Multiwavelet neural networks construction study

Multiwavelet neural networks are a new class of artificial neural networks and their activation functions are multiwavelet functions or multi-scale functions. The activation functions in different neural networks including feedforward, radial basis functions, wavelets and multiwavelets are discussed and compared. The relations between multiwavelet transformations and artificial neural network are analyzed in detail. Two constructions of multiwavelet neural networks based on multiwavelet frame and continuous multiwavelet transformation are proposed in the paper, and corresponding learning algorithms are given. In the end, the future problems on multiwavelet neural networks are put forward and discussed.