Training multilayer perceptrons parameter by parameter

A new fast training algorithm for multi-layer perceptrons (MLP) is presented. This new algorithm, named parameter by parameter optimization algorithm (PBPOA), is proposed based on the idea of layer by layer (LBL) algorithm. The inputs errors of output layer and hidden layer are taken into consideration. Four classes of solution equations for parameters of networks are deducted respectively. The presented algorithm doesn't need the calculation of the gradient of error function at all. In each iteration step, the weight or threshold can be optimized directly one by one with other variables fixed. Effectiveness of the presented algorithm is demonstrated by two benchmarks, in which faster convergence rate of training are obtained in contrast with the BP algorithm with momentum (BPM) and the conventional LBL algorithm.