University of Jyväskylä

Visit Lecture: At the Crossroads between Computational Geometry and Multiobjective Optimization. Associate Professor Michael Emmerich, Leiden University, Netherlands

Start date: Jun 20, 2017 10:00 AM

End date: Jun 20, 2017 12:00 PM

Location: Mattilanniemi, Ag C 421.1

Prof. Emmerich will give a visiting lecture on June 20 at 10-12 in Ag C 421.1. He will talk about their applied projects and give a brief overview of his group's activities at the eginning of the talk. Please find below his abstract and a brief bio.

You are warmly welcome!


Algorithms in multiobjective optimization are concerned with computations on point sets in the objective space, which is typically the m-dimensional Euclidean space.  The understanding of the geometrical nature of the Pareto dominance order, diversity indicators, visualization techniques, and the computation of certain integrals over the non-dominated space are essential when assessing the computational feasibility of certain optimization algorithm components and visualization techniques.  Examples are the computation of attainment surfaces, the computation of hypervolume indicators and related integrals over the dominated space, the computation of Hausdorff distance on level sets, and subset selection problems with respect to such indicators. Related to this are gradient computations for indicators and integral computations, such as the probability of improvement and multicriteria expected in surrogate assisted optimization.  In this talk latest findings in the complexity theory of geometrical computations in multiobjective optimization will be discussed, and it will provide some examples for whiich exact results are known and how they affect algorithm design.  In particular we will look at some interesting differences and commonalities that  occur at the transition from 2-D to 3-D case.  For more than 4 dimensions sharp results on complexity are widely unavailable and the talk will summarize the latest findings on upper and lower complexity bounds. 

Averaged Hausdorff distance, expected, epsilon-indicator, hypervolume improvement, hypervolume indicator. 

Short bio:

Dr. Michael Emmerich is Associate Professor at LIACS, Leiden University, and leader of the Multicriteria Optimization and Decision Analysis (MODA) research group. He was born in 1973 in Coesfeld (Germany) and received his doctorate in 2005 from Technical University of Dortmund (Hans.-Paul Schwefel and Peter Buchholz, promoters). He carried out projects as a researcher at Center of Applied Systems Analysis/ICD e.V. (Germany), Dept. Computer Science, TU Dortmund, IST Lisbon, University of the Algarve (Portugal), ACCESS Material Science e.V. (Germany), and the FOM/AMOLF institute on Fundamental Science of Matter (Netherlands). He is known for pioneering work on surrogate model-assisted and indicator-based multiobjective optimization,. Dr. Emmerich has co-authored more than 30 research articles in the listed journals, and 70 research papers in conference proceedings, of which 5 papers received the best paper award. He has been scientific organizer of three international Lorentz center workshops on multiobjective optimization and on set oriented numerics. Moreover, he recently joined the steering committee of the Evolutionary Multi-Criterion Optimization conference and since 2005 he is teaching a master computer science course on "Multiobjective Optimization and Decision Analysis" at Leiden University.


Filed under: