Comparison of Several Stochastic and Deterministic Derivative-Free Global Optimization Algorithms
In this paper popular open-source solvers are compared against Globalizer solver, which is developed at the Lobachevsky State University. The Globalizer is designed to solve problems with black-box objective function satisfying the Lipschitz condition and shows competitive performance with other similar solvers. The comparison is done on several sets of challenging multi-extremal benchmark functions. Also this work considers a method of heuristic hyperparameters control for the Globalizer allowing to reduce amount of initial tuning before optimization. The proposed scheme allows substantially increase convergence speed of the Globalizer by switching between “local” and “global” search phases in runtime.
KeywordsDeterministic global optimization Stochastic global optimization Algorithms comparison Derivative-free algorithms Black-box optimization Multi-extremal problems
- 8.Johnson, S.G.: The NLOpt nonlinear-optimization package. http://ab-initio.mit.edu/nlopt. Accessed 24 Dec 2018
- 9.Jones, E., Oliphant, T., Peterson, P., et al.: SciPy: open source scientific tools for Python (2001–). http://www.scipy.org/. Accessed 24 Dec 2018
- 12.Kennedy, J., Eberhart, R.: Particle swarm optimization. In: Proceedings of ICNN 1995 - International Conference on Neural Networks, vol. 4, pp. 1942–1948 (1995)Google Scholar
- 15.Madsen, K., Zertchaninov, S.: A new branch-and-bound method for global optimization (1998)Google Scholar
- 24.Strongin, R.: Numerical Methods in Multiextremal Problems (Information-Statistical Algorithms). Nauka, Moscow (1978). (In Russian)Google Scholar