Нова Модель Штучного Нейрону для Побудови Нейронних Мереж у Задачі Апроксимації

Vladyslav Gorbatiuk; Olena Chumachenko

New Artificial Neuron Model for Building Neural Networks in Approximation Tasks

Authors

Vladyslav Gorbatiuk dept. of technical cybernetics NTUU «Igor Sikorsky KPI»
Olena Chumachenko dept. of technical cybernetics NTUU «Igor Sikorsky KPI»

Keywords:

approximation, neural networks, artificial neuron, ReLU neuron

Abstract

A new artificial neuron model is presented, which generalizes and in certain way improves known models that belong to ReLU “family”: ReLU, PReLY and maxout models. Its theoretical advantages in regards to mentioned known models are analyzed. Experimental comparison tests on real data sets were conducted, confirming the approximation effectiveness of the suggested model.