Determining the Neural Network Topology from the Viewpoint of Kuhn's Philosophy and Popper's Philosophy

摘要

Determining the number of hidden layers and the number of neurons are very important and have a large influence on deep neural network(DNN) performance. In some studies, there is no clear guideline on how to determine the number of hidden layers or neurons optimally; even the roles and functions of both are explained minimally. Although it is difficult, researches to determine the number of hidden layers and neurons must continue to be carried out, because both will greatly determine the performance of DNN. According to Kuhn, the method for determining neural network topology in deciding the number of hidden layers and hidden neurons is still in pre-paradigm phase. New studies continue to be made in an effort to find methods that can be generally accepted, so that they will become normal sciences. The proposed new methods can be tested by using Popper's falsification which will determine whether the methods can eventually become normal sciences or not.