XIANG Guoquan, DONG Daozhen. ACTIVATION FUNCTIONS AND A METHOD TO IMPROVE THE TRAINING OF NEURAL NETWORKS BY USING BP ALGORITHMJ. Journal of Computer Research and Development, 1997, 34(2).
Citation:
XIANG Guoquan, DONG Daozhen. ACTIVATION FUNCTIONS AND A METHOD TO IMPROVE THE TRAINING OF NEURAL NETWORKS BY USING BP ALGORITHMJ. Journal of Computer Research and Development, 1997, 34(2).
XIANG Guoquan, DONG Daozhen. ACTIVATION FUNCTIONS AND A METHOD TO IMPROVE THE TRAINING OF NEURAL NETWORKS BY USING BP ALGORITHMJ. Journal of Computer Research and Development, 1997, 34(2).
Citation:
XIANG Guoquan, DONG Daozhen. ACTIVATION FUNCTIONS AND A METHOD TO IMPROVE THE TRAINING OF NEURAL NETWORKS BY USING BP ALGORITHMJ. Journal of Computer Research and Development, 1997, 34(2).
ACTIVATION FUNCTIONS AND A METHOD TO IMPROVE THE TRAINING OF NEURAL NETWORKS BY USING BP ALGORITHM
This paper studies the influence of activation functions in BP algorithms, and comes to a conclusion: The more cliffy the function is ,the quicker the convergence of the algorithm. A new method in BP algorithm is proposed in order to avoid local minimum, that is, add trainning data step by step.
DownLoad: