A neural network training algorithm utilizing multiple sets of linear equations

A fast algorithm is presented for the training of multilayer perceptron neural networks. In each iteration, there are two passes through the training data. In the first pass, linear equations are solved for the output weights. In the second data pass, linear equations are solved for hidden unit weight changes. Full batching is used in both data passes. An algorithm is described for calculating the learning factor for use with the hidden weights. It is shown that the technique is significantly faster than standard output weight optimization-backpropagation.