Scaffolding the RETE network

Abstract

Scaffolding is a novel method for improving conjunctive match efficiency that incorporates a TMS (truth maintenance system) into an RETE matcher, thereby exploiting match redundancy between separate runs. The resulting inactivations are reactivations of the data dependencies replace expensive RETE testing with simpler, less costly NETL-like propagations. Experimental results obtained by profiling with a non-toy expert system substantiate scaffolding's between-trial transfer capability. The experiments demonstrate that there can be significant transfer of match knowledge between runs on different, homologous problems. The construction and retrieval of these matches are efficiently built into the algorithm. Scaffolding's utility and relation to other artificial-intelligence methods are discussed.<>