Learning teaching strategies in an Adaptive and Intelligent Educational System through Reinforcement Learning

Abstract

One of the most important issues in Adaptive and Intelligent Educational Systems (AIES) is to define effective pedagogical policies for tutoring students according to their needs. This paper proposes to use Reinforcement Learning (RL) in the pedagogical module of an educational system so that the system learns automatically which is the best pedagogical policy for teaching students. One of the main characteristics of this approach is its ability to improve the pedagogical policy based only on acquired experience with other students with similar learning characteristics. In this paper we study the learning performance of the educational system through three important issues. Firstly, the learning convergence towards accurate pedagogical policies. Secondly, the role of exploration/exploitation strategies in the application of RL to AIES. Finally, a method for reducing the training phase of the AIES.

Reinforcement learning of pedagogical policies in adaptive and intelligent educational systems

“…Simi larly, initializing the value function with pre recorded experience tuples may accelerate the learning of the action policy [10]. In pre vious work [7], we have verified empirically with simulated stu dents that the size of the learning phase can be reduced by initializing the system with a pedagogical strategy, even when the initialization does not completely match with the current stu dents' needs.…”

Reinforcement Learning in a POMDP Based Intelligent Tutoring System for Optimizing Teaching Strategies

“…Chi and co-workers performed empirical evaluation on the application of RL to adaptive pedagogical strategies [8]. In [9], the researchers evaluated the learning performance of the educational system through three issues: The learning convergence, exploration/exploitation strategies, and reduction of training phase.…”

“…∈ ′∈ (9) where is the root action of , is the observation after the agent takes , is the subtree in which is connected to the node of by the edge of , ℛ( , ) is the expected immediate reward after is taken in , and ( | , ′) is the observation probability for the agent to observe after it takes and enters ′.…”

Aprendizagem simulada na nuvem

“…Conforme Iglesias et al (2009), quando se trata de sistemas em avaliação é difícil de convencer uma quantia suficiente de estudantes a participar desse tipo de experimento. Para isso, foi necessário realizar uma adaptação na ferramenta de simulação MADEA (Modelagem Automática e Dinâmica de Estilos de Aprendizagem), a qual foi desenvolvida por Dorça (2012) em sua tese.…”