CAREER: Learning to Cooperate and Compete

该奖项支持对一系列项目的研究，所有这些项目都解决了个人或团体在包含其他个人和团体的环境中学习提高生产率的问题。涵盖的多智能体学习场景的范围从完全合作的小组，其中每个人的唯一目标是为群体利益服务，到纯粹竞争的情况，其中个人学习利用竞争对手的弱点。在另一个维度上，该奖项既对待个人主义的学习方案，其中每个代理人自己学习，也对待共享的学习方法，其中两个或更多的代理人积极参与相互学习。为了在合作小组中实现共享学习，小组基础设施是必要的。该奖项下的工作将研究团队架构和灵活的承诺框架，使代理能够根据环境动态的需要，在可预测性和响应性之间做出明智的决定。

The award supports research on a collection of projects, all of which address the problem of individuals or groups learning to be more productive in an environment containing other individuals and groups. The gamut of multi-agent learning scenarios covered ranges from completely cooperative groups, where the sole goal of each individual is to server group interests, to purely competitive situations in which individuals learn to exploit the weaknesses of competitors. On another dimension, the award treats both individualistic learning schemes, where each agent learns by itself, and shared learning approaches, where two or more agents actively participate in learning about each other. To enable shared learning in cooperative groups, a group infrastructure is necessary. Work under this award will study a group architecture and a flexible commitment framework that allows agent to make sensible decision about predictability-versus-responsiveness tradeoffs as necessitated by environmental dynamics.