Title :
Modified EBP algorithm with instant training of the hidden layer
Author :
Wilamowski, Bogdan M.
Author_Institution :
Dept. of Electr. Eng., Wyoming Univ., Laramie, WY, USA
Abstract :
Several algorithms for training feedforward neural networks including the steepest decent EBP (error backpropagation) and Lavenberg-Marquardt are compared. Various techniques to improve convergence of the EBP are also reviewed. A very fast training algorithm, with instant training of the hidden layer is introduced. For easy problems it has a similar convergence rate as the Lavenberg-Marquardt (LM) method. The algorithm sustains the fast convergence rate also for the cases when the LM algorithm fails and the EBP algorithm has practically unacceptable slow convergence rate
Keywords :
backpropagation; feedforward neural nets; Lavenberg-Marquardt algorithm; convergence rate; feedforward neural networks; hidden layer; instant training; steepest decent error backpropagation; Backpropagation algorithms; Convergence; Feedforward neural networks; Jacobian matrices; Least squares methods; Neural networks; Neurons; Newton method; Stability; Stochastic processes;
Conference_Titel :
Industrial Electronics, Control and Instrumentation, 1997. IECON 97. 23rd International Conference on
Conference_Location :
New Orleans, LA
Print_ISBN :
0-7803-3932-0
DOI :
10.1109/IECON.1997.668437
