Evaluating GFlowNet from partial episodes for stable and flexible policy-based training

Puhua Niu, Shili Wu, Xiaoning Qian · 2026

Read Evaluating GFlowNet from partial episodes for stable and flexible policy-based training by Puhua Niu, Shili Wu, Xiaoning Qian (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-01047.txt.

Description

Niu, Wu, and Qian bridge value-based and policy-based training for Generative Flow Networks, which sample candidate sets along one-way acyclic graphs. They show flow balance over partial runs yields a sound policy evaluator for divergence, and propose a balance goal to learn it. The approach strengthens policy-based training's trust while adding freedom: tuned backward policies and offline data use both fit with ease, as synthetic and real tasks confirm.

Catalog record

CREATORPuhua Niu, Shili Wu, Xiaoning Qian
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE102096
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.01047

Browse the full library · About the Tower of Babel