arXiv：cs.LG· Sangmin Lee, Youngju Na, Chanmi Lee, Sung-eui Yoon·· 4 小时前AI 评分37
MoSDOT：通过保留支撑集的蒸馏实现多智能体协调
Multi-Agent Coordination via Support-Preserving Distillation
AI 导读
针对离线 MARL 中基于流的教师模型会将相近噪声样本路由至冲突协调模式、导致误差经蒸馏传播给学生的问题，作者提出 Mode-Support Semi-Discrete Optimal Transport（MoSDOT），将多模态回放数据归纳为带指定容量的有限模式支撑集，并用条件半离散最优传输为每个噪声样本分配单一模式。
正文
Abstract:Offline MARL increasingly relies on generative policies to model multimodal joint behavior, typically by distilling a centralized teacher into decentralized one-step actors under the CTDE. We identify a failure mode at the teacher training stage: standard flow-based teachers pair noise with replay targets independently, so nearby noise samples can be routed toward conflicting coordination modes. The teacher then produces samples between valid modes, and because the distillation loss regresses each local actor onto the conditional mean of the teacher's output given local input, this error is not absorbed but propagated to the student. To remove this teacher-side artifact, we propose Mode-Support Semi-Discrete Optimal Transport (MoSDOT), which summarizes multimodal replay into a finite mode support with prescribed capacities and uses conditional semi-discrete optimal transport to assign each noise sample to a single mode before teacher training. We additionally study a shared-randomness variant that uses a shared noise component at execution to expose the residual gap intrinsic to strict-product execution. On controlled diagnostics and offline MARL benchmarks, MoSDOT improves endpoint quality and routing consistency, particularly on datasets exhibiting multimodal joint behavior.
| Comments: | Accepted at NeurIPS 2026 (Main Track, Poster) |
| Subjects: | Machine Learning (cs.LG); Multiagent Systems (cs.MA); Robotics (cs.RO) |
| Cite as: | arXiv:2610.10087 [cs.LG] |
| (or arXiv:2610.10087v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.10087 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Sangmin Lee [view email]
[v1]
Wed, 7 Oct 2026 13:48:27 UTC (1,351 KB)
来源：arXiv：cs.LG · arxiv.org