Diffusion Generative Flow Samplers: Improving learning signals through partial trajectory optimization

TL;DR

This paper introduces Diffusion Generative Flow Samplers (DGFS), a novel method that improves high-dimensional density sampling by utilizing partial trajectory optimization and intermediate learning signals, leading to more accurate normalization constant estimates.

Contribution

DGFS extends existing sampling methods by enabling partial trajectory optimization with flow functions, inspired by GFlowNets, to enhance learning signals and sampling accuracy.

Findings

DGFS outperforms prior methods in estimating normalization constants.

Partial trajectory optimization improves learning efficiency.

Intermediate signals enhance the training process.

Abstract

We tackle the problem of sampling from intractable high-dimensional density functions, a fundamental task that often appears in machine learning and statistics. We extend recent sampling-based approaches that leverage controlled stochastic processes to model approximate samples from these target densities. The main drawback of these approaches is that the training objective requires full trajectories to compute, resulting in sluggish credit assignment issues due to use of entire trajectories and a learning signal present only at the terminal time. In this work, we present Diffusion Generative Flow Samplers (DGFS), a sampling-based framework where the learning process can be tractably broken down into short partial trajectory segments, via parameterizing an additional "flow function". Our method takes inspiration from the theory developed for generative flow networks (GFlowNets), allowing us to make use of intermediate learning signals. Through various challenging experiments, we demonstrate that DGFS achieves more accurate estimates of the normalization constant than closely-related prior methods.