Authors:
Bruno Barracosa | EDF R&D | France
Julien Bect | Université Paris-Saclay, CNRS, CentraleSupélec, Laboratoire des signaux et systèmes | France
Héloïse Baraffe | EDF R&D | France
Juliette Morin | EDF R&D | France
Gilles Malarange | EDF R&D | France
Emmanuel Vazquez | Université Paris-Saclay, CNRS, CentraleSupélec, Laboratoire des signaux et systèmes | France
Interest in multi-objective simulation-based optimization—in other words, multi-criteria optimization based on stochastic simulators—can be traced back to at least the mid-70s. But it is only quite recently that algorithms aiming to provide an estimate of the entire Pareto set, and/or Pareto front, started to appear. In situations where individual simulations have a non-negligible run time (or cost), one cannot hope to obtain an arbitrarily accurate estimation of these sets given a limited budget of time (or money). Bayesian optimization, where stochastic (often Gaussian) process models are used to quantify the optimizer's uncertainty about the properties of the simulator, is a natural candidate for such situations. This talk will review and compare existing Bayesian methods for multi-objective simulation-based optimization, and propose some new ideas.