Two simple training schemes, GMPO and GMPG, let diffusion and flow models act as offline-RL policies with competitive or better scores on standard benchmarks, and a new framework, GenerativeRL, standardizes such experiments.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Revisiting Generative Policies: A Simpler Reinforcement Learning Algorithmic Perspective
Two simple training schemes, GMPO and GMPG, let diffusion and flow models act as offline-RL policies with competitive or better scores on standard benchmarks, and a new framework, GenerativeRL, standardizes such experiments.