A New Training Method for Feedforward Neural Networks Based on Geometric Contraction Property of Activation Functions

TL;DR

This paper introduces a novel training approach for feedforward neural networks that leverages the geometric contraction property of activation functions, leading to faster learning and improved classification accuracy.

Contribution

It presents a new training method that simplifies the network's nonlinearity by removing the activation function's nonlinearity at the output layer, enhancing training efficiency.

Findings

Faster learning speed in experiments

Lower classification error rates

Effective for activation functions with geometric contraction property

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.