Title :
Periodic activation function for fast on-line EKF training and pruning
Author :
Chang, Sheng-Jiang ; Wong, Kwok-Wo ; Leung, Chi-sing
Author_Institution :
Dept. of Electron. Eng., City Univ. of Hong Kong, Kowloon Tong, Hong Kong
fDate :
11/12/1998 12:00:00 AM
Abstract :
In the use of the local extended Kalman filter algorithm for network training and pruning, the authors propose replacing the monotonical increasing sigmoid activation function with a periodic function. Computer simulations show that the rate of convergence is substantially improved and that more insignificant weights can be pruned
Keywords :
Kalman filters; convergence; feedforward neural nets; learning (artificial intelligence); computer simulation; convergence; feedforward neural network; local extended Kalman filter algorithm; on-line training; periodic activation function; pruning;
Journal_Title :
Electronics Letters
DOI :
10.1049/el:19981554
