Title :
A study of removing hidden neurons in cascade-correlation neural networks
Author :
Liang, Xun ; Ma, Long
Author_Institution :
Inst. of Comput. Sci. & Technol., Peking Univ., Beijing, China
Abstract :
In this paper, the techniques of removing hidden neurons in cascade-correlation neural networks are studied. The removing processes are based on the orthogonal projection and the crosswise propagation (CP) method. They avoid randomly deleting hidden neurons. The linear compression method and the two methods to remove the hidden neurons, which are most possibly redundant are presented and the examples are also given.
Keywords :
neural nets; probability; cascade correlation neural networks; crosswise propagation method; hidden neuron removal process; linear compression method; orthogonal projection; probability; Biological neural networks; Computer science; Erbium; Intelligent networks; Multi-layer neural network; Multilayer perceptrons; Network topology; Neural networks; Neurons; Vectors;
Conference_Titel :
Neural Networks, 2004. Proceedings. 2004 IEEE International Joint Conference on
Print_ISBN :
0-7803-8359-1
DOI :
10.1109/IJCNN.2004.1380073