Abstract
We introduce a novel black box optimization method, which we term Polynomial-Model-Based Optimization (PMBO). The main idea of this algorithm originates from Bayesian Optimization (BO) [16], however, instead of inferring a stochastic process, a polynomial surrogate of the objective function is fitted and exploited in the regions where the objective is expected to minimize. We compare PMBO with state-of-the-art optimization algorithms for common test functions and a real-world optimization problem. The results demonstrate PMBO to be capable of inferring the inherent structure of the objective function, resulting in its efficiency. This suggests that PMBO might be the pivotal choice for the optimization of highly structured real-world problems.

This publication has 16 references indexed in Scilit: