Incremental neural network construction by using correlation as cost function

Authors:
X. X. Wang;D. J. Brown
Affiliations:
Intelligent Systems and Diagnostics Group, Department of Electronic and Computer Engineering, University of Portsmouth, UK;Intelligent Systems and Diagnostics Group, Department of Electronic and Computer Engineering, University of Portsmouth, UK
Venue:
Design and application of hybrid intelligent systems
Year:
2003

Citing 6
Cited 0

The Strength of Weak Learnability

Machine Learning
Bayesian interpolation

Neural Computation
Constructive incremental learning from only local information

Neural Computation
Neural Networks for Pattern Recognition

Neural Networks for Pattern Recognition
A decision-theoretic generalization of on-line learning and an application to boosting

EuroCOLT '95 Proceedings of the Second European Conference on Computational Learning Theory
Decision trees can initialize radial-basis function networks

IEEE Transactions on Neural Networks

Quantified Score

Hi-index	0.01

Visualization

Abstract

Incremental neural network construction by using Gaussian function is studied in this paper. Correlation between the Gaussian function and the training data is used as the cost function to fit each neuron. Compared to least square cost function, correlation cost function pays more attention to the peak area of the Gaussian function which is important in incremental learning method. So the correlation cost function can strengthen the local property of Gaussian like Radian Basis Functions. In addition, a weighted optimisation method based on the AdaBoost algorithm is proposed and used in neuron position and shape determination. Compared to the traditional gradient-based method, it has the advantage of being easy implemented and can be applied where the cost function is non-smooth. In addition, it is robust because there is no matrix inverse solving problem in the algorithm. The experimental results based on these algorithms are also given in the paper.