A machine learning approach for modeling and its applications
Shuxiang Xu, Yunling Liu, Byeong Ho Kang, Wanlin Gao · eCite Digital Repository (University of Tasmania) · 2013
This paper proposes a new learning algorithm for Higher Order Neural Networks for the purpose of modelling and applies it in three benchmark problems. Higher Order Neural Networks (HONNs) are Artificial Neural Networks (ANNs) in which the net input to acomputational neuron is a weighted sum of its inputs and products of its inputs (rather than just a weighted sum of its inputs as in traditional ANNs). It was well known that HONNs can implement invariant pattern recognition. The new learning algorithm proposed is an Extreme Learning Machine (ELM) algorithm. ELM randomly chooses hidden neurons and analytically determines the output weights. With ELM algorithm only the connection weights between hidden layer and output layer are adjusted. This paper proposes an ELM algorithm forHONN models and applies it in an image processing problem, a medical problem, and an energy efficiency problem. The experimental results demonstrate the advantages of HONN models with the ELM algorithm in such aspects as significantly faster learning and improvedgeneralization abilities (in comparison with standard HONN and traditional ANN models).