Lenny Pitt

Ph.D. Yale University, 1985

Research Statement

Professor Pitt's research focuses on formal mathematical models of how learning agents interact with their environments, precise definitions for what it means to learn, and characterizations of exactly which types of concepts or behaviors are provably learnable, given the model of interaction and the definition of learning. A wide variety of techniques are employed, including those borrowed from theory of computation, analysis of algorithms, computational complexity theory, logic, machine learning, and probability and statistics.

Projects have involved investigations into the learnability of various types of logical rules, including propositional Boolean formulas in certain normal forms, description logics (first-order KL-one type languages), and geometric classifiers in Euclidian space. In some cases, provably efficient learning algorithms have been developed, while in others it has been proved that the existence of such algorithms would result in complexity-theoretic consequences widely believed to be unlikely. A recent interdisciplinary project with faculty from Computer Science and Cognitive Science addresses the impact of background knowledge and concept use in feature extraction and concept learning. Another current project focuses on the computational complexity of data mining problems.

Professor Pitt is also involved in educational outreach activities for K-12 classrooms, with emphasis on the development of fun "hands-on" activities for teaching discrete mathematics and computer science at the elementary and secondary level.