Skip to yearly menu bar Skip to main content


Spotlight Poster

Local Search GFlowNets

Minsu Kim · Yun Taeyoung · Emmanuel Bengio · Dinghuai Zhang · Yoshua Bengio · Sungsoo Ahn · Jinkyoo Park

Halle B #177
[ ]
Wed 8 May 1:45 a.m. PDT — 3:45 a.m. PDT

Abstract:

Generative Flow Networks (GFlowNets) are amortized sampling methods that learn a distribution over discrete objects proportional to their rewards. GFlowNets exhibit a remarkable ability to generate diverse samples, yet occasionally struggle to consistently produce samples with high rewards due to over-exploration on wide sample space. This paper proposes to train GFlowNets with local search, which focuses on exploiting high-rewarded sample space to resolve this issue. Our main idea is to explore the local neighborhood via backtracking and reconstruction guided by backward and forward policies, respectively. This allows biasing the samples toward high-reward solutions, which is not possible for a typical GFlowNet solution generation scheme, which uses the forward policy to generate the solution from scratch. Extensive experiments demonstrate a remarkable performance improvement in several biochemical tasks. Source code is available: \url{https://github.com/dbsxodud-11/ls_gfn}.

Live content is unavailable. Log in and register to view live content