Adaptive learning using a qualitative feedback loop

Most of the example-based learning algorithms developed so far are limited by the fact that they learn unidirectionally, i.e., they just transform the presented examples into a fixed internal representation form and do not adapt their learning strategy according to the results of this transformation process. Only a few learning algorithms incorporate such a feedback from an evaluation of the learned problem representation to the input for the next learning step. But all those rely on quantitative evaluation of the problem representation only, qualitative criteria are always neglected. In this paper we present the automatic learning environment ALEX which allows for adaptive learning by applying a feedback loop based on quantitative and qualitative evaluation of the problem representation. We follow the idea that the quality of a problem representation determines further knowledge acquisition activities in a certain problem domain. Hence, we derive qualitative evaluation criteria for problem representations and exemplify their successful applicability for an inductive learning strategy, namely example-based learning, in ALEX