CAREER: Decision Making and Learning in Dynamic Multiagent Systems

职业：动态多智能体系统中的决策和学习

• 批准号: 0133845
• 负责人: Junling Hu
• 金额: $ 29.97万
• 依托单位: University of Rochester
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2002
• 资助国家: 美国
• 起止时间: 2002-03-01 至 2004-02-29
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0133845&HistoricalAwards=false
• 关键词: CAREER; Decision; Making; Learning; Dynamic

This project investigates two important elements in intelligent multiagent systems: decision making and learning. Based on a theoretical model of dynamic games, the PI will investigate a new optimality criterion for decision making. Second, the PI will define a non-stationary strategy within a dynamic game framework, providing a language to describe the policy choices. Third, the PI will provide a framework to describe the interaction between learning and action choice, and redefine the concept of online learning. Fourth, the PI will search for efficient online learning algorithms to reach an optimal non-stationary strategy. Fifth, this research on non-stationary strategy learning will be linked to research on models for dynamic single-agent systems. Finally, the proposed methodology and algorithms will be implemented in agent applications for E-commerce domains and computer games. Potential applications include bidding agents for online auctions, pricebots for online retailers and computer characters for strategic games. The proposed research is integrated with the PI's education plan, which focuses on promoting the concept of intelligence for multiagent systems. The PI will teach decision making and learning in dynamic games in her Ph.D. seminar class, and construct a graduate course on agent design for multiagent systems and their applications in electronic commerce.

该项目研究智能多智能体系统中的两个重要元素：决策和学习。基于动态博弈的理论模型，PI 将研究一种新的决策最优标准。其次，PI将在动态博弈框架内定义非平稳策略，提供描述政策选择的语言。第三，PI将提供一个框架来描述学习和行动选择之间的交互，并重新定义在线学习的概念。第四，PI将寻找有效的在线学习算法以达到最佳的非平稳策略。第五，非平稳策略学习的研究将与动态单智能体系统模型的研究联系起来。最后，所提出的方法和算法将在电子商务领域和计算机游戏的代理应用程序中实现。潜在的应用包括在线拍卖的投标代理、在线零售商的价格机器人和战略游戏的计算机角色。拟议的研究与 PI 的教育计划相结合，该计划的重点是推广多智能体系统的智能概念。 PI 将在她的博士课程中教授动态游戏中的决策和学习。研讨会班，并构建多代理系统代理设计及其在电子商务中的应用研究生课程。