Sites Inria

Version française



Best paper at GECCO conference

The Genetic and Evolutionary Computation Conference (GECCO) conference is the major and most selective yearly conference on evolutionary computation.  The paper "Local-Meta-Model CMA-ES for Partially Separable Functions'', of Zyed Bouzarkouna (IFP and Tao project-team), Anne Auger (Tao project-team) et Didier Yu Ding (IFP), obtained the best paper award in the Evolution Strategies / Evolutionary Programming track.

In the context of black-box optimization of numerical objective functions, stochastic search algorithms explore the search space in a random manner and face the so-called curse of dimensionality: the search space volume grows exponentially with the dimension. Consequently problems solved rarely have more than a few hundred of variables. When the problem has some structure like partial separability, exploiting this structure allows to break this curse of dimensionality and hence solve problems with a larger number of variables. This paper is the first to propose an efficient way to exploit partial separability as problem structure within the CMA-ES optimization algorithm coupled with meta-models. It opens the way for other research in this direction.

Keywords: Evolutionary computation Tao Saclay - Île-de-France Anne Auger