An Improved Learning Algorithm with Tunable Kernels for Complex-Valued Radial Basis Function Neural Networks

摘要

In this paper, as an extension of real-valued orthogonal least-squares regression with tunable kernels (OLSRTK), a complex-valued OLSRTK is presented which can be used to construct a suitable sparse regression model. In order to enhance the real-valued OLSRTK, the random traversal process and method of filtering center are adopted in complex-valued OLSRTK. Then, the complex-valued OLSRTK is applied to train complex-valued radial basis function neural networks. Numerical results show that better performance can be achieved by the developed algorithm than by the original real-valued OLSRTK.