A Novel Adaptive Kernel for the RBF Neural Networks
This work addresses the need for more effective kernel methods in neural networks for estimation problems, though it appears incremental as it builds on existing RBF frameworks.
The paper tackled the problem of improving radial basis function neural networks by proposing a novel adaptive kernel that fuses Euclidean and cosine distances, dynamically adjusting weights via gradient descent, and it outperformed manual fusion methods on nonlinear system identification, pattern classification, and function approximation tasks.
In this paper, we propose a novel adaptive kernel for the radial basis function (RBF) neural networks. The proposed kernel adaptively fuses the Euclidean and cosine distance measures to exploit the reciprocating properties of the two. The proposed framework dynamically adapts the weights of the participating kernels using the gradient descent method thereby alleviating the need for predetermined weights. The proposed method is shown to outperform the manual fusion of the kernels on three major problems of estimation namely nonlinear system identification, pattern classification and function approximation.