Self-organized Invasive Parallel Optimization with Self-repairing Mechanism
ISSN der Zeitschrift
PARS: Parallel-Algorithmen, -Rechnerstrukturen und -Systemsoftware: Vol. 28, No. 1
Gesellschaft für Informatik e.V.
The parallelization of optimization algorithms is very beneficial when the function evaluations of optimization problems are time consuming. However, parallelization gets very complicated when we deal with a large number of parallel resources. In this paper, we present a framework called Self-organized Invasive Parallel Optimization (SIPO) in which the resources are self-organized. The optimization starts with a small number of resources which decide the number of further required resources on-demand. This means that more resources are stepwise added or eventually released from the platform. In this paper, we study an undesired effect in such a self-organized system and propose a self-repairing mechanism called Recovering-SIPO. These frameworks are tested on a series of multi-objective optimization problems.