Ask the Scientist:
What is Optimal Teaching, and what did the recent Optimal Teaching Workshop at UC San Diego entail?
The word "optimal" in "optimal teaching" is intended to underline how stochastic optimal control theory, as well as the related fields of machine learning and reinforcement learning, can contribute to the study of how humans teach as well as to the development of automated teaching systems.
The past decade has seen tremendous growth in the development of inexpensive sensors such as high-resolution cameras as well as computer vision and machine learning algorithms for mapping the sensor values into meaningful information about a student's state, such as whether he/she is frustrated, bored, engaged, etc.
In order to use these sensors to teach more effectively, however, it is useful to employ a principled mathematical framework such as control theory to integrate sensor inputs into the decision making process. For instance, if a student's performance drops suddenly, is it because the material was too difficult, or because the student stopped trying? If the student's face indicated that he had "disengaged" from the task, how should that sensor value affect the teacher's next action? Control theory provides a framework for using sensor inputs to both update the teacher's belief about how the student is doing, as well as to take actions to maximize the student's expected learning gains.
The purpose of the Optimal Control Workshop, which the TDLC hosted on May 4 of this year, was to bring together senior researchers from the intelligent tutoring systems, neuroscience, cognitive science, machine learning, and psychology communities to discuss the state-of-the-art and current challenges to automated teaching, including but not limited to the application of stochastic optimal control theory to teaching. For more information about the Optimal Teaching Workshop, please click here.