Graphical models factorize a global probability distribution/energy function as the product/ sum of local functions. A major inference task, known as MAP in Markov Random Fields and MPE in Bayesian Networks, is to find a global assignment of all the variables with maximum a posteriori probability/minimum energy. A usual distinction on MAP solving methods is complete/incomplete, i.e. the ability to prove optimality or not. Most complete methods rely on tree search, while incomplete methods rely on local search. Among them, we study Variable Neighborhood Search (VNS) for graphical models. In this paper, we propose an iterative approach above VNS that uses (partial) tree search inside its local neighborhood exploration. The proposed approach performs several neighborhood explorations of increasing search complexity, by controlling two parameters, the discrepancy limit and the neighborhood size. Thus, optimality of the obtained solutions can be proven when the neighborhood size is maximal and with unbounded tree search. We further propose a parallel version of our method improving its anytime behavior on difficult instances coming from a large graphical model benchmark. Last we experiment on the challenging minimum energy problem found in Computational Protein Design, showing the practical benefit of our parallel version. A solver is available at https://github.com/toulbar2/toulbar2.

The quadratic 3-dimensional assignment problem (Q3AP) is an extension of the well-known NP-hard quadratic assignment problem. It has been proved to be one of the most difficult combinatorial optimization problems. Local search (LS) algorithms are a class of heuristics which have been successfully applied to solve such hard optimization problem. These methods handle with a single solution iteratively improved by exploring its neighborhood in the solution space. In this paper, we propose an iterated tabu search for solving the Q3AP. The design of this algorithm is essentially based on a new large neighborhood structure. Indeed, in LS heuristics, designing operators to explore large promising regions of the search space may improve the quality of the obtained solutions. However, designing such neighborhood is at the expense of a highly computationally process. Therefore, the use of graphics processing units (GPUs) provides an efficient complementary way to speed up the search. The proposed GPU-based iterated tabu search has been experimented on 5 different Q3AP instances. The obtained results are convincing both in terms of efficiency, quality and robustness of the provided solutions at run time.

In this paper we propose a parallel hybrid genetic method for solving Quadratic 3-dimensional Assignment Problem (Q3AP). This problem is proved to be computationally NP-hard. The parallelism in our algorithm is of two hierarchical levels. The first level is an insular model where a number of GAs (genetic algorithms) evolve in parallel. The second level is a parallel transformation of individuals in each GA. Implementation has been done using ParadisEO1 framework, and the experiments have been performed on GRID5000, the French nation-wide computational grid. To evaluate our method, we used three benchmarks derived from QAP instances of QAPLIB and the results are compared with those reported in the literature. The preliminary results show that the method is promising. The obtained solutions are close to the optimal values and the execution is efficient.