Anti-Hebbian rule for faster backpropagation learning

This paper introduces an algorithm to speed up the backpropagation learning rules. The algorithm is based on providing lateral connections among the neurons of every hidden layer. These connections are trained using an anti-Hebbian learning rule which decorrelates the outputs of these nodes. The decorrelation process minimizes any redundant information transferred from an internal layer to the next and therefore enables the network to capture the statistical properties of the mapping much faster. The algorithm is applied to some benchmark problems and the results are compared to those obtained using the conventional backpropagation network