Model-free POMDP optimisation of tutoring systems with echo-state networks
Abstract
Intelligent Tutoring Systems (ITSs) are now recognised as an interesting alternative for providing learning opportunities in various domains. The Reinforcement Learning (RL) approach has been shown reliable for finding efficient teaching strategies. However, similarly to other human-machine interaction systems such as spoken dialogue systems, ITSs suffer from a partial knowledge of the interlocutor's intentions. In the dialogue case, engineering work can infer a precise state of the user by taking into account the uncertainty provided by the spoken understanding language module. A model-free approach based on RL and Echo State Newtorks (ESNs), which retrieves similar information, is proposed here for tutoring.