Solver Learning for Predicting Changes in Dynamic Constraint Satisfaction Problems
Jon Spragg, Arnaud Lallouet, Andre Legtchenko, Éric Monfroy, AbdelAli Ed-Dbali, Ying Lü, Lara S. Crawford, Wheeler Ruml, Markus P. J. Fromherz, Christophe Guettier, William S. Havens, Bistra Dilkina, Nicola Policella, Amedeo Cesta, Angelo Oddi, Stephen F. Smith, Alfio Vidotto, Kenneth N. Brown, Christine Wei Wu, Andrëı Legtchenko · 2004
We present a way of integrating machine learning capabilities in constraint reasoning systems by the use of partially defined constraints called Open Constraints. This enables a form of constraint reasoning with incomplete information: we use a machine learning algorithm to guess the missing part of the constraint and we put immediately this knowledge into the operational form of a solver. This approaches extends the field of applicability of constraint reasoning to problems which are difficult to model using classical constraints, and also potentially improves the efficiency of dynamic constraint solving. We illustrate our framework on online constraint solving applications which range from mobile computing to robotics.