Rule-extraction by backpropagation of polyhedra

The core problem of rule-extraction from feed-forward networks is an inversion problem. In this article, we solve this inversion problem by backpropagating unions of polyhedra. We obtain as a by-product a new rule-extraction technique for which the fidelity of the extracted rules can be made arbitrarily high.

[1]  Huan Liu,et al.  NeuroLinear: From neural networks to oblique decision rules , 1997, Neurocomputing.

[2]  Alexander Schrijver,et al.  Theory of linear and integer programming , 1986, Wiley-Interscience series in discrete mathematics and optimization.

[3]  Sebastian Thrun,et al.  Inversion in Time , 1990, EURASIP Workshop.

[4]  Gene H. Golub,et al.  Matrix computations , 1983 .

[5]  Joachim Diederich,et al.  Survey and critique of techniques for extracting rules from trained artificial neural networks , 1995, Knowl. Based Syst..

[6]  David K. Smith.  Theory of Linear and Integer Programming , 1987 .