两篇新研究论文介绍了用于训练多智能体系统（MAS）的新型框架，它们超越了简单的单次编排。MAS-Loop 专注于闭合编排器和优化器之间的反馈循环，利用执行轨迹诊断和修复故障，从而提高性能，并为 MAS 奖励建模提供了新数据集。R^2Orch 采用递归自我对抗，重点关注可学习的任务选择和技能撤回，使编排器能够在没有手工设计的课程的情况下进行训练，并在各种推理和交互任务上取得最先进的成果。 AI
影响 这些框架提供了训练更强大、更适应性强的多智能体系统的新方法，有可能提高复杂推理和交互任务的性能。
排序理由 两篇介绍多智能体系统新框架的学术论文。
在 arXiv cs.MA (Multiagent) 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
arXiv:2610.38946v1 Announce Type: cross Abstract: Training-time multi-agent system (MAS) design seeks to learn reusable orchestration policies, yet prevailing approaches remain largely one-shot: an Orchestrator generates a system but only observes and learns from its terminal out…
Training-time multi-agent system (MAS) design seeks to learn reusable orchestration policies, yet prevailing approaches remain largely one-shot: an Orchestrator generates a system but only observes and learns from its terminal outcome. This paradigm discards execution trajectorie…
R²Orch uses recursive self-play with learnable task selection and skill withdrawal to train orchestrators without hand-designed curricula, improving performance across reasoning and interactive tasks.