Multi-gradient: A fast converging and high performance learning algorithm

In this paper, we propose a new learning algorithm for multilayer neural networks. In the backpropagation learning algorithm, weights are adjusted to reduce the error or cost function that reflects the difference between the computed and desired outputs. In the proposed learning algorithm, we consider each term of the output layer as a function of weights and adjust the weights directly so that the output layers produce the desired outputs. Experiments show the proposed algorithm consistently performs better than the back-propagation learning algorithm.