An improved back-propagation/Cauchy machine network

Abstract

To overcome the shortcomings of the backpropagation (Bp) algorithm, namely, slow convergence, local minimum, and paralysis problems, a combined backpropagation/Cauchy (Bp/Cauchy) machine has been proposed by Wasserman (1990). In this paper, a switching condition is introduced to improve the backpropagation/Cauchy machine network. To illustrate the effectiveness of the proposed method, examples of xor and the learning of a unknown function are included. Results show that the improved Bp/Cauchy machine is more effective in learning than the original Bp/Cauchy machine.<>