GAUSS: An Automatic Algorithm Selection System for Quadrature

This paper introduces GAUSS a system for the automatic selection of quadrature routines in numerical computation. Given a problem in numerical integration and constraints on time and accuracy, GAUSS comes up with an efficient algorithm to solve it. It banks on a performance database of various algorithms and test problems, an automatic feature identification module and a knowledge methodology that represents information about the numerical domain in terms of predicate rules. Inductive Logic Programming is used to induce learning and make inferences in this domain. Experimental Results are presented. It is found that GAUSS comes up with several heuristics that have been earlier formulated by the domain experts.