Recursively Adapted Radial Basis Function Networks and its Relationship to Resource Allocating Networks and Online Kernel Learning

This paper proposes a recursively adapted radial basis function network and provides additional insights into several well-known techniques such as radial basis function networks, resource allocating networks and stochastic gradient descent in reproducing kernel Hilbert spaces. Through this perspective, resource allocating networks are investigated in a more principled way so that issues of convergence and generalization can be mathematically analyzed in the least mean square framework.