Peter Merz

Fitness landscape analysis and memetic algorithms for the quadratic assignment problem

In this paper, a fitness landscape analysis for several instances of the quadratic assignment problem (QAP) is performed, and the results are used to classify problem instances according to their hardness for local search heuristics and meta-heuristics based on local search. The local properties of the fitness landscape are studied by performing an autocorrelation analysis, while the global structure is investigated by employing a fitness distance correlation analysis. It is shown that epistasis, as expressed by the dominance of the flow and distance matrices of a QAP instance, the landscape ruggedness in terms of the correlation length of a landscape, and the correlation between fitness and distance of local optima in the landscape together are useful for predicting the performance of memetic algorithms-evolutionary algorithms incorporating local search (to a certain extent). Thus, based on these properties, a favorable choice of recombination and/or mutation operators can be found. Experiments comparing three different evolutionary operators for a memetic algorithm are presented.

A genetic local search algorithm for solving symmetric and asymmetric traveling salesman problems

The combination of local search heuristics and genetic algorithms is a promising approach for finding near-optimum solutions to the traveling salesman problem (TSP). An approach is presented in which local search techniques are used to find local optima in a given TSP search space, and genetic algorithms are used to search the space of local optima in order to find the global optimum. New genetic operators for realizing the proposed approach are described, and the quality and efficiency of the solutions obtained for a set of symmetric and asymmetric TSP instances are discussed. The results indicate that it is possible to arrive at high quality solutions in reasonable time.

Genetic local search for the TSP: new results

The combination of local search heuristics and genetic algorithms has been shown to be an effective approach for finding near-optimum solutions to the traveling salesman problem. Previously proposed genetic local search algorithms for the symmetric and asymmetric traveling salesman problem are revisited and potential improvements are identified. Since local search is the central component in which most of the computation time is spent, improving the efficiency of the local search operators is crucial for improving the overall performance of the algorithms. The modifications of the algorithms are described and the new results obtained are presented. The results indicate that the improved algorithms are able to arrive at better solutions in significantly less time.

New Genetic Local Search Operators for the Traveling Salesman Problem

In this paper, an approach is presented to incorporate problem specific knowledge into a genetic algorithm which is used to compute near-optimum solutions to traveling salesman problems (TSP). The approach is based on using a tour construction heuristic for generating the initial population, a tour improvement heuristic for finding local optima in a given TSP search space, and new genetic operators for effectively searching the space of local optima in order to find the global optimum. The quality and efficiency of solutions obtained for a set of TSP instances containing between 318 and 1400 cities are presented.

A comparison of memetic algorithms, tabu search, and ant colonies for the quadratic assignment problem

A memetic algorithm (MA), i.e. an evolutionary algorithm making use of local search, for the quadratic assignment problem is presented. A new recombination operator for realizing the approach is described, and the behavior of the MA is investigated on a set of problem instances containing between 25 and 100 facilities/locations. The results indicate that the proposed MA is able to produce high quality solutions quickly. A comparison of the MA with some of the currently best alternative approaches-reactive tabu search, robust tabu search and the fast ant colony system-demonstrates that the MA outperforms its competitors on all studied problem instances of practical interest.

Greedy and Local Search Heuristics for Unconstrained Binary Quadratic Programming

In this paper, a greedy heuristic and two local search algorithms, 1-opt local search and k-opt local search, are proposed for the unconstrained binary quadratic programming problem (BQP). These heuristics are well suited for the incorporation into meta-heuristics such as evolutionary algorithms. Their performance is compared for 115 problem instances. All methods are capable of producing high quality solutions in short time. In particular, the greedy heuristic is able to find near optimum solutions a few percent below the best-known solutions, and the local search procedures are sufficient to find the best-known solutions of all problem instances with n ≤ 100. The k-opt local searches even find the best-known solutions for all problems of size n ≤ 250 and for 11 out of 15 instances of size n = 500 in all runs. For larger problems (n = 500, 1000, 2500), the heuristics appear to be capable of finding near optimum solutions quickly. Therefore, the proposed heuristics—especially the k-opt local search—offer a great potential for the incorporation in more sophisticated meta-heuristics.

Fitness Landscapes, Memetic Algorithms, and Greedy Operators for Graph Bipartitioning

The fitness landscape of the graph bipartitioning problem is investigated by performing a search space analysis for several types of graphs. The analysis shows that the structure of the search space is significantly different for the types of instances studied. Moreover, with increasing epistasis, the amount of gene interactions in the representation of a solution in an evolutionary algorithm, the number of local minima for one type of instance decreases and, thus, the search becomes easier. We suggest that other characteristics besides high epistasis might have greater influence on the hardness of a problem. To understand these characteristics, the notion of a dependency graph describing gene interactions is introduced. In particular, the local structure and the regularity of the dependency graph seems to be important for the performance of an algorithm, and in fact, algorithms that exploit these properties perform significantly better than others which do not. It will be shown that a simple hybrid multi-start local search exploiting locality in the structure of the graphs is able to find optimum or near optimum solutions very quickly. However, if the problem size increases or the graphs become unstructured, a memetic algorithm (a genetic algorithm incorporating local search) is shown to be much more effective.

Memetic Algorithms and the Fitness Landscape of the Graph Bi-Partitioning Problem

In this paper, two types of fitness landscapes of the graph bipartitioning problem are analyzed, and a memetic algorithm — a genetic algorithm incorporating local search — that finds near-optimum solutions efficiently is presented. A search space analysis reveals that the fitness landscapes of geometric and non-geometric random graphs differ significantly, and within each type of graph there are also differences with respect to the epistasis of the problem instances. As suggested by the analysis, the performance of the proposed memetic algorithm based on Kernighan-Lin local search is better on problem instances with high epistasis than with low epistasis. Further analytical results indicate that a combination of a recently proposed greedy heuristic and Kernighan-Lin local search is likely to perform well on geometric graphs. The experimental results obtained for non-geometric graphs show that the proposed memetic algorithm (MA) is superior to any other heuristic known to us. For the geometric graphs considered, only the initialization phase of the MA is required to find (near) optimum solutions.

On the effectiveness of evolutionary search in high-dimensional NK-landscapes

NK-landscapes offer the ability to assess the performance of evolutionary algorithms on problems with different degrees of epistasis. In this paper, we study the performance of six algorithms in NK-landscapes with low and high dimension while keeping the amount of epistatic interactions constant. The results show that compared to genetic local search algorithms, the performance of standard genetic algorithms employing crossover or mutation significantly decreases with increasing problem size. Furthermore, with increasing K, crossover based algorithms are in both cases outperformed by mutation based algorithms. However, the relative performance differences between the algorithms grow significantly with the dimension of the search space, indicating that it is important to consider high-dimensional landscapes for evaluating the performance of evolutionary algorithms.