International audienceTouch sensitive interface enable new interaction methods like using gesture commands. To easily memorize more than a dozen of gesture commands, it is important to be able to customize them. The classifier used to recognize drawn symbols must hence be customizable, able to learn from very few data samples, and evolving, able to learn and improve during its use. This work studies several methods of labeling run-time data for the classifier on-line training. We compare seven supervision strategies, depending on user interactions, and system self-evaluation capacities (notion of reject). We show in this paper that the strategy giving the best results is to learn from data implicitly validated by the user, and from data exp...