No abstract

No abstract

No abstract

Abstract. In this paper we present an extension of a previously developed generic student model based on Bayesian Networks. A new layer has been added to the model to include prerequisite relationships. The need of this new layer is motivated from different points of view: in practice, this kind of relationships are very common in any educational setting, but also their use allows for improving efficiency of both adaptation mechanisms and the inference process. The new prerequisite layer has been evaluated using two different experiments: the first experiment uses a small toy example to show how the BN can emulate human reasoning in this context, while the second experiment with simulated students suggests that prerequisite relationships can improve the efficiency of the diagnosis process by allowing increased accuracy or reductions in the test length.

This work describes MPQ-learning, an temporal-difference method that approximates the set of all non-dominated policies in multi-objective Markov decision problems, where rewards are vectors and each component stands for an objective to maximize. Unlike other approximations to Multi-objective Reinforcement Learning, MPQ-learning does not require additional parameters or preference information, and can be applied to non-convex Pareto frontiers. We also present the results of the application of MPQ-learning to some benchmark problems and compare it to a linearization procedure.