1 AIT Asian Institute of Technology

Rule extraction from neural networks

AuthorManandhar, Suman K.
Call NumberAIT Thesis no.CS-95-15
Subject(s)Neural networks (Computer science)

NoteA thesis submitted in partial fulfillment of the requirement for the degree of Master of Engineering
PublisherAsian Institute of Technology
AbstractNeural networks have been described as opaque systems by their critics because the knowledge learned by the network is implicitly coded in the weight matrices. These weight matrices, however, can be analyzed using computer programs and rules can be extracted that explicitly state what the network has learned. A rule extraction algorithm called NeuRuleMaker has been described and implemented. Three case studies have been taken as real-world examples and the extracted rulebase in each case is shown to perform close to the neural network itself from which the rules were extracted. It is argued that rule extraction can be a valuable tool for expert system developers and may help popularize "consumer grade" expert systems.
Year1995
TypeThesis
SchoolSchool of Engineering and Technology (SET)
DepartmentDepartment of Information and Communications Technologies (DICT)
Academic Program/FoSComputer Science (CS)
Chairperson(s)Sadananda, Ramakoti;
Examination Committee(s)Stueart, Robert D.;Yulu, Qi;
Scholarship Donor(s)The Government of Austria;
DegreeThesis (M.Eng.) - Asian Institute of Technology, 1995


Usage Metrics
View Detail0
Read PDF0
Download PDF0