Escaping local minima with local derivative-free methods: a numerical investigation

We investigate the potential of applying a state-of-the-art, local derivative-free solver, Py-BOBYQA to global optimization problems. In particular, we demonstrate restarts procedure – as distinct from multistart methods allow escape minima (where ordinarily it would terminate at first minimum found). also introduce an adaptive variant which yields improved performance on As is model-based trust-region method, compare largely with other for (global) models are important, such Bayesian and response surface methods; consider state-of-the-art representative deterministic stochastic codes, DIRECT CMA-ES. find numerically that procedures in effective helping minima, when compared using no Py-BOBYQA. Additionally, has comparable solvers all accuracy/budget regimes, both smooth noisy settings. variants best performing multiplicative noise problems high-accuracy regimes. by-product, some preliminary conclusions can be drawn relative have tested default