1980 Reports

Learning Meta-Rule Control of Production Systems from Execution Traces

Harrison, Malcolm C.; Stolfo, Salvatore

In the last decade, work in Artificial Intelligence has stressed the importance of having both declarative and procedural forms of knowledge available. Recently, a number of workers have pointed out that procedural information can be regarded as control information which can be used to control the sequencing of application of the declarative components, and have remarked on the advantages of keeping these two types of information separate [see, for example, Sigart Newsletters #63, June 1977 and #70, Feb. 1980]. In the last few years we have been looking at the problem of automatically inferring this control component from the behaviour of the declarative component [7]. In this paper we describe briefly some experiments we have done, discuss some of the difficulties we have encountered, and show how they might be overcome.