报 告 人：金耀初 (Yaochu Jin)教授欧洲科学院院士、IEEE Fellow Bielefeld University

报告时间：2023年3月20日下午15：30-16:30

报告地点：信息学管228

邀 请 人：刘建昌 教授

Abstract：This talk begins with a brief introduction to heterogeneous multi-objective optimization and its application background. Then, some recent ideas of enhancing the efficiency of Bayesian optimization of heterogeneous multi-objective problems by transferring knowledge from computationally cheap objectives to expensive objectives will be presented. We show that knowledge transfer can be achieved by sharing hyperparameters in Gaussian processes and generating synthetic data by either building a co-surrogate between the two objectives or using domain adaptation. Extensive empirical studies verify the effectiveness of various knowledge transfer methods in Bayesian heterogeneous multi-objective optimization.

报告人简介：金耀初 (Yaochu Jin)教授目前为德国比勒费尔德大学工程学院“洪堡人工智能教席教授”，兼任英国萨里大学计算机系“计算智能”讲席教授，欧洲科学院院士，IEEE Fellow。曾任芬兰国家创新局“Finland Distinguished Professor”、澳大利亚悉尼科技大学“杰出访问学者”、东北大学“长江学者特聘教授”。目前担任IEEE Computational Intelligence Society主席和Complex & Intelligent Systems主编，曾担任IEEE Transactions on Cognitive and Developmental Systems主编。长期从事人工智能与系统科学的理论、算法和工程应用研究，特别是数据驱动的复杂系统进化优化、进化多目标机器学习、联邦学习与安全机器学习、演化发育系统与形态发育机器人学等。金耀初教授已发表学术论文500余篇，获美国、欧盟和日本专利9项。据Google Scholar，其论文被引用总次数38,000余次，h-index 97，入选Web of Science 2019-2022年度“全球高被引科学家”。