Symmetry-reinforced Nogood Recording from Restarts

Christophe Lecoutre (1) , Sébastien Tabary (1)

1 CRIL - Centre de Recherche en Informatique de Lens

Nogood recording from restarts is a form of lightweight learn- ing that combines nogood recording with a restart strategy. At the end of each run, nogoods are extracted from the current (rightmost) branch of the search tree. These nogoods can be used to prevent parts of the search space from being explored more than once. In this paper, we propose to reinforce nogood recording (from restarts) by exploiting symmetries: every time the solver has to be restarted, not only the nogoods that are extracted from the current branch are recorded, but also some additional nogoods that can be computed by means of the previously identi ed problem symmetries. This mechanism of computing symmetric nogoods can be iterated until a xed-point is reached, and controlled (if necessary) by limiting the number and/or the size of recorded nogoods.

Domaines: Intelligence artificielle [cs.AI]

Christophe Lecoutre, Sébastien Tabary. Symmetry-reinforced Nogood Recording from Restarts. 11th International Workshop on Symmetry in Constraint Satisfaction Problems (SymCon'11), 2011, Perugia, Italy. pp.13-27. ⟨hal-00870896⟩