Distributed random walks for fitness landscape analysis

Fitness landscape analysis is used to mathematically characterize optimization problems. In order to perform fitness landscape analysis on continuous-valued optimization problems, a sample of the fitness landscape needs to be taken. A common way to perform this sampling is to use random walk algorithms. This paper proposes a new random walk algorithm for continuous-valued optimization problems, called the distributed random walk algorithm. The algorithm is based on the premise that multiple short random walks of the same type will provide better coverage of the decision space and more robust fitness landscape measures than a single long random walk. The distributed random walk algorithm is simple to implement, and the computational overhead is insignificant compared to random walk algorithms in the literature. The results of the study indicate that the distributed random walk algorithm achieves both of these objectives. Furthermore, the benefits of the distributed random walk algorithm are shown to be much more significant when small step sizes are used in the random walks.