Building localized basis function networks using context dependent clustering

Abstract

Networks based on basis set function expansions, such as the Radial Basis Function (RBF), or Separable Basis Function (SBF) networks, have non-linear parameters that are not trivial to optimize. Clustering techniques are frequently used to optimize positions of localized functions. Context-dependent fuzzy clustering techniques improve convergence of parameter optimization, leading to better networks and facilitating formulation of prototype-based logical rules that provide low-complexity models of data. © Springer-Verlag Berlin Heidelberg 2008.