A Space-Efficient Technique of Policy Trees for an Intelligent Tutoring System on POMDP

Abstract

Adaptive tutoring is essentially a decision process, in each step of which an optimal tutoring action is chosen and taken. In recent years, researchers have been increasingly interested in applying the partially observable Markov decision process (POMDP) model to build intelligent tutoring systems (ITSs). The POMDP model may enable an ITS to optimize tutoring when uncertainties exist. Computing in a POMDP is characterized by exponential costs, in both time and space, which obstruct the POMDP model for practical applications. In our research, we develop techniques to improve computing efficiency in a POMDP based ITS. In this paper, we report a space-efficient technique of policy trees. The technique may help achieve high space efficiency by grouping policy trees and dynamically creating the policy trees to be evaluated in making a decision.