Generative Flow Networks for Discrete Probabilistic Modeling

TL;DR

This paper introduces energy-based generative flow networks (EB-GFN), a new probabilistic modeling approach for high-dimensional discrete data that combines GFlowNets with energy functions for efficient sampling and mode mixing.

Contribution

The paper develops a novel energy-based GFlowNet framework that jointly trains a GFlowNet and an energy function for improved discrete data modeling.

Findings

EB-GFN effectively models high-dimensional discrete data.

The approach enables efficient mode mixing via large-block Gibbs sampling.

Code implementation is publicly available.

Abstract

We present energy-based generative flow networks (EB-GFN), a novel probabilistic modeling algorithm for high-dimensional discrete data. Building upon the theory of generative flow networks (GFlowNets), we model the generation process by a stochastic data construction policy and thus amortize expensive MCMC exploration into a fixed number of actions sampled from a GFlowNet. We show how GFlowNets can approximately perform large-block Gibbs sampling to mix between modes. We propose a framework to jointly train a GFlowNet with an energy function, so that the GFlowNet learns to sample from the energy distribution, while the energy learns with an approximate MLE objective with negative samples from the GFlowNet. We demonstrate EB-GFN's effectiveness on various probabilistic modeling tasks. Code is publicly available at https://github.com/zdhNarsil/EB_GFN.