A fast and robust pattern recognition using a new algorithm for training feed-forward neural networks

A fast and robust algorithm is presented for training multilayer feedforward neural networks as an alternative to the backpropagation algorithm. The number of iterations required by the new algorithm to converge is less than 10% of what is required by the backpropagation algorithm. Also, it is less affected by the choice of initial weights and setup parameters. The algorithm uses a modified form of the backpropagation algorithm to minimize the mean-squared error between the desired and actual outputs with respect to the inputs to the nonlinearities. This is in contrast to the standard algorithm which minimizes the mean-squared error with respect to the weights. The new algorithm is called “Predictor of Linear Output” (PLO), in terms of its function. The estimated linear signals, generated by the modified backpropagation algorithm, are used to produce an updated set of weights through a system of linear equations at each node