EN FR
EN FR


Section: New Results

Classification algorithms in Bayesian optimization

Participants : Régis Duvigneau, Matthieu Sacher [Ecole Navale] , Frédéric Hauville [Ecole Navale] , Olivier Le Maître [CNRS-LIMSI] .

A Gaussian-Process based optimization algorithm is proposed to efficiently determine the global optimum for expensive simulations, when some evaluations may fail, due to unrealistic configurations, solver crash, degenerated mesh, etc. The approach is based on coupling the classical Bayesian optimization method with a classification algorithm, to iteratively identify the regions where the probability of failure is high.

The method is applied to the optimization of foils and sails in the context of racing yachts [14], [18], [23], [28], in particular for the America's Cup in collaboration with Groupama team. This work is part of M. Sacher's PhD work at Ecole Navale.