A New Training Method for Feedforward Neural Networks Based on Geometric Contraction Property of Activation Functions
Neural and Evolutionary Computing
2018-08-14 v2 Machine Learning
Abstract
We propose a new training method for a feedforward neural network having the activation functions with the geometric contraction property. The method consists of constructing a new functional that is less nonlinear in comparison with the classical functional by removing the nonlinearity of the activation function from the output layer. We validate this new method by a series of experiments that show an improved learning speed and better classification error.
Keywords
Cite
@article{arxiv.1606.05990,
title = {A New Training Method for Feedforward Neural Networks Based on Geometric Contraction Property of Activation Functions},
author = {Petre Birtea and Cosmin Cernazanu-Glavan and Alexandru Sisu},
journal= {arXiv preprint arXiv:1606.05990},
year = {2018}
}