| Abstract |
In this talk I will introduce the Gene-pool Optimal Mixing Evolutionary Algorithm (GOMEA) and illustrate its advantageous properties over classical "blind" Evolutionary Algorithms (EAs). GOMEA belongs to the class of Model-Based EAs (MBEAs) and focuses particularly on efficiently learning and exploiting so-called linkage models that describe dependencies between the variables that are used to encode solutions to the optimization problem at hand. Especially when combined with the so-called Linkage Tree (LT) model, GOMEA exhibits highly robust performance on well-known benchmark problems, often achieving polynomial scalability of a lower order than other well-known state-of-the-art EAs. Recent work expands the GOMEA as originally introduced for classical binary representations to other domains, including permutations, real values, and tree-based genetic programming for both single- and multi-objective optimization problems. In all cases, the excellent performance of GOMEA is maintained. I will in particular consider solving optimization problems both from a Black-Box Optimization (BBO) perspective where nothing (is assumed) to be known about the problem being solved as well as from a Grey-Box Optimization (GBO) perspective where partial evaluations are possible. Especially in the latter case, GOMEA is capable of obtaining (near-)optimal results for problems with millions of variables in less than an hour on a normal desktop computer. I will end the talk by presenting the projects that my subgroup of Medical Informatics at CWI is currently involved in, outlining how EAs in general, and GOMEAs in particular, are used to solve real-world problems, in collaboration with hospitals and other academic partners. |