Analysis of Scalable Parallel Evolutionary Algorithms [CEC7501]

Inherent parallelism is regarded as one of the most important advantages of evolutionary algorithms. This paper aims at makin g an initial study on the speedup of scalable parallel evolutionary algorithms. First the scalable parallel evolutionary algo rithms are described; then the speedup of such scalable algorithms is defined based on the first hitting time; Using the new definition, the relationship between population diversity and superlinear speedup is analyzed; finally a case study demonstra tes how population diversity plays a crucial role in generating the superlinear speedup.