The goal of this work is the development of new efficient algorithms that learn classifying rules ("concepts") from examples, as well as the characterization of those types of concepts for which such efficient learning algorithms exist. Algorithms for inference os string patterns for logic programs, and for various types of boolean functions will be developed, and new learning algorithms in naturally motivated domains will be sought. The approach will also focus on extending known structural and combinatorial characterizations of learnable concept classes so as to be more widely applicable, and so as to address the learnability of concepts when the learning algorithms may pose various queries to a human expert.//
