Resolving Efficiency Bottleneck of the Bellman Equation in Adaptive Teaching

Abstract

AbstractAn intelligent tutoring system (ITS) performs adaptive teaching, which is essentially a process of making decisions. In each teaching step, the system chooses an optimal action based on information of the student’s current knowledge state. The partially observable Markov decision process (POMDP) model is a useful tool for building an ITS. The model enables the system to choose optimal teaching actions even when information of student knowledge states is incomplete and/or unclear. In a POMDP, the agent makes decisions by evaluating a Bellman equation, which has exponential complexity. The computing cost for evaluating the Bellman equation has been a major bottleneck in building POMDP-based ITSs. In this paper, we report our techniques for improving computing efficiency in evaluating the Bellman equation in an ITS. The techniques include partitioning and reducing the state space and solution space.KeywordsIntelligent tutoring systemComputer supported educationAdaptive teachingPartially observable Markov decision processComputational complexity