Research On Unsupervised Neural Network Based On Information Theory Learning

Self-organizing map neural network and neural gas network are two typical unsupervised learning feedforward neural networks.They have strong self-organizing,and self-learning abilities.However,the experimental results show that those two kinds of neural.networks still have some shortcomings,e.g.they may achieve poor performance when there are noise in the given training samples.In order to improve the anti-noise ability of self-organizing map neural network and neural gas network,the unsupervised neural networks based on information theory learning are studied in this thesis.Two strategies for improving the unsupervised neural networks are proposed in the thesis.The main works are as follows:1.Self-organizing mapping neural network based on correntropy is proposed.The proposed method is utilizes correntropy to replace the Euclidean distance measure,which makes the improved network model have better performance.In addition,for the optimization problem of the proposed method,the semi-quadratic optimization technique is used to solve the problem.The experimental results verify that the proposed method has better performance.2.A improve robust neural gas network is proposed.In the proposed method,the M-estimator is used to replace the Euclidean distance between the input vector and the weight vector in the objective function of the traditional neural gas network,and the optimization problem based on the M-estimator is solved by the gradient descent method.The experimental results show that the proposed method has stronger robustness.