Saved in:
Bibliographic Details
Main Authors: Wang, Yan, Deng, Ke, Ren, Yongli
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.18671
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911288076009472
author Wang, Yan
Deng, Ke
Ren, Yongli
author_facet Wang, Yan
Deng, Ke
Ren, Yongli
contents Cooperative multi-agent reinforcement learning (MARL) commonly adopts centralized training with decentralized execution (CTDE), where centralized critics leverage global information to guide decentralized actors. However, centralized-decentralized mismatch (CDM) arises when the suboptimal behavior of one agent degrades others' learning. Prior approaches mitigate CDM through value decomposition, but linear decompositions allow per-agent gradients at the cost of limited expressiveness, while nonlinear decompositions improve representation but require centralized gradients, reintroducing CDM. To overcome this trade-off, we propose the multi-agent cross-entropy method (MCEM), combined with monotonic nonlinear critic decomposition (NCD). MCEM updates policies by increasing the probability of high-value joint actions, thereby excluding suboptimal behaviors. For sample efficiency, we extend off-policy learning with a modified k-step return and Retrace. Analysis and experiments demonstrate that MCEM outperforms state-of-the-art methods across both continuous and discrete action benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2511_18671
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Multi-Agent Cross-Entropy Method with Monotonic Nonlinear Critic Decomposition
Wang, Yan
Deng, Ke
Ren, Yongli
Machine Learning
Multiagent Systems
Cooperative multi-agent reinforcement learning (MARL) commonly adopts centralized training with decentralized execution (CTDE), where centralized critics leverage global information to guide decentralized actors. However, centralized-decentralized mismatch (CDM) arises when the suboptimal behavior of one agent degrades others' learning. Prior approaches mitigate CDM through value decomposition, but linear decompositions allow per-agent gradients at the cost of limited expressiveness, while nonlinear decompositions improve representation but require centralized gradients, reintroducing CDM. To overcome this trade-off, we propose the multi-agent cross-entropy method (MCEM), combined with monotonic nonlinear critic decomposition (NCD). MCEM updates policies by increasing the probability of high-value joint actions, thereby excluding suboptimal behaviors. For sample efficiency, we extend off-policy learning with a modified k-step return and Retrace. Analysis and experiments demonstrate that MCEM outperforms state-of-the-art methods across both continuous and discrete action benchmarks.
title Multi-Agent Cross-Entropy Method with Monotonic Nonlinear Critic Decomposition
topic Machine Learning
Multiagent Systems
url https://arxiv.org/abs/2511.18671