Approximation with Diffusion-Neural-Network

Neural information processing models largely assume that the samples for training a neural network are sufficient. Otherwise there exist a non-negligible error between the real function and estimated function from a trained network. To reduce the error in this paper we suggest a diffusion-neural-network (DNN) to learn from a small sample. First, we show the principle of information diffusion using properties of quasitriangular fuzzy numbers. After that, we apply this principle to construct the DNN. Finally, we give an example to show that the approximation with DNN is better than the conventional back propagation network.