报告人:香港中文大学John C.S. Lui教授,IEEE/ACM Fellow
报告时间:2019年10月20日(周日) 上午10:00
报告地点:信息科学与工程学院 原106教室(现220)
报告摘要:
The Shapley value is a cornerstone in cooperative game theory and also has been widely applied in machine learning and data science, etc. However, it is also known to be computationally expensive when the game includes many players. Since the cooperation preference is an important factor for a variety of real life applications, we first “generalize” the classic Shapley value to capture the cooperation preference. More specifically, we develop mathematical models to solicit two types of cooperation preferences: (1) group-wise preferences and (2) pair-wise preferences. We then extend the Shapley value to capture theses preferences. Secondly, we design computationally efficient randomized algorithms to approximate our generalized Shapley value with theoretical guarantees on accuracy. To show the utility of our framework, we demonstrate how to apply it to divide rewards in a crowdsourcing system, and divide the revenue among ISPs in deploying new Internet architectures. We also provide insights as to how preference may influence the reward (or revenue) division.
报告人简介:
吕自成教授目前是香港中文大学计算机科学与工程系李卓敏荣誉教授,他于UCLA获得计算机博士学位,随后加入了IBM实验室,参与了有关文件系统和并行I/O架构的研发项目,后来加入港中文计算机科学与工程系。吕教授曾是UCLA、哥伦比亚大学、马里兰大学学院公园分校、普渡大学、马萨诸塞大学阿姆斯特分校和意大利的都灵大学的访问学者。研究方向聚焦于机器学习在网络科学、网络经济学、网络/系统安全、大规模分布式系统和性能测评理论方面的研究与应用。吕教授获得了诸多教学与科研方面的奖项,包括香港中文大学校长模范教学奖和香港中文大学职员杰出研究奖(2011-2012)。他获得了IFIP WG 7.3 Performance 2005, IEEE/IFIP NOMS 2006,SIMPLEX'14,ACM RecSys’17等重要国际会议的最佳学术论文奖以及ACM Mobihoc’18和ASONAM’17会议的最佳论文提名奖。吕教授是IFIP WG 7.3,ACM,IEEE等多个重要协会的会士,Croucher基金会的高级研究专家,以及现任的ACM SIGMETRICS会议主席。