David C. Wilkins
Machine Learning for Expert Systems: Automated Knowledge Acquisition
Machine learning has great significance for the future use of computers. Autonomous computer systems of the future will need far more knowledge than humans can explicitly transfer; this requires that computers learn independently. The goal of Professor Wilkins' research is to develop techniques that will allow expert systems to learn the large amounts of domain-specific knowledge that are necessary for achieving expert-level performance. This problem of automated knowledge acquisition is widely considered to be the most important research problem associated with expert systems. The most successful means used by humans to acquire expertise is an apprenticeship learning period, the paradigmatic example being medical internship. His research produces methods that allow expert systems to expand their expertise automatically using apprenticeship learning techniques.
Professor Wilkins plans to complete the revisions to two major books in the area of automated knowledge acquisition that have been accepted for publication. The first book, Automated Knowledge Acquisition: The Odysseus Apprenticeship Learning System (MIT Press), describes the first successful demonstration of techniques that allow classification expert systems to improve their performance automatically using human apprenticeship methods. This was the focus of his Ph.D. dissertation research, which was carried out at the Knowledge Systems Lab at Stanford University, where the field of expert systems was pioneered. During his first two years at Illinois, Professor Wilkins and his research group extended this initial work. One major extension was the creation of the MINERVA expert system shell, which allows an expert system to introspect on its reasoning processes, which is essential for apprenticeship learning.
The second book, which Professor Wilkins is co-editing with Bruce Buchanan, is titled Readings in Knowledge Acquisition and Learning: Toward Automation of Program Construction and Debugging (Morgan Kaufmann). It includes commentary from both editors that provides the first unified account of the large field of automated knowledge acquisition and should have a significant impact on the direction of future research in the field.