Предложен подход к формированию ансамблей нейросетей, основанный на кодировании нескольких архитектур в одном бинарном дереве с применением генетического программирования.
An approach to the design of neural network ensembles is proposed, which is based on the coding of several architectures in one binary tree using genetic programming. The method provides automatic optimization of the structure and composition of the ensemble. As a result of experiments on classification and regression problems, it was found that the method shows efficiency comparable to or superior to alternative models in most cases.