Learning Logic Formulas from Data

Speaker:  Klaus Truemper - University of Texas at Dallas
  Tuesday, June 15, 2004 at 5:00 PM
Among the important tools for the formulation of logic-based intelligent systems are methods that extract logic formulas from data. For example, for systems handling credit rating, medical diagnosis, or natural language processing, such extraction methods are very useful. In this talk we present one such method, called Lsquare, for the extraction task. Lsquare supports the derivation of shortest, longest, and optimized formulas. The method requires that the data contain only True/False values. Often, that condition is not satisfied. For example, some values may be rational numbers or elements of finite, but possibly large, sets. We describe a second method, called Cutpoint, that transforms such values to True/False values to which Lsquare can be applied. Finally, we cover some computational results that prove the two methods to be effective and useful.

Ca' Vignal 3 - Piramide, Floor 0, Hall Verde

Contact person
Maria Paola Bonacina

Publication date
April 22, 2004