Title :
The hidden layer design of the MVQ neural network
Author :
Abouali, A.H. ; Porter, W.A.
Author_Institution :
Egyptian Res. Center, Cairo, Egypt
Abstract :
We introduce the first part of neural network classifiers design methodology. The design has a lot of the desired features. The design is based on a preprocessing stage of the multiple class vector quantization (MVQ) algorithm. The algorithm extracts the information from the training set. The outcome of this stage fully defines the first hidden layer of the network. The methodology not only has better performance but also provides insights to why and how the neural network works
Keywords :
learning (artificial intelligence); multilayer perceptrons; neural net architecture; vector quantisation; hidden layer design; multiple class vector quantization neural network; neural network classifiers; training set; Algorithm design and analysis; Backpropagation; Data mining; Design methodology; Fuzzy sets; Nearest neighbor searches; Neural networks; Neurons; Process design; Vector quantization;
Conference_Titel :
System Theory, 1998. Proceedings of the Thirtieth Southeastern Symposium on
Conference_Location :
Morgantown, WV
Print_ISBN :
0-7803-4547-9
DOI :
10.1109/SSST.1998.660103
