ABSTRACT
This chapter proposes a multi-input Bernoulli-polynomial neuronet (MIBPN) on the basis of function approximation theory. The MIBPN is trained by a weights-and-structure-determination (WASD) algorithm with twice-pruning (TP). The WASD algorithm can obtain the optimal weights and structure for the MIBPN, and overcome the weaknesses of conventional back-propagation (BP) neuronets such as slow training speed and local minima. The chapter focuses on the non-orthogonal polynomials so as to extend the investigation. It analyses the model of FIBPN is presented and the theoretical basis. The number of hidden neurons and the training time of the two BP neuronets are set to be the same as those of the proposed FIBPN for each target function so as to compare the two BP neuronets with the FIBPN under the same conditions. In order to determine the optimal neuronet structure, a corresponding TP-aided WASD algorithm has been proposed based on the previously presented weights-direct-determinationweights-direct-determination subalgorithm.
