14h00 - 15h00
Which Gaussian Process for Bayesian Optimization?
Bayesian Optimization (BO) is a popular approach to the global optimization of costly non-convex functions in moderate dimension. BO is based on Gaussian processes that are iteratively learned and serve as a model to control the exploration-exploitation trade-off through an acquisition criterion. Much of the recent research on BO has focused on new acquisition criteria and the specialization to specific problems (e.g., uncertain or multi-fidelity contexts). In this talk, on the contrary, we will consider a standard black-box single objective problem and a standard acquisition criterion, the expected improvement. The focus will be the Gaussian process (GP) and how it can be modified to improve the optimization. Three directions for progress will be discussed: the trend of the GP, the adaptation to higher dimension by variable selection, and the collaboration between the global and a local GP.