On The Role Of Execution Order In Hybrid Evolutionary Algorithms

Many real-world problems can be formulated as the optimization of a continuous function. Furthermore, these problems are becoming increasingly more complex every year, reaching, or even exceeding, the thousand of variables. Evolutionary Algorithms have been traditionally successful at solving this kind of problems, due to their good balance in terms of solution quality and computation time. However, the aforementioned growth in the size of the problems requires of novel approaches to deal with the increased complexity of larger solutions spaces. Hybrid evolutionary algorithms are a powerful alternative in these scenarios as they are able to combine the strengths of multiple search methods to solve more complex problems. These hybrid approaches normally do not pay attention to the execution order of their components, being the most frequent strategy to always run them in a predefined sequence. In this contribution we study the role of execution order in hybrid evolutionary algorithms within the context of the multiple offspring sampling framework, one of the best algorithms in large-scale global optimization. As shown in the experimentation, a proper execution order policy can boost the performance of MOS to improve the results of other state-of-the-art algorithms.