Experimental evidence and theoretical advances over the years have created an academic consensus regarding majority voting systems, namely that, under certain conditions, the group performs better than its components. However, the underlying reason for such conditions, e.g., stochastic independence of agents, is not often explored and may help to improve performance in known setups by changing agent behavior, or find new ways of combining agents to take better advantage of their characteristics. In this work, an investigation is conducted for homogeneous groups of independent agents playing the game of Checkers. The analysis aims to find the relationship between the change in performance caused by majority voting, the group size, and the underlying decision process of each agent, which is mapped to its source of non-determinism. A characteristic unbalance in Checkers, due to an apparent initiative disadvantage, serves as a pivot for the study, from which decisions can be separated into beneficial or detrimental biases. Experimental results indicate that performance changes caused by majority voting may be beneficial or not, and are linked to the game properties and player skill. Additionally, a way of improving agent performance by manipulating its non-determinism source is briefly explored.
CITATION STYLE
Carvalho, D. S., Nguyen, M. L., & Iida, H. (2017). An analysis of majority voting in homogeneous groups for checkers: understanding group performance through unbalance. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 10664 LNCS, pp. 213–223). Springer Verlag. https://doi.org/10.1007/978-3-319-71649-7_18
Mendeley helps you to discover research relevant for your work.