Automatic generation, validation and utilization of a knowledge-base: An artificial neural network approach

Thesis/Dissertation · Sun Jan 01 04:00:00 EST 1989

OSTI ID:6037472

Yoon, YoungOhc

Generation of a knowledge base has proved difficult in the development of rule-based expert systems, due to the requirement for explicit rules. This problem arises because much of human knowledge is implicit, especially expert knowledge. It is known that a domain in which knowledge is implicit can not render a clear and complete specification of decision rules. The knowledge acquisition phase may result in the loss of critical information in casting implicit knowledge into explicit rules. Moreover, most expert systems, whose knowledge is in the form of explicit rules, are unable to add or modify an existing knowledge base from experience. Therefore, an alternative method is needed to extract the domain knowledge from experts and to encode it into a knowledge base automatically. The study proposes an artificial neural network (ANN) as an alternative method to construct a knowledge base for the purpose of mitigating these difficulties. This proposal offers several methodologies for building, validating, and utilizing an expert system with an ANN. First, methods are designed to generate the hypothetical examples for the purpose of reducing the amount of time in acquiring a sufficient number of real examples. Second, representation methods of a knowledge based network are proposed. Third, methods are presented to obtain realistic estimates of the performance of a connectionist system and to measure the stability of a chosen model. Fourth, two explanation methods are proposed in order to make an ANN system eligible to be an expert system. Through presentation of these explanation methods, the study overcomes a major limitation of connectionist expert systems, since this system can explain decisions and the underlying knowledge base in the way conventional rule-based systems can. An experimental expert system is developed with proposed methodologies in a medical domain.

• Research Organization: Texas Univ., Arlington, TX (USA)
• OSTI ID: 6037472
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
ARTIFICIAL INTELLIGENCE
AUTOMATION
BENCH-SCALE EXPERIMENTS
DATA
DECISION MAKING
DESIGN
EXPERT SYSTEMS
INFORMATION
KNOWLEDGE BASE
NEURAL NETWORKS
NUMERICAL DATA
PERFORMANCE TESTING
SELECTION RULES
TECHNOLOGY UTILIZATION
TESTING
THEORETICAL DATA
VALIDATION