An improved infomax algorithm of independent component analysis applied to fMRI data

Independent component analysis (ICA) is a technique that attempts to separate data into maximally independent groups. Several ICA algorithms have been proposed in the neural network literature. Among these algorithms applied to fMRI data, the Infomax algorithm has been used more widely so far. The Infomax algorithm maximizes the information transferred in a network of nonlinear units. The nonlinear transfer function is able to pick up higher-order moments of the input distributions and reduce the redundancy between units in the output and input. But the transfer function in the Infomax algorithm is a fixed Logistic function. In this paper, an improved Infomax algorithm is proposed. In order to make transfer function match the input data better, the we add a changeable parameter to the Logistic function and estimate the parameter from the input fMRI data in two methods, 1. maximizing the correlation coefficient between the transfer function and the cumulative distribution function (c.d.f), 2. minimizing the entropy distance based on the KL divergence between the transfer function and the c.d.f. We apply the improved Infomax algorithm to the processing of fMRI data, and the results show that the improved algorithm is more effective in terms of fMRI data separation.