Mean-Field Diffuser: Scaling Offline MARL to Thousands of Agents
Wenhao Li, Xiangfeng Wang, Bo Jin · 2026
Read Mean-Field Diffuser: Scaling Offline MARL to Thousands of Agents by Wenhao Li, Xiangfeng Wang, Bo Jin (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-2605-30190.txt.
Introduction
Wenhao Li, Xiangfeng Wang and Bo Jin introduce MF-Diffuser, lifting planning to the Wasserstein space, where a small subset of agents captures population motion. Their approach has a value-weighted entropy objective balancing fidelity with return maximization, and a gradual strategy growing the agent population. They establish suboptimality bounds, and prove the policy forms an approximate mean-field Nash equilibrium with convergence guarantees. Tests on three benchmarks show best returns in most settings. The largest gains come on suboptimal data and at scales over a thousand agents.
Catalog record
| CREATOR | Wenhao Li, Xiangfeng Wang, Bo Jin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 246145 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.30190