Comparison of the MATSuMoTo library for expensive optimization on the noiseless black-box Optimization Benchmarking testbed

Numerical black-box optimization problems occur frequently in engineering design, medical applications, finance, and many other areas of our society's interest. Often, those problems have expensive-to-calculate objective functions for example if the solution evaluation is based on numerical simulations. Starting with the seminal paper of Jones et al. on Efficient Global Optimization (EGO), several algorithms tailored towards expensive numerical black-box problems have been proposed. The recent MATLAB toolbox MATSuMoTo (short for MATLAB Surrogate Model Toolbox) is the focus of this paper and is benchmarked within the Black-box Optimization Benchmarking framework BBOB. A comparison with other already previously benchmarked algorithms for expensive numerical black-box optimization with the default setting of MATSuMoTo highlights the strengths and weaknesses of MATSuMoTo's cubic radial basis functions surrogate model in combination with a Latin Hypercube initial design in the range of 50 times dimension many function evaluations.