A weight initialization method for improving training speed in feedforwardneural network

An algorithm for determining the optimal initial weights of feedforward neu ral networks based on the Cauchy's inequality and a linear algebraic method is developed. The algorithm is computational efficient. The proposed metho d ensures that the outputs of neurons are in the active region and increase s the rate of convergence. With the optimal initial weights determined, the initial error is substantially smaller and the number of iterations requir ed to achieve the error criterion is significantly reduced. Extensive tests were performed to compare the proposed algorithm with other algorithms. In the case of the sunspots prediction, the number of iterations required for the network initialized with the proposed method was only 3.03% of those s tarted with the next best weight initialization algorithm. (C) 2000 Elsevie r Science B.V. All rights reserved.