Implicit Generation and Generalization in Energy-Based Models

TL;DR

This paper advances energy-based models (EBMs) by introducing scalable training techniques, demonstrating their effectiveness on high-dimensional datasets, and showcasing their versatility in tasks like classification, inpainting, and trajectory prediction.

Contribution

The authors develop scalable MCMC training methods for EBMs on neural networks and demonstrate their effectiveness across diverse high-dimensional data and tasks.

Findings

EBMs achieve better samples than likelihood models and approach GAN performance.

EBMs cover all data modes and enable tasks like inpainting and compositional generation.

EBMs attain state-of-the-art results in out-of-distribution and adversarial robustness.

Abstract

Energy based models (EBMs) are appealing due to their generality and simplicity in likelihood modeling, but have been traditionally difficult to train. We present techniques to scale MCMC based EBM training on continuous neural networks, and we show its success on the high-dimensional data domains of ImageNet32x32, ImageNet128x128, CIFAR-10, and robotic hand trajectories, achieving better samples than other likelihood models and nearing the performance of contemporary GAN approaches, while covering all modes of the data. We highlight some unique capabilities of implicit generation such as compositionality and corrupt image reconstruction and inpainting. Finally, we show that EBMs are useful models across a wide variety of tasks, achieving state-of-the-art out-of-distribution classification, adversarially robust classification, state-of-the-art continual online class learning, and coherent long term predicted trajectory rollouts.