Numerical Aspects of Extreme Learning Machine Implementation to Regression Problems

Abstract

An Extreme Learning Machine (ELM) – a neural network with fixed hidden layer and adjustable output weights is able to solve complicated regression (approximation) problems, but the standard selection of input weights and biases may lead to ill-conditioning of the output weights calculation and result in high values of the output weights. Two modifications of standard ELM are discussed: deterministic generation of hidden nodes parameters and modifications of activation functions to improve numerical properties of the algorithm.