BriefGPT.xyz
Mar, 2022
用对比学习实现分散式多智能体通信的学习
Learning to Ground Decentralized Multi-Agent Communication with Contrastive Learning
HTML
PDF
Yat Long Lo, Biswa Sengupta
TL;DR
研究使用自我监督学习的方法，通过最大化给定轨迹信息的消息之间的互信息，使用一种新的视角诱导出一个共同语言，在通信关键的环境中取得了更好的学习表现和速度，以及学习出比现有方法更一致的共同语言，而且不需要引入额外的学习参数。
Abstract
For communication to happen successfully, a
common language
is required between agents to understand information communicated by one another. Inducing the
emergence
of a
→