Modeling the Student with Reinforcement Learning

We describe a methodology for enabling an intelligent teaching system to make high level strategy decisions on the basis of low level student modeling information. This framework is less costly to construct, and superior to hand coding teaching strategies as it is more responsive to the learner’s needs. In order to accomplish this, reinforcement learning is used to learn to associate superior teaching actions with certain states of the student’s knowledge. Reinforcement learning (RL) has been shown to be flexible in handling noisy data, and does not need expert domain knowledge. A drawback of RL is that it often needs a significant number of trials for learning. We propose an off-line learning methodology using sample data, simulated students, and small amounts of expert knowledge to bypass this problem.