Blackbox: A procedure for parallel optimization of expensive black-box functions

TL;DR

This paper introduces Blackbox, a parallel optimization procedure for expensive black-box functions that uses response surface methodology with radial basis functions, initial Latin hypercube sampling, and a modified CORS algorithm, scalable on multicore processors.

Contribution

It presents a novel parallel optimization method combining response surface modeling and space rescaling, with an implementation in Python for efficient black-box function optimization.

Findings

Effective scaling on multicore processors

Utilizes radial basis functions for response modeling

Incorporates Latin hypercube sampling and modified CORS algorithm

Abstract

This note provides a description of a procedure that is designed to efficiently optimize expensive black-box functions. It uses the response surface methodology by incorporating radial basis functions as the response model. A simple method based on a Latin hypercube is used for initial sampling. A modified version of CORS algorithm with space rescaling is used for the subsequent sampling. The procedure is able to scale on multicore processors by performing multiple function evaluations in parallel. The source code of the procedure is written in Python.