Proximal Point Nash Learning from Human Feedback

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Tiapkin, Daniil, Calandriello, Daniele, Belomestny, Denis, Moulines, Eric, Naumov, Alexey, Rasul, Kashif, Valko, Michal, Menard, Pierre
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866910062161690624
author Tiapkin, Daniil
Calandriello, Daniele
Belomestny, Denis
Moulines, Eric
Naumov, Alexey
Rasul, Kashif
Valko, Michal
Menard, Pierre
author_facet Tiapkin, Daniil
Calandriello, Daniele
Belomestny, Denis
Moulines, Eric
Naumov, Alexey
Rasul, Kashif
Valko, Michal
Menard, Pierre
contents Traditional Reinforcement Learning from Human Feedback (RLHF) often relies on reward models, frequently assuming preference structures like the Bradley--Terry model, which may not accurately capture the complexities of real human preferences (e.g., intransitivity). Nash Learning from Human Feedback (NLHF) offers a more direct alternative by framing the problem as finding a Nash equilibrium of a game defined by these preferences. While many works study the Nash learning problem directly in the policy space, we instead consider it under a more realistic policy parametrization setting. We first analyze a simple self-play policy gradient method, which is equivalent to Online IPO. We establish high-probability last-iterate convergence guarantees for this method, but our analysis also reveals a possible stability limitation of the underlying dynamics. Motivated by this, we embed the self-play updates into a proximal point framework, yielding a stabilized algorithm. For this combined method, we prove high-probability last-iterate convergence and discuss its more practical version, which we call Nash Prox. Finally, we apply this method to post-training of large language models and validate its empirical performance.
format Preprint
id arxiv_https___arxiv_org_abs_2505_19731
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Proximal Point Nash Learning from Human Feedback
Tiapkin, Daniil
Calandriello, Daniele
Belomestny, Denis
Moulines, Eric
Naumov, Alexey
Rasul, Kashif
Valko, Michal
Menard, Pierre
Machine Learning
Traditional Reinforcement Learning from Human Feedback (RLHF) often relies on reward models, frequently assuming preference structures like the Bradley--Terry model, which may not accurately capture the complexities of real human preferences (e.g., intransitivity). Nash Learning from Human Feedback (NLHF) offers a more direct alternative by framing the problem as finding a Nash equilibrium of a game defined by these preferences. While many works study the Nash learning problem directly in the policy space, we instead consider it under a more realistic policy parametrization setting. We first analyze a simple self-play policy gradient method, which is equivalent to Online IPO. We establish high-probability last-iterate convergence guarantees for this method, but our analysis also reveals a possible stability limitation of the underlying dynamics. Motivated by this, we embed the self-play updates into a proximal point framework, yielding a stabilized algorithm. For this combined method, we prove high-probability last-iterate convergence and discuss its more practical version, which we call Nash Prox. Finally, we apply this method to post-training of large language models and validate its empirical performance.
title Proximal Point Nash Learning from Human Feedback
topic Machine Learning
url https://arxiv.org/abs/2505.19731