Search academic texts

Information × Registration Number 0223U002296, 0121U000191 , R & D reports Title Development of GMDH neural networks with active neurons for inductive models construction of complex systems popup.stage_title Head Moroz Olha G, Кандидат технічних наук Registration Date 14-02-2023 Organization International Research and Training Center of Information Technologies and Systems UNESKO IIP popup.description2 It was found that the GMDH algorithm of the neural network type prevails over the main deterministic deep neural networks of direct propagation, as it has a dynamic number of layers, which is adjusted by the network in the learning process and is devoid of retraining. Among the main disadvantages of the GMDH neural network are the following: exponential growth of the degree of the polynomial in the quadratic partial description, the possibility of losing informative and consolidating non-informative arguments in the model. To eliminate them, it is proposed to implement the idea of introducing neurons of different structure, which is optimized by a hybrid combinatorial-genetic algorithm  Product Description popup.authors Moroz Olga H popup.nrat_date 2023-02-14 Close