Ubiquitous environments may become a reality in a foreseeable future and research is aimed on making them more and more adapted and comfortable for users. Our work consists on applying reinforcement learning techniques in order to adapt services provided by a ubiquitous assistant to the user. The learning produces a context model, associating actions to perceived situations of the user. Associations are based on feedback given by the user as a reaction to the behavior of the assistant. Our method brings a solution to some of the problems encountered when applying reinforcement learning to systems where the user is in the loop. For instance, the behavior of the system is completely incoherent at the beginning and needs time to converge. The user does not accept to wait that long to train the system. The user's habits may change over time and the assistant needs to integrate these changes quickly. We study methods to accelerate the reinforced learning process.