Hybridization as Cooperative Parallelism for the Quadratic Assignment Problem

Abstract

International audienceThe Quadratic Assignment Problem is at the core of several real-life applications. Finding an optimal assignment is computationally very difficult, for many useful instances. The best results are obtained with hybrid heuristics, which result in complex solvers. We propose an alternate solution where hybridization is obtain by means of parallelism and cooperation between simple single-heuristic solvers. We present experimental evidence that this approach is very efficient and can effectively solve a wide variety of hard problems, often surpassing state-of-the-art systems

On Integrating Population-Based Metaheuristics with Cooperative Parallelism

“…ParEOTS [19] is a powerful hybrid solver for QAP combining Extremal Optimization and Tabu Search. Being interested in memetic algorithms, we tried to add a Genetic Algorithm (GA) as third component to this hybrid solver.…”

“…These methods are detailed in [18], [19]. We extend the CPLS model by having two independent pools of configurations inside the head node: the diversification pool (DivPool) and the intensification pool (IntPool).…”

“…All experiments have been carried out on server with 4 × 16-core AMD Opteron 6376 CPUs, running at 2.3 GHz and 128 GB of RAM. This is the same machine as used for ParEOTS evaluation in [19] but we could only use 32 cores (while the reference paper presents results for 128 cores).…”

“…The expected outcome is that a solution which may be stagnating for one solver, has a chance to be improved on by another metaheuristic. CPLS has been successfully used to develop a very efficient hybrid solver for QAP, called ParEOTS, by implicitly combining Extremal Optimization and Tabu Search [19].…”

“…Theoretically it is possible to plug a population-based metaheuristics into CPLS, we thus tried to hybridize a genetic algorithm with the local search procedures used in ParEOTS (extremal optimization and tabu search) [19]. It turned out that the resulting solver performed worse than ParEOTS!…”