Immunity clone strategy based ICA
Fang Liu, Fan Jia · 2004
In this paper, immunity clone algorithm is introduced to the learning of separating matrix in independent component analysis (ICA). For no restriction on the derivative of objective function, we can derive the matrix without help of traditional gradient descend algorithms. An objective function with a smaller error to approximate the negentropy is adopted. This algorithm has the advantages of simpleness, stable and global convergence, and is verified with computer simulation.