation of samples, ensuring convergence to the true data distribution and retaining density information at optimality.

Contribution

It proposes a flexible adversarial training framework with theoretical guarantees and practical approximation techniques for energy-based GANs.

Findings

Discriminator accurately recovers data distribution energy.

Framework ensures generator converges to true data distribution.

Empirical results validate theoretical analysis.

Abstract

In this paper, we propose to equip Generative Adversarial Networks with the ability to produce direct energy estimates for samples.Specifically, we propose a flexible adversarial training framework, and prove this framework not only ensures the generator converges to the true data distribution, but also enables the discriminator to retain the density information at the global optimal. We derive the analytic form of the induced solution, and analyze the properties. In order to make the proposed framework trainable in practice, we introduce two effective approximation techniques. Empirically, the experiment results closely match our theoretical analysis, verifying the discriminator is able to recover the energy of data distribution.