The Application of Pareto Local Search to the Single-Objective Quadratic Assignment Problem

Pareto optimization, that includes the simultaneous optimization of multiple conflict objectives, has been employed as a high level strategy to reduce the effect of local optima (Segura et al. 2013). This approach was first introduced in (Louis and Rawlins 1993), and later reinvestigated and termed as multi-objectivization in (Knowles, Watson, and Corne 2001). Since then it has been studied by many researchers actively. The idea of multi-objectivization is to translate the target single-objective optimization problem into a multi-objective one, and then to solve the later using a Preto optimization technique. There have been several studies on multi-objectivization with various applications (Segura et al. 2013), resulting in two main groups of multi-objectivization: methods that decompose the primary objective into multiple conflicting objectives (Knowles, Watson, and Corne 2001), and methods that optimizes at least one additional “helper” objective simultaneously with the primary objective (Jensen 2004). Both approaches rely on devising new effective problemdependent objectives, which is normally a tedious task. This paper briefly presents the application of Pareto local search (PLS) (Paquete, Chiarandini, and Stützle 2004), as a Pareto optimization technique, to the single-objective quadratic assignment problem. The idea is to use PLS instead of local search, to optimize the primary objective together with an additional augmented function. The augmented objective function is defined using a general penaltybased approach, an idea that comes from Guided Local Search (GLS) (Voudouris, Tsang, and Alsheddy 2010). This results in a multi-objectivization approach that is simple and general.