报告题目：Recent Advances in Infill Criteria for Surrogate-Assisted Evolutionary Optimization

报 告 人：金耀初教授（英国萨里大学计算科学系“计算智能”首席教授）

报告时间：2018年5月2日上午9：30-10.30

报告地点： 安徽大学磬苑校区理工D楼318室

主办单位：安徽大学计算机科学与技术学院

欢迎各位老师、同学届时前往！

科学技术处

2018年4月28日

报告摘要：

Infill criteria have been shown to be effective for model management in Gaussian process assisted evolutionary optimization of low-dimensional expensive problems. However, they are prevented from being applied to high-dimensional problems due to two serious problems. First, the computational complexity of constructing Gaussian processes increases dramatically when the number of training samples increases. Second, the uncertainty information estimated by the Gaussian process becomes less reliable, degrading the effectiveness of the infill criteria. This talk presents a brief account of recent advances in addressing the above challenges in surrogate-assisted evolutionary optimization of high-dimensional problems. Heterogeneous ensemble learning models are generated to replace the Gaussian process to improve the scalability of computational complexity for high-dimensional problems. In addition, a multi-objective infill criterion has been proposed to replace existing single objective infill criteria. Empirical results demonstrate the effectiveness of the proposed methods.
报告人简介：

金耀初分别于1988、1991及1996年在浙江大学电机系获学士、硕士及博士学位，并于2001年在德国波鸿鲁尔大学神经信息研究所获工学博士学位。现为英国萨里大学计算科学系“计算智能”首席教授, “自然计算与应用”研究组主任，萨里大学“数学与计算生物学中心”共同负责人。金耀初博士是中组部“****”专家，教育部“****”讲座教授，芬兰国家技术创新局“芬兰讲座教授”。目前担任《IEEE认知与发育系统汇刊》主编， Springer《复杂与智能系统》共同主编，IEEE 杰出演讲人。IEEE Fellow。已出版专/编著及会议论文集9本，发表学术论文200余篇。论文被引用总次数13500余次 (据Google Scholar）, 其中SCI引用4200余次，h-index 为55。获美国、欧盟和日本专利共9项。先后在30多个国际会议上作特邀大会或主题报告。