Pedagogical Activity Selection:
Drawing Insight From Sequential
Decision Making Under Uncertainty
Emma Brunskill
Computer Science Department
Carnegie Mellon University
What activity should be given to a student at each interaction to best
help the student learn? This question represents one of the key
challenges facing any human or automated teacher, and good answers to
it could have a profound impact on education. Sequential decision
making under uncertainty offers us a powerful and flexible frameworks
for modeling adaptive pedagogical activity selection. I will discuss
two of my projects that pose tutor activity selection as sequential
decision making under uncertainty: an algorithm for scaling up to
the large domains common in education, and an analysis of the impact of
individual variance in decision process parameters on the resulting
best policies. I'll also briefly mention open issues.