Analysis and synthesis of feedforward neural networks using discrete affine wavelet transformations Neural Networks IEEE Transactions on英文电子书.pdf

IEEE TRANSACTIONS ON NEURAL NETWORKS. VOL. 1.NO. I . JANUARY I 9 Y i 73 Analysis and Synthesis of Feedforward Neural Networks Using Discrete Affine Wavelet Transformations Y. C. Pati and P. S. Krishnaprasad, Fellow, IEEE Abstract-In this paper we develop a representation of a class networks have demonstrated a somewhat miraculous ability to of feedforward neural networks in terms of discrete affine wavelet “learn” (closely approximate) the desired map. A number of transforms. It is shown that by appropriate grouping of terms, rigorous mathematical proofs have been provided to explain feedforward neural networks with sigmoidal activation functions can be viewed as architectures which implement affine wavelet the ability of feedforward neural networks to approximate decompositions of mappings. This result follows simply from maps [I], [2], [lo], [ l l ] . Several of these proofs have been the observation that standard feedforward network architectures based on arguments of density, of the class of maps that can possess an inherent translation-dilation structure and every node be implemented by a feedforward network, in various function implements the same activation function. It is shown that the spaces. However, these methods do not naturally give rise wavelet transform formalism provides a mathematical frame- work within which it is possible to perform both analysis and to syste