Benchmarking a pool-based execution with GA and PSO workers on the BBOB noiseless testbed
Mario García-Valdéz, Juan Julián Merelo · Proceedings of the Genetic and Evolutionary Computation Conference Companion · 2017
In this work, we evaluate an asynchronous population-based algorithm following a pool-based approach. In Pool-based algorithms, a collection of workers collaborates through a shared population repository. In particular, we followed the EvoSpace approach in which workers asynchronously interact with a population pool by taking samples of the population to perform a standard search on the samples, to then return newly evolved solutions back to the pool. For this purpose, we use the BBOB Noiseless Testbed and a hybrid algorithm combining two kinds or workers: PSO and GA. We find that a Pool- based approach outperforms the canonical GA and PSO algorithms in nearly all cases. The results of these tests suggest that a Pool Based approach can be used to implement hybrid algorithms that can improve the performance of canonical population-based optimization algorithms.