Generation of Expert Systems by Partial Evaluation
Abstract
In this report, Partial Evaluation (PE) is shown to be a useful technique for machine learning. Our implemented version, LDS, generates expert diagnostic knowledge bases using PE. One of the benefits of the system is to simplify knowledge acquisition. PE is applicable whenever the problems to be solved share reasoning methods that can be represented by a common model with variables, and the knowledge to partially instantiate those variables is readily available. The general model used in this report is a causal model of the failures observed in power supplies, supplemented by knowledge that links structures to function. Unlike Explanation-based Learning, learning by Partial Evaluation proceeds by specializing a general model rather than generalizing a specific example. The output of learning is a specific causal model for an input circuit and is ready to be used by the performance system, which is an expert diagnostician.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1990
- Accession Number
- ADA220427
Entities
People
- David P. Benjamin
- Leonard Friedman
Organizations
- University of Southern California