Title :
An analytical approach based on information theory for neural network architecture
Author_Institution :
Dept. of Electr. Eng., Istanbul Tech. Univ., Turkey
Abstract :
In this study on the neural network architecture following its training period, with only one hidden layer and some constraints, the number of hidden nodes have been calculated by using the concepts of mean information quantity which was defined as an entropy, and the importance of sigmoid function has been emphasized as the necessary condition of analytical approach used.
Keywords :
entropy; information theory; neural net architecture; neural nets; parallel architectures; analytical approach; constraints; entropy; hidden layer; hidden nodes; information theory; mean information quantity; necessary condition; neural network architecture; sigmoid function; training period; Entropy; Information analysis; Information theory; Neural networks; Uncertainty;
Conference_Titel :
Neural Networks, 1993. IJCNN '93-Nagoya. Proceedings of 1993 International Joint Conference on
Print_ISBN :
0-7803-1421-2
DOI :
10.1109/IJCNN.1993.713919
