Abstract: A technique for collecting changes to working memory objects made by rule execution in an artificial intelligence production system avoids frequent use of a matching algorithm by delaying the match processing of the collected changes to a memory object until completion of an executing rule. A change to an object wrought by execution of a rule is signified in a control block for that object. Once a first change has occurred, subsequent changes caused before execution of the rule is complete will be made to the object and indicated by the change block. When execution of the rule is complete, the changes coalesced in the object itself are registered in the system by introduction of the changed object into the matching algorithm. This avoids match processing the object each time it is changed during execution of the rule.

Abstract: A demand-driven AI production system utilizing a RETE network for comparison matching in a condition/data match, rule-selection, and rule-firing execution cycle in which the RETE network is modified to maintain a list of instantiations satisfying the match conditions expressed in each node of the RETE network, passing of tokens to descendant nodes upon a comparison match, maintaining patterns to all ancestor nodes through which the tokens have passed, and traversing the patterns as a path for avoiding those RETE pattern matchings redundant between a previous match and a current match in progress.