ADVANCED ALGORITHM OF EVOLUTION STRATEGIES OF COVARIATION MATRIX ADAPTATION

The paper considers the extension of CMA-ES algorithm using mixtures distributions for finding optimal hyperparameters neural networks. Hyperparameter optimization, formulated as optimization black box objective function, which is a necessary condition automation and high performance machine learning approaches. an efficient without derivatives, one alternatives in combination hyperparameter methods. developed based on assumption multi-peak density distribution parameters complex systems. Compared to other methods, computationally inexpensive supports parallel computations. Research results show that can be competitive, especially concurrent assessment mode. However, much broader more detailed comparison still needed, will include test tasks various modifications, such adding constraints. Based Monte Carlo method, it was shown new improve search by average 12%.