Chapter

SOFSEM 2012: Theory and Practice of Computer Science

Volume 7147 of the series Lecture Notes in Computer Science pp 612-624

Knowledge Compilation with Empowerment

  • Lucas BordeauxAffiliated withCarnegie Mellon UniversityMicrosoft Research
  • , Joao Marques-SilvaAffiliated withCarnegie Mellon UniversityMicrosoft ResearchUniversity CollegeIST/INESC-ID

* Final gross prices may vary according to local VAT.

Get Access

Abstract

When we encode constraints as Boolean formulas, a natural question is whether the encoding ensures a “propagation completeness” property: is the basic unit propagation mechanism able to deduce all the literals that are logically valid? We consider the problem of automatically finding encodings with this property. Our goal is to compile a naïve definition of a constraint into a good, propagation-complete encoding. Well-known Knowledge Compilation techniques from AI can be used for this purpose, but the constraints for which they can produce a polynomial size encoding are few. We show that the notion of empowerment recently introduced in the SAT literature allows producing encodings that are shorter than with previous techniques, sometimes exponentially.