On collective bandit behaviour

Abstract: The collective decision process of Gambusia affinis (the mosquitofish) is investigated from the standpoint of online machine learning algorithms. A new algorithm, the Collaborative Exp3 algorithm, is derived from the adversarial bandits framework to model how groups of fish make collective decisions leading to consensus. Thanks to maximum likelihood estimation, parameters are tuned and comparisons between data and algorithm performances are addressed. This work provides promising results in the scope of recovering information transfer within fish groups as well as to understand the individual mechanisms involved in the collective decision process. It is the first published approach to connect online machine learning algorithms with data, hence bridging a gap between theory and biological practice.