Optimal Learning with Progressive Accuracy for Function Representations in Orthogonal Wavelet Neural Network (WNN)

This paper illustrates the procedure that takes advantage of the properties of discrete wavelet frames so as to improve the learning efficiency of static model representations. The focus is on using orthonormal basis functions due to its convergence properties and compactly supported in frequency domain. The network trained with stochastic gradient type algorithm is presented. Results obtained for modeling two simulated processes are compared with reported bench mark results and demonstrate the effectiveness of the proposed method.