Multi-Kernel Fusion for RBF Neural Networks

Abstract A simple yet effective architectural design of radial basis function neural networks (RBFNN) makes them amongst the most popular conventional networks. The current generation network is equipped with multiple kernels which provide significant performance benefits compared to previous using only a single kernel. In existing multi-kernel RBF algorithms, formed by convex combination base/primary kernels. this paper, we propose novel RBFNN in every base kernel has its own (local) weight. This flexibility provides better such as faster convergence rate, local minima and resilience against stucking poor minima. These gains are achieved at competitive computational complexity contemporary algorithms. proposed algorithm thoroughly analysed for gain mathematical graphical illustrations also evaluated on three different types problems namely: (i) pattern classification, (ii) system identification (iii) approximation. Empirical results clearly show superiority state-of-the-art approaches.