Proxifield：基于语义邻近度的去中心化多智能体通信
原标题：Proxifield: Decentralized Multi-Agent Communication through Semantic Proximity
AI 摘要
该论文提出 Proxifield，一种去中心化的多智能体通信协议，通过智能体演化中的语义邻近度构建稀疏通信图，无需模型训练或集中式规划器，在推理时利用直接寻址、信息需求、计划对齐和信息互补四种路由信号连接智能体。在无人机搜救和 HiddenBench 集体推理两个领域，Proxifield 在最大规模模型（397B）下优于 Star 和 Shared Context 基线，且随团队规模增大优势从 5.4% 扩大到 59.5%。在智能体永久失效条件下，Proxifield 保留 73.6% 的无故障任务奖励，显著高于 Shared Context 的 58.3% 和 Star 的 38.8%。
正文节选
Proxifield: Decentralized Multi-Agent Communication through Semantic Proximity Abstract As LLM capabilities have expanded, multi-agent communication has emerged as an increasingly active area of research. Prevailing protocols often adopt rigid structures that introduce coordination bottlenecks and can degrade as the number of agents increases. We introduce Proxifield, a round-adaptive multi-agent protocol with decentralized agent decision-making that constructs sparse communication graphs from t