QGFN: Controllable Greediness with Action Values

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Lau, Elaine, Lu, Stephen Zhewen, Pan, Ling, Precup, Doina, Bengio, Emmanuel
Formato: Preprint
Publicado: 2024
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866916463840854016
author Lau, Elaine
Lu, Stephen Zhewen
Pan, Ling
Precup, Doina
Bengio, Emmanuel
author_facet Lau, Elaine
Lu, Stephen Zhewen
Pan, Ling
Precup, Doina
Bengio, Emmanuel
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2402_05234
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle QGFN: Controllable Greediness with Action Values
Lau, Elaine
Lu, Stephen Zhewen
Pan, Ling
Precup, Doina
Bengio, Emmanuel
Machine Learning
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.
title QGFN: Controllable Greediness with Action Values
topic Machine Learning
url https://arxiv.org/abs/2402.05234