Construction of neural networks for piecewise approximation of continuous functions

A feedforward neural network structure which can be directly constructed to approximate arbitrary continuous functions is proposed. This neural network is devised by introducing a space tessellation which is a covering of the Euclidean space by nonoverlapping hyperpolyhedral convex cells. The plastic weights of the proposed neural network can be calculated to implement the mapping for the training data. This reduces training time and alleviates the difficulties of local minima in training. The piecewise local interpolation capability of the proposed network improves the performance in generalization for new data