Simulated annealing: theory and applications
Simulated annealing: theory and applications
Learning internal representations by error propagation
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1
The Utility of Knowledge in Inductive Learning
Machine Learning
Performance evaluation of MAX: the Maintenance Administrator Expert System
CIKM '93 Proceedings of the second international conference on Information and knowledge management
Theory refinement combining analytical and empirical methods
Artificial Intelligence
Genetic Algorithms in Search, Optimization and Machine Learning
Genetic Algorithms in Search, Optimization and Machine Learning
Machine Learning
NYNEX MAX: A Telephone Trouble Screening Expert
IAAI '91 Proceedings of the The Third Conference on Innovative Applications of Artificial Intelligence
Hi-index | 0.00 |
Our research addresses the challenge of learning to troubleshoot a telephone network using data provided by Nynex (the parent company of Nynex New England and Nynex New York, formerly New England Telephone and New York Telephone). Nynex has implemented a rule-based expert system, the Maintenance Administrator Expert (Max), which determines a malfunction's location for customer-reported telephone troubles. In particular, Max troubleshoots the local loop, the part of the telephone network from the central office to the customer's premises. Max, as well as other systems generated from the original system, are an important part of the operations of Nynex New York and Nynex New England, the largest phone companies for those regions.Like all expert systems, Max requires occasional maintenance to its knowledge base. In addition, many different sites in New York and New England use Max, and there are small differences in how each site diagnoses reported troubles. Max's designers have facilitated this customization via numeric parameters (indicating, for example, when a voltage is too high) that each site sets or that the designers adjust periodically to improve Max's performance. Our goal is to develop strategies to tune these parameters for improved performance on the examples. Because of the many troubles processed by Max and Max-related systems, a phone company can save a significant amount of money with even a small improvement in performance (as little as 1%).In this article, we describe the troubleshooting problem and Max in more detail and discuss how we collected the data for the study. Then we describe Opti-Max, our approach for revising Max's parameters, and describe the results of using Opti-Max to improve these parameters.