An adaptable parallel search of knowledge bases with beam search

Abstract

An online computer-supported expert system is described which is designed for searching complex knowledge bases. Searching is performed in parallel with beam search, so that only a limited number of specific features are examined on each level. Parameters for beam search are modified during application of the software package, so that self adapting capability of the beam search through an expert system is incorporated in the design of a package. The proposed approach is especially effective for image processing and speech recognition. The difference between making a sequential program parallel and using natural parallelism is shown. The proposed method on a mesh network of transputers or on a MIMD (multiple-instruction, multiple-data-stream) parallel computer, PARSYS, has the properties of a neural network.<>