multistrategy task-adaptive learning: MTL

(Michalski, Wnek, Kaufman, Utz, Vafaie, J. Zhang)

This project is concerned with developing a novel methodology for multistrategy learning, based on the Inferential Theory of Learning. The proposed methodology, called multistrategy task-adaptive learning (MTL) integrates a range of learning strategies, in particular, two basic and mutually complementary learning paradigms: empirical learning and analytical learning (see Figure beside). Empirical learning assumes that the learner does not have much prior knowledge relevant to the task of learning, while analytic learning assumes that the learner has sufficient knowledge to solve the problem in principle, but that knowledge is not directly applicable or efficient. Empirical learning is based primarily on inductive inference from facts, while analytical learning is based primarily on deductive inference from prior knowledge.

Other major learning strategies that are integrated in MTL include constructive induction, analogical learning, and abstraction. Constructive induction employs background knowledge to generate problem-relevant descriptive concepts, and through them derives the most plausible inductive hypotheses. Analogical learning transfers knowledge from one problem domain to another through an analysis of similarities between concepts or problem solving methods. Abstraction transfers a description from a high-detail level to a low-detail and more goal-oriented level.

MTL postulates that the learning strategy, or a combination thereof, should be based on the analysis of the learning task at hand. A learning task is defined by the input, learner's prior knowledge and the learner's goal(s). The learning goal(s) are viewed as a central factor in controlling a learning process. This research provides foundations for building advanced learning systems, and applying them to such tasks as knowledge acquisition, planning, problem solving, intelligent robots and knowledge extraction from databases.

Selected References

Michalski, R.S., "Toward a Unified Theory of Learning: Multistrategy Task-adaptive Learning," in Readings in Knowledge Acquisition and Learning: Automating the Construction and Improvement of Expert Systems, B.G. Buchanan and D.C. Wilkins, Morgan Kaufmann, San Mateo, 1993.

For more references, seeĀ publications section.


MLI Copyright © 2014 Machine Learning and Inference Laboratory
College of Health and Human Services, George Mason University
4400 University Dr, MSN 1J3, Fairfax, VA 22030, U.S.A