Title :
Updating learning rates for backpropagation network
Author_Institution :
Dept. of Marine Technol., Newcastle upon Tyne Univ., UK
Abstract :
A new approach for improving the convergence rate of backpropagation network is proposed in the paper. This method updates the learning rate parameter for each individual weight before each weight is updated. Simulation on the XOR problem shows that when compared to the conventional backpropagation algorithm, the improved algorithm reduces the number of training iterations and CPU time by up to seventy and fifty times, respectively.
Keywords :
backpropagation; computational complexity; neural nets; XOR problem; backpropagation neural network; convergence rate; training iterations; updating learning rates; Adaptive systems; Backpropagation algorithms; Equations; Marine technology; Stability;
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.1993.713979
