Derivative-Free Optimization I

WC-03: Derivative-Free Optimization I
Stream: Derivative-Free Optimization
Room: Nash
Chair(s): Massimo Roma

Regret Bounds for Noise-Free Bayesian Optimization
Sattar Vakili
Bayesian optimisation is a powerful method for non-convex black-box optimization in low data regimes. However, the question of establishing tight performance bounds for common algorithms in the noiseless setting remains a largely open question. We establish new and tightest bounds for two algorithms, namely GP-UCB and Thompson sampling, under the assumption that the objective function is smooth in terms of having a bounded norm in a Matérn RKHS. Furthermore, we do not assume perfect knowledge of the kernel of the Gaussian process emulator used within the Bayesian Optimization loop.

On some Bayesian optimization recipes applied to mono-multi-fidelity constrained black box problems
Nathalie Bartoli, Thierry Lefebvre, Youssef Diouane, Sylvain Dubreuil, Joseph Morlier
This work aims at developing new methodologies to optimise computational costly complex systems (e.g., aeronautical engineering systems). The proposed surrogate-based method (often called Bayesian Optimization) uses adaptive sampling to promote a trade-off between exploration and exploitation. Our in-house implementation, called SEGOMOE, handles a high number of design variables and nonlinearities by combining mixtures of experts (local surrogate models ) for the objective and/or the constraints. An extension to multi-fidelity is also included when a variety of information is available.

On the use of Derivative-Free Optimization in design Discrete Event Simulation models for Emergency Departments
Massimo Roma, Alberto De Santis, Tommaso Giovannelli, Stefano Lucidi, Mauro Messedaglia
The most widely used tool for studying patient flow through an Emergency Department (ED) is Discrete Event Simulation (DES). However, to achieve high reliability of such a DES model, an accurate calibration procedure is firstly required in order to determine a good estimate of the model input parameters. In this work we propose a Simulation-Optimization approach integrating DES models with a Derivative-Free Optimization method in order to determine the best values of model input parameters. The approach we propose, has been widely experimented on a large ED in Rome.

Theme: Overlay by Kaira