Fast Computational Approach to the Levenberg-Marquardt Algorithm for Training Feedforward Neural Networks

Abstract This paper presents a parallel approach to the Levenberg-Marquardt algorithm (LM). The use of train neural networks is associated with significant computational complexity, and thus computation time. As result, when network has big number weights, becomes practically ineffective. article new computations in learning algorithm. proposed solution based on vector instructions effectively reduce high time this was tested several examples involving problems classification function approximation, next it compared classical method. detail idea shows obtained acceleration for different problems.