Section: New Results

Filtering

Participants : Nicolas Beldiceanu, Alban Derrien, Jérémie Du Boisberranger, Jean-Guillaume Fages, Arnaud Letort, Xavier Lorca, Thierry Petit, Charlotte Truchet, Mohamed Wahbi.

• Given a matrix model, with the same constraint defined by a finite-state automaton on each row and a global cardinality constraint on each column, [12] exploits double counting to derive necessary conditions on the cardinality variables of the global cardinality constraints from the automata. (participants: Beldiceanu)

• By using the observation that most global constraints can be reformulated as a conjunction of a total function constraint together with a constraint that can be easily reified (e.g. a linear constraint involving two variables), [13] introduces a simple way for deriving reified global constraints. (participants: Beldiceanu)

• In the context of distributed constraint solving [22] , [25] introduce two filtering algorithms that extend Asynchronous Forward Checking (AFC). The last one outperforms AFC specially on sparse problems. (participants: Wahbi)

• We improve the energetic reasoning checker of the cumulative constraint by decreasing the number checked intervals by a factor seven. We prove this approach can be generalized to the ER filtering algorithm. Furthermore, in a context of makespan minimization of hard problems, our experiments demonstrate that associating this checker with a Time-Table propagator is more efficient than using the best state-of-the-art propagators, such as Time-Table Edge-Finding. This work is at the core of Alban Derrien's doctoral research (Alban is a PhD student of TASC). It was published at the doctoral program of CP2013 [28] . (participants: Derrien, Petit)

• [29] introduces a probabilistic model for the bound consistency algorithm of the alldifferent constraint in order to decrease the number of times the constraint is woken without making new deductions during constraint propagation. (participants: Du Boisberranger, Lorca, Truchet).

• Initially motivated by the shift minimisation personal task scheduling problem [30] shows how to integrate difference constraints into the AtMostNValue constraint in order to get a better estimation about the minimum number of distinct values. (participants: Fages, Lorca)

• Motivated by scalability issues, and based on the idea of accelerating the convergence to the fix-point by filtering several cumulative constraints in parallel, [33] and [47] presents a sweep based algorithm for a conjunction of cumulative constraints. (participants: Beldiceanu, Letort)

• [46] come up with a more efficient filtering algorithm than the one introduced for the cost regular constraint for dealing with constraints for which the set of solutions can be represented by an automaton with counters. (participants: Beldiceanu)