荷蘭萊頓大學Michael Emmerich博士報告通知
承辦單位：國際智能感知與計算研究中心
“智能感知與圖像理解”教育部重點實驗室
國家111計劃創新引智基地
IEEE西安分會
IET西安分會
報告地點：主樓Ⅱ-227報告廳
報告時間：2014年7月11日 9:00
報告題目：Recent Advances in Set Oriented and Multiobjective Optimization
報告人： Michael Emmerich
報告簡介
There are many scenarios where it is desirable to search for a set of alternative solutions instead of searching for a single best solution. For instance, such problems occur in portfolio selection, multiobjective optimization, and level set approximation. A rigorous method of algorithm design is to establish first a performance indicator that specifies an order among different solution sets and then design an algorithm that seeks to optimize this indicator. In the talk this paradigm will be exemplified for different scenarios, that is receiver operator curve performance optimization (a joint work with Xi’an and Hefei University), Pareto front approximation and the approximation of level sets (or sets of constraint-satisfying solutions). The rigorous measures (area under convex hull, free hypervolume indicator, and, respectively, average Hausdorff distance) are used in the design of efficient gradient-based and evolutionary algorithms for the approximation of sets. Examples from drug discovery, biological network identification, and machine learning will show the usefulness of these methods in practical scenarios.
專家簡介
Dr. Michael Emmerich, Leiden Institute of Advanced Computer Science, Leiden University, The Netherlands.
Dr. Michael Emmerich graduated in 1999 from the Technical University of Dortmund and after working for several years as a research consultant for the German chemical industry, and, since 2003 as a research manager in the Collaborative Research Center Computational Intelligence, he received his PhD in 2005 from Dortmund University. His promoters were Prof. Hans-Paul Schwefel (Evolutionary Optimization) and Prof. Peter Buchholz (Quantitative Systems, Markov Processes). Since 2005, Michael Emmerich is appointed as Assistant Professor in the Computer Science Department of Leiden University (Algorithms and Natural Computing Group), and also worked during intermediate appointments in other institutes such as Instituto Superior Technológico , Lisbon (Portugal) and FOM/AMOLF in Amsterdam.
Dr. Michael Emmerich’s research topic are the design and analysis of stochastic and deterministic algorithms for different classes of black-box optimization. Moreover he is an expert in machine learning, in particular Bayesian methods such as Gaussian Processes, and complex networks for dynamical systems analysis. He is co-developer for well established algorithms for multiobjective optimization (SMS-EMOA), which introduced hypervolume-based selection in evolutionary algorithms. He has worked on the development for applied algorithmic solutions for problems in science and engineering, in particular in the domain of drug discovery and process optimization.
