Neurons-Samples Theorem
Guannan Liu · International Journal of Computational Methods · 2022
Artificial neural network (NN) has become one of the most widely used machine learning (ML) models for problems in science and engineering, including the fast-developing artificial intelligence (AI) technology. In training an NN model for a problem, one of the most frequently asked questions is how many neurons or layers of neurons should be used for a given dataset with a number of samples (or data points). This paper provides an answer to this critical question, by presenting a Neurons-Samples Theorem, which states, in short, that the number of neurons should be equal or less than the number of samples used to train the NN.