A framework for data-driven adaptive instruction

Speaker
Joseph E. Beck
Robotics Institute, Carnegie Mellon University

When
-

Where
Wean Hall 5409

Description

This talk describes ADVISOR, an architecture for selecting tutorial actions in a data-driven, principled manner. The first component of ADVISOR observes students using an intelligent tutor, and from their interactions with the tutor induces a model of how students behave. The second component of ADVISOR is responsible for selecting tutorial decisions that are optimal, according to a specified criterion, for this student. This approach is in contrast with theory-driven approaches, which typically cognitive or pedagogical theories to determine how the machine should behave. This talk discusses different methods for making decisions in the ADVISOR framework, including reinforcement learning, heuristic search, and Monte Carlo simulation. Finally, this talk discusses new capabilities that ADVISOR provides to system designers, including the ability to determine where research and development effort is most profitably spent.

Speaker's Bio

Joseph E. Beck is a postdoctoral fellow at CMU in the robotics institute working on Project LISTEN. He received his B.S. in math, computer science, and cognitive science from CMU, and recently received his Ph.D. in computer science at the University of Massachusetts Amherst, specializing in applying machine learning techniques to user modeling and intelligent tutoring construction. His areas of interest include reasoning about users, empirical evaluations, machine learning, and intelligent tutoring systems.

Speaker's Website
http://www.cs.cmu.edu/~krelth/

Host
Jack Mostow