Expected flow networks in stochastic environments and two-player zero-sum games

TL;DR

This paper introduces expected flow networks (EFlowNets) for stochastic environments and adversarial flow networks (AFlowNets) for two-player zero-sum games, demonstrating superior performance in tasks like protein design and Connect-4.

Contribution

The paper extends GFlowNets to stochastic and adversarial settings, proposing EFlowNets and AFlowNets with improved performance over existing methods.

Findings

EFlowNets outperform other GFlowNet formulations in stochastic tasks.

AFlowNets find over 80% of optimal moves in Connect-4.

AFlowNets outperform AlphaZero in tournaments.

Abstract

Generative flow networks (GFlowNets) are sequential sampling models trained to match a given distribution. GFlowNets have been successfully applied to various structured object generation tasks, sampling a diverse set of high-reward objects quickly. We propose expected flow networks (EFlowNets), which extend GFlowNets to stochastic environments. We show that EFlowNets outperform other GFlowNet formulations in stochastic tasks such as protein design. We then extend the concept of EFlowNets to adversarial environments, proposing adversarial flow networks (AFlowNets) for two-player zero-sum games. We show that AFlowNets learn to find above 80% of optimal moves in Connect-4 via self-play and outperform AlphaZero in tournaments.