Correction Queries in Active Learning
Scientific Applications of Language Methods
Imperial College Press
In this work we investigate the learning model in which a learner gets information of the hidden concept by using different types of correction query (CQs). Specifically, three types of queries are taken into account, the so-called prefix, length bounded and edit distance CQs.
In order to state the power of these models, we have considered two learning scenarios. In the first one, computational complexity issues are neglected and the matter is to decide, for a given class of concepts, whether learning in a finite number of steps is possible. Here we show power relationships among CQ models themselves, between these models and well-established query models, and with learning models in a different paradigm, namely the Gold-style learning in the limit model.
The last part of this chapter is focused on the second learning scenario where computational complexity issues are important: here the point is to decide for a given class whether there exists poly time learners. We provide new poly time learning algorithms using prefix correction queries for the class of pattern languages and the class of k-reversible languages. In addition, we show power relationships among the new models themselves and the membership query model.