An Experimental Investigation of Iterated Ants for the Quadratic Assignment Problem

Ant Colony Optimization (ACO) algorithms construct solutions each time starting from scratch, that is, from an empty solution. Similar to ACO, Iterated Greedy is a constructive stochastic local search (SLS) method. However, differently from ACO, Iterated Greedy starts the solution construction from partial solutions. In this paper we examine the performance of a variation of MAX–MIN Ant System, one of the most successful ACO algorithms, that exploits this idea of starting the solution construction from partial solutions when applied to the quadratic assignment problem. This particular application problem is chosen, since another source of inspiration for this article are earlier researches on the usage of external memory in ACO, an idea that actually also results in the usage of partial solutions to seed the solution construction in ACO algorithms. Differently from previously reported results on external memory usage, our computational results are more pessimistic in the sense that the idea of using partial solutions for seeding the constructions does not necessarily lead to improvements, when this idea is integrated into very high-performing ACO algorithms.