Integration of magnified gradient function and weight evolution with deterministic perturbation into back-propagation
Sin-Chun Ng, Chi-Chung Cheung, S.H. Leung · Electronics Letters · 2003
An integrated approach of magnified gradient function and weight evolution with deterministic perturbation to improve the performance of back-propagation learning is proposed. Simulation results show that, in terms of the convergence rate and the percentage of global convergence, the integrated approach always outperforms the other traditional methods.