Thursday 24 Oct 2013: The Geometry of Evolutionary Algorithms
Dr Alberto Moraglio - Computer Science, University of Exeter
Harrison 103 15:00-16:00
The various flavors of Evolutionary Algorithms look very similar when cleared of algorithmically irrelevant differences such as domain of application and phenotype interpretation. Representation-independent algorithmic characteristics like the selection scheme can be freely exchanged between algorithms. Ultimately, the origin of the differences of the various flavors of Evolutionary Algorithms is rooted in the solution representation and relative genetic operators. Are these differences only superficial? Is there a deeper unity encompassing all Evolutionary Algorithms beyond the specific representation? Is a general mathematical framework
unifying search operators for all solution representations at all possible? The talk will introduce a formal, but intuitive, unified point of view on Evolutionary Algorithms across representations based on geometric ideas, which provides a possible answer to the above questions. It also presents the benefits for both theory and practice brought by this novel perspective.
The geometric view on search operators has a number of interesting consequences that the talk will overview. These include (i) a straightforward view on the fitness landscape seen by recombination operators, (ii) a formal unification of many pre-existing search operators across representations, (iii) a principled way of designing crossover operators for new representations, (iv) a principled way of generalizing search algorithms from continuous to combinatorial spaces, and (v) the potential for a unified theory of evolutionary algorithms across representations.