Partitions of unity improve neural function approximators

Neural function approximators with localized receptive fields are sometimes riddled with disturbing interpolation artifacts. A general principle is proposed to remove these defects. Such approximators should be designed as partitions of unity within their domains. This principle explains earlier empirical results, and its effectiveness is demonstrated by the removal of spurious interpolation artifacts of a radial basis functions (RBF) network. Using well-known partitions of unity, further improvements can be easily obtained. This is demonstrated by converting the piecewise constant functions of standard cerebellar model articulation controller (CMAC) nets into arbitrary smooth functions (C/sup infinity /-CMACs).<<ETX>>

[1]  D.A. Handelman,et al.  Theory and development of higher-order CMAC neural networks , 1992, IEEE Control Systems.

[2]  S. Lang.  Introduction to Differentiable Manifolds , 1964 .

[3]  David H. Wolpert,et al.  A Mathematical Theory of Generalization: Part II , 1990, Complex Syst..

[4]  C. Micchelli.  Interpolation of scattered data: Distance matrices and conditionally positive definite functions , 1986 .

[5]  T Poggio,et al.  Regularization Algorithms for Learning That Are Equivalent to Multilayer Networks , 1990, Science.

[6]  Ken-ichi Funahashi,et al.  On the approximate realization of continuous mappings by neural networks , 1989, Neural Networks.

[7]  George Cybenko,et al.  Approximation by superpositions of a sigmoidal function , 1992, Math. Control. Signals Syst..

[8]  John C. Platt.  Leaning by Combining Memorization and Gradient Descent , 1990, NIPS.

[9]  D. Hammerstrom,et al.  Neural networks at work , 1993, IEEE Spectrum.

[10]  David H. Wolpert,et al.  A Mathematical Theory of Generalization: Part I , 1990, Complex Syst..

[11]  Andrew G. Barto,et al.  Connectionist learning for control: an overview , 1990 .

[12]  M. J. D. Powell,et al.  Radial basis functions for multivariable interpolation: a review , 1987 .

[13]  Kurt Hornik,et al.  Multilayer feedforward networks are universal approximators , 1989, Neural Networks.

[14]  James S. Albus,et al.  New Approach to Manipulator Control: The Cerebellar Model Articulation Controller (CMAC)1 , 1975 .