Seminar on Computational Learning and Adaptation

   Constraint-Based Learning From Error: Theory and Applications

Stellan Ohlsson
Department of Psychology
University of Illinois at Chicago
 

Cognitive systems, whether human or artificial, can learn from the errors they make when negotiating an unfamiliar task environment. To benefit from such an error, the learner must detect the error, attribute it to a particular knowledge structure, and choose among the many possible revisions of that structure. In this talk, I present a computational algorithm which addresses these problems in a general way that relies in reinterpreting declarative knowledge as consisting of constraints rather than propositions. The resulting theory throws new light on certain patterns in human learning, and it has generated a design theory for intelligent tutoring systems. Constraint-based tutoring systems implemented by Antonija Mitrovic at Canterbury University demonstrably support human learning, and some have become viable, Web-accessible products in the commercial market, thus completing the process of research and development.

Return to the seminar schedule