Learning Energy-Based Models With Adversarial Training

TL;DR

This paper introduces an adversarial training method for energy-based models that effectively models data support, generates diverse realistic images, and offers stability and robustness advantages over traditional EBM training.

Contribution

It presents a novel adversarial training approach for EBMs, improving stability, diversity, and robustness in generative modeling compared to existing methods.

Findings

Generates diverse, realistic images with competitive quality.

Training is stable and well-suited for image translation.

Exhibits strong out-of-distribution adversarial robustness.

Abstract

We study a new approach to learning energy-based models (EBMs) based on adversarial training (AT). We show that (binary) AT learns a special kind of energy function that models the support of the data distribution, and the learning process is closely related to MCMC-based maximum likelihood learning of EBMs. We further propose improved techniques for generative modeling with AT, and demonstrate that this new approach is capable of generating diverse and realistic images. Aside from having competitive image generation performance to explicit EBMs, the studied approach is stable to train, is well-suited for image translation tasks, and exhibits strong out-of-distribution adversarial robustness. Our results demonstrate the viability of the AT approach to generative modeling, suggesting that AT is a competitive alternative approach to learning EBMs.