Multigradient: A new neural network learning algorithm for pattern classification

In this paper, we propose a new learning algorithm for multilayer feedforwa rd neural networks, which converges faster and achieves a better classifica tion accuracy than the conventional backpropagation learning algorithm for pattern classification. In the conventional backpropagation learning algori thm, weights are adjusted to reduce the error or cost function that reflect s the differences between the computed and the desired outputs. In the prop osed learning algorithm, we view each term of the output layer as a functio n of weights and adjust the weights directly so that the output neurons pro duce the desired outputs. Experiments with remotely sensed data show the pr oposed algorithm consistently performs better than the conventional backpro pagation learning algorithm in terms of classification accuracy and converg ence speed.