Date: Thu, 07 Nov 1996 19:08:10 GMT Server: NCSA/1.5 Content-type: text/html Last-modified: Thu, 19 Sep 1996 20:01:11 GMT Content-length: 9083 Jude W. Shavlik's Home Page

Jude W. Shavlik

Associate Professor
Computer Sciences Department
University of Wisconsin
1210 W. Dayton St.
Madison, WI 53706-1685

E-mail: shavlik@cs.wisc.edu
Telephone: (608) 262-7784
Fax: (608) 262-9777

Ph.D., University of Illinois, Urbana, 1988
Interests: machine learning, neural networks, artificial intelligence, informational retrieval, computational biology

Table of Contents

Research Summary

We are primarily developing machine learning systems that combine the strengths of symbolic approaches to artificial intelligence with those of connectionist AI. A major focus is improving the dialog between human teachers and machine learners. Traditionally, this interaction is limited to the teacher providing labelled training examples to the machine. Toward the goal of widening the ``communication pipeline'' between human and machine, we have been developing a language for providing, in a natural manner and at any time, general-purpose advice to a machine learner. In our approach, the human advice-giver observes the behavior of the learner and occasionally makes suggestions, expressed in a simple language. Based on techniques developed in our work on knowledge-based neural networks, these instructions are inserted directly into learner. Subsequent connectionist (neural network) learning further integrates and refines the advice.

Currently, we are extending the language used to advise our learning algorithms, studying new ways of incorporating this advice into neural networks, investigating the extraction of human-comprehensible rules from trained neural networks, and developing methods for choosing good representations for training examples. We are also developing parallel algorithms, on the department's Condor system and our CM-5 computer, for machine learning and computational biology.

Shavlik (1992) and (1996) provide an overview of our approach to knowledge-based neural networks. Recent developments appear in the papers referenced on this page, as well as in the "home pages" of the students listed below.

PhD Students

Selected Recent Publications

Click here to see our recent titles and abstracts (you can also grab all our abstracts in one file or directly access our ftp directory of postscript versions of recent papers.

Courses Recently Taught

Some Interesting Links


Last modified: Fri Jul 5 18:40:39 1996 by Jude Shavlik
shavlik@cs.wisc.edu