On the performance of lazy matching in production systems

Abstract

Production systems are an established method for encoding knowledge in an expert system. The semantics of produc-tion system languages and the concomitant algorithms for their evaluation, RETE and TREAT, enumerate the set of rule instantiations and then apply a strategy that selects a single instantiation for firing. Often rule instantiations are calculated and never fired. In a sense, the time and space re-quired to eagerly compute these unfired instantiations is wasted. This paper presents preliminary results about a new match technique, lazy matching. The lazy match algorithm folds the selection strategy into the search for instantiations, such that only one instantiation is computed per cycle. The algorithm improves the worst-case asymptotic space com-plexity of incremental matching. Moreover, empirical and