A self-adaptive neural network for blind signal separation and experiment
Xue Hu · Journal of Hefei University of Technology · 2002
A novel online learning algorithm with selfadaptive learning rates for blind separation of signals is presented to improve the convergence rate. The KullbackLeibler divergence is used as cost function,and the standard stochastic gradient descent method adopted. The learning rate is adjusted during the learning process according to a set of differential equations. Computer simulation results show that the independent sources can be extracted from the hybrid mixture of image signals by using the presented algorithm.