Evolutionary Computation and Decision Making
Evolutionary Computation and Decision Making
Keynote talk: Mariano Luque, Professor of Mathematics in Economics, University of Malaga, Spain
TITLE
An introduction to the preference-based and interactive EMO algorithms.
ABSTRACT
Evolutionary Multiobjective Optimization (EMO) algorithms have become very popular for solving multiobjective optimization problems in the last decades. Their purpose was to approximate the whole set of Pareto optimal solutions, that is, the Pareto optimal front (PF), with a diverse set of well-distributed non-dominated solutions (i.e., they are a posteriori methods). EMO algorithms have widely demonstrated their capabilities in handling multiple objective functions and their effectiveness in solving real-world applications involving problems with different characteristics (variables of different nature, non-convex or discontinuous functions, etc.). Despite this, identifying the most preferred solution (MPS) in the approximation set generated by an EMO algorithm may not be easy for a decision maker (DM) . To overcome this, and also to reduce the computational resources, preference information can be incorporated into the search process of EMO algorithms to only approximate a part of the PF defined by the DM’s preferences. This results either in preference-based EMO algorithms (if the preferences are given initially, as in a priori methods), or in interactive EMO algorithms (if the preferences are gradually given in an iterative way). This allows for shortening the computational time needed per run or iteration compared to EMO algorithms, thus reducing long waiting times when interacting with the DM if the time available for deciding is limited.
In this presentation, we will see an introduction to these types of EMO algorithms, preference-based and interactive EMO algorithms. To do so, we will point out which types of preferences are most commonly used, and how the set of solutions that reflect these preferences are usually computed. In many of these cases, the algorithms used are adapted versions of existing EMO algorithms to approximate the whole PF. We will pay special attention to cases based on reference points or aspiration levels, as this is the most commonly used type. Finally, key aspects to apply these techniques to real world problems will be discussed.
SHORT BIOGRAPHY
Mariano Luque was born in Madrid (Spain), in 1972. He is Full Professor in Quantitative Methods in Economics since 2015, in the Department of Applied Economics (Mathematics), University of Málaga (SPAIN). He holds a degree in Mathematics (1995) with an academic average score of 9 out of 10, a Bachelor Thesis in Mathematics (1997) and a PhD in Quantitative Methods in Economics (2000). His research is focused on multicriteria techniques, especially interactive methods, reference point-based methods, interval programming and evolutionary algorithms. He works on applications in the fields of economics of education, sustainability issues, workers’ satisfaction, electricity generation, etc. He has published about sixty papers JCR in journals such as Evolutionary Computation, Swarm and Evolutionary Computation, European Journal of Operational Research, Omega, Management Science, etc. He has been a Visiting Researcher at several prestigious universities and institutions, such as the Manchester Business School and Helsinki School of Economics. He has also given several talks at different universities such as the University of Manchester, University of Jyvaskyla, University of Valencia, etc. In addition, he has participated and led numerous research projects (european, national and regional).