Title: Impact of mixed-variable management by probability features in an Evolutionary Algorithm
Abstract: The management of mixed variables, i.e., of different natures (continuous, integer, discrete, categorical) in optimization problems is a challenge, especially in the presence of categorical variables since there is no concept of order between their possible “values”. In a black-box optimization context, a way to handle categorical variables has been proposed within a particle swarm algorithm to improve its performance. It consists in defining a probability feature for each “value” of each categorical variable which will evolve during the optimization process. These probabilities are then used to update the population of the algorithm.
In this talk, we present an adaptation of this method to be used in the evolutionary algorithm (EA) implemented in Minamo, the in-house design space exploration and multi-disciplinary optimization platform developed in the applied research center Cenaero. More specifically, we propose several ways to use these probabilities inside the genetic operators of the EA and we compare them in order to get a modified version of the EA which is more efficient in the presence of categorical variables.
The seminar will take place in Room S08 at the Faculty of Sciences.