Sequential Time Splitting and Bounds Communication for a Portfolio of Optimization Solvers

Roberto Amadini (1, 2) , Peter J. Stuckey (3, 4)

1 DISI - Department of Computer Science and Engineering [Bologna]
2 FOCUS - Foundations of Component-based Ubiquitous Systems
3 NICTA - National ICT Australia [Sydney]
4 University of Melbourne

Scheduling a subset of solvers belonging to a given portfolio has proven to be a good strategy when solving Constraint Satisfaction Problems (CSPs). In this paper, we show that this approach can also be effective for Constraint Optimization Problems (COPs). Unlike CSPs, sequential execution of optimization solvers can communicate informa-tion in the form of bounds to improve the performance of the following solvers. We provide a hybrid and flexible portfolio approach that com-bines static and dynamic time splitting for solving a given COP. Empiri-cal evaluations show the approach is promising and sometimes even able to outperform the best solver of the porfolio.

Domaines: Informatique [cs]

Roberto Amadini, Peter J. Stuckey. Sequential Time Splitting and Bounds Communication for a Portfolio of Optimization Solvers. Principles and Practice of Constraint Programming (CP 2014), Sep 2014, Lyon, France. pp.108 - 124, ⟨10.1007/978-3-319-10428-7_11⟩. ⟨hal-01091664⟩