Efficiency enhancement of estimation of distribution algorithms by a compressed tournament selection

Abstract

Estimation of Distribution Algorithms (EDA) are population-based meta-heuristics that use probabilistic models to describe the relationships between variables of a problem. From these models, new solutions can be sampled without disruption of Building Blocks, leading to a more effective search. The key aspect of EDAs is the model-building step and better models should enable a better exploration of the search space. However, high-quality models require computationally expensive algorithms, resulting in too costly EDAs. This paper proposes an efficient implementation of the tournament selection algorithm (TS), called Compressed Tournament Selection (CTS). CTS avoids the insertion of repeated solutions in the population during the TS. Therefore, model-building can be performed on a reduced population (equivalent to the original), improving the EDAs' performance as a whole. In the experiments the method was applied to the Extended Compact Genetic Algorithm and the results showed that by using an appropriate tournament size, CTS can lead to high speed-ups without a decrease of model quality. Besides, CTS keeps the characteristics of TS and can be easily used to enhance the efficiency of any EDA based on the tournament selection algorithm.