Learning Probabilistic Models from Generator Latent Spaces with Hat EBM

TL;DR

This paper introduces Hat EBM, a novel probabilistic modeling approach that integrates generator networks into an energy-based framework, enabling explicit output distribution modeling without complex inference.

Contribution

It presents a new method to incorporate generator networks into EBMs without latent inference or Jacobian calculations, broadening probabilistic modeling capabilities.

Findings

Strong performance on ImageNet synthesis at 128x128

Effective refinement of existing generator outputs

Successful learning of EBMs with non-probabilistic generators

Abstract

This work proposes a method for using any generator network as the foundation of an Energy-Based Model (EBM). Our formulation posits that observed images are the sum of unobserved latent variables passed through the generator network and a residual random variable that spans the gap between the generator output and the image manifold. One can then define an EBM that includes the generator as part of its forward pass, which we call the Hat EBM. The model can be trained without inferring the latent variables of the observed data or calculating the generator Jacobian determinant. This enables explicit probabilistic modeling of the output distribution of any type of generator network. Experiments show strong performance of the proposed method on (1) unconditional ImageNet synthesis at 128x128 resolution, (2) refining the output of existing generators, and (3) learning EBMs that incorporate non-probabilistic generators. Code and pretrained models to reproduce our results are available at https://github.com/point0bar1/hat-ebm.