Title :
A convenient method to prune multilayer neural networks via transform domain backpropagation algorithm
Author_Institution :
Dept. of Electron. Eng., Jiao Tong Univ., Shanghai, China
Abstract :
It is proved that the transform domain backpropagation (BP) algorithm with a variable learning rate is an effective algorithm for accelerating the convergence of a multilayer neural network. It is shown that the transform domain BP algorithm can also be applied to prune neural networks conveniently and to accelerate the convergence to some extent. This is based on the fact the correlation within the input pattern of every layer can be removed via an orthogonal transform
Keywords :
backpropagation; neural nets; correlation; multilayer neural networks; pruning; transform domain backpropagation; variable learning rate; Acceleration; Algorithm design and analysis; Backpropagation algorithms; Convergence; Discrete transforms; Frequency; Multi-layer neural network; Neural networks; Neurons;
Conference_Titel :
Neural Networks, 1992. IJCNN., International Joint Conference on
Conference_Location :
Baltimore, MD
Print_ISBN :
0-7803-0559-0
DOI :
10.1109/IJCNN.1992.227051
