QGFN: Controllable Greediness with Action Values

Elaine Lau, Stephen Zhewen Lu, Ling Pan, Doina Precup, Emmanuel Bengio

Keywords: GFlowNets, generative models, reinforcement learning, molecule design

TL;DR: We combine a GFlowNet policy and an action-value estimate, $Q$ into mixture policies that get better rewards without losing diversity

Abstract: Generative Flow Networks (GFlowNets; GFNs) are a family of energy-based generative methods for combinatorial objects, capable of generating diverse and high-utility samples. However, consistently biasing GFNs towards producing high-utility samples is non-trivial. In this work, we leverage connections between GFNs and reinforcement learning (RL) and propose to combine the GFN policy with an action-value estimate, $Q$, to create greedier sampling policies which can be controlled by a mixing parameter. We show that several variants of the proposed method, QGFN, are able to improve on the number of high-reward samples generated in a variety of tasks without sacrificing diversity.

Submission Number: 48