Strategies in Combined Learning via Logic Programs

• Authors: Evelina Lamma;Fabrizio Riguzzi;Luís Moniz Pereira
• Venue: Machine Learning - Special issue on multistrategy learning
• Year: 2000

We discuss the adoption of a three-valued setting forinductive concept learning. Distinguishing between what is true, whatis false and what is unknown can be useful in situations wheredecisions have to be taken on the basis of scarce, ambiguous, ordownright contradictory information. In a three-valued setting, welearn a definition for both the target concept and its opposite,considering positive and negative examples as instances of twodisjoint classes. To this purpose, we adopt Extended Logic Programs(ELP) under a Well-Founded Semantics with explicit negation(WFSX) as the representation formalism for learning, and show howELPs can be used to specify combinations of strategies in adeclarative way also coping with contradiction and exceptions.Explicit negation is used to represent the opposite concept,while default negation is used to ensure consistency and tohandle exceptions to general rules. Exceptions are representedby examples covered by the definition for a concept that belongto the training set for the opposite concept.Standard Inductive Logic Programming techniques are employed tolearn the concept and its opposite. Depending on the adoptedtechnique, we can learn the most general or the least generaldefinition. Thus, four epistemological varieties occur,resulting from the combination of most general and least generalsolutions for the positive and negative concept. We discuss thefactors that should be taken into account when choosing andstrategically combining the generality levels for positive andnegative concepts.In the paper, we also handle the issue of strategic combinationof possibly contradictory learnt definitions of a predicate andits explicit negation.All in all, we show that extended logic programs underwell-founded semantics with explicit negation add expressivityto learning tasks, and allow the tackling of a number ofrepresentation and strategic issues in a principled way.Our techniques have been implemented and examples run on astate-of-the-art logic programming system with tabling whichimplements WFSX.