Finding New Rules for Incomplete Theories: Explicit Biases for Induction with Contextual Information

Abstract

This chapter reviews two disparate machine learning approaches that have seen rigorous review: (1) explanation-based learning (EBL) and (2) similarity-based learning (SBL). EBL is a deductive approach in which a definition of a concept is learned usually after observing only a single example of that concept. When EBL cannot derive a complete explanation, the partial explanation forms a context in which learning takes place. EBL assumes that a system is given an explicit theory of the domain that is complete, correct, and tractable. SBL is an empirical technique that involves the comparison of a large number of input examples. SBL suffers because of its lack of an explicit domain theory. Information extracted from partial explanations and from complete explanations, can be exploited by SBL to do better induction of the missing domain knowledge. The extracted information constitutes a strong bias for SBL. EBL and SBL have been applied to problems in a variety of domains. Both methods are rife with well-defined problems.