Independent component analysis works in many application. This paper is submitted for improve the performance of previous algorithm. There is many parameters in ICA network. Unfortunately, some of them is selected in problem dependent. Learning rate is one of them. This paper propose the new adaptive learning algorithm.
Suggested algorithm is two method. One is for scalar type learning algorithm. The other is vector type learning algorithm. The basis of the algorithm is maximum entropy. They improve the separated output's quality. On-line ICA's performance is improved in adaptive manner.