Unsupervised Hebbian learning in neural networks

In this paper, a survey of a particular class of unsupervised learning rules for neural networks is presented. These learning rules are based on variants of Hebbian correlation learning to update the connection weights of two-layer network architectures consisting of an input layer with n units and an output layer with m units. It will be demonstrated that the networks are able to perform a variety of important data analysis tasks, including Principal Component Analysis (PCA), Minor Component Analysis (MCA) and Independent Component Analysis (ICA).