Ver Post

On return and risk in evolutionary multiobjective optimization

6 Março 2012, 09:19 - António Sérgio Constantino Folgado Ribeiro

Monday, March 12, 2012 | 14:00 - 15:00 | QA1.3 (Alameda Campus) | Free and Open to All

Title: On return and risk in evolutionary multiobjective optimization

Abstract:In this work, the task of selecting a diverse subset of (non-dominated) solutions from a larger set of candidate solutions according to Decision Maker (DM) preference information in evolutionary algorithms is reinterpreted as a (financial) portfolio selection problem.Fitness assignment may then be performed by finding an optimal,risk-adjusted portfolio of candidate solutions, e.g., based on the Sharpe-ratio performance index, which amounts to solving a convex quadratic programming problem in the simplest case.

One particular instance of this general paradigm combines Fonseca and Fleming's preferability relation with the hypervolume indicator in order to arrive at a goal-driven, diversity-promoting, combined fitness-assignment and bounded-archiving procedure for evolutionary multiobjective optimization (EMO) algorithms. The resulting evolutionary multiobjective optimizer is then evaluated against two existing optimizers, namely NSGA II and SMS-EMOA, on a number of multiobjective knapsack problem instances. Experimental results show that the proposed approach is highly competitive on the problems studied, and motivate further research on the connection between risk modelling and diversity promotion in EMO.

Speaker:Carlos M. Fonseca, joint work with Iryna Yevseyeva and Michael Emmerich

Affiliation:Associate Professor at the Department of Informatics Engineering and member of the Centre for Informatics and Systems of the University of Coimbra, Portugal