Mj. Palakal et Av. Hudli, AUGMENTING PARALLEL REASONING MECHANISMS WITH LEARNING NETWORKS IN EXPERT-SYSTEMS, Engineering applications of artificial intelligence, 8(5), 1995, pp. 515-525
Efficient reasoning mechanisms and a facility to learn the reasoning p
rocess can significantly enhance the performance of expert systems. Th
is paper proposes parallel algorithms that implement the reasoning mec
hanisms, and a Learning Network that can learn the patterns of the rea
soning process in art Expert System Model (ESM). As a result, either b
ackward or forward reasoning cart be accomplished with a high degree o
f parallelism, and the recall of a learned conclusion can be accomplis
hed in constant time. The parallel algorithms are implemented using ar
tificial neural networks. The parallel implementation of the algorithm
makes use of several orders of simple neural networks, each realizing
a simple task. These simple neural networks are organized vertically,
thereby achieving a second level of parallelism. The paper presents a
novel way to handle both forward and backward chaining reasoning mech
anisms in parallel. The secondary neural network model (the learning n
etwork) monitors and learns the reasoning patterns carried out by the
reasoning networks. Once adequately learned, this learning network can
generate reasoning based on ''experience''. Generating reasons based
on experience eliminates the need for unnecessary search and significa
ntly enhances the response time. Several simulation results show that
both parallel realization and learning of the reasoning process can si
gnificantly improve the performance of expert systems.