global optimization; multimodal function; parallelization; opt-aiNet algorithm; graphics processing unit
Parallelization is one of possible approaches for obtaining better results in terms of algorithm performance and overcome the limits of the sequential computation. In this paper, we present a study of parallelization of the opt-aiNet algorithm which comes from Artificial Immune Systems, one part of large family of population based algorithms inspired by nature. The opt-aiNet algorithm is based on an immune network theory which incorporates knowledge about mammalian immune systems in order to create a state-of-the-art algorithm suitable for the multimodal function optimization. The algorithm is known for a combination of local and global search with an emphasis on maintaining a stable set of distinct local extrema solutions. Moreover, its modifications can be used for many other purposes like data clustering or combinatorial optimization. The parallel version of the algorithm is designed especially for modern graphics processing units. The preliminary performance results show very significant speedup over the computation with traditional central processor units.