Estimating Networks of Interaction in Fish Schools with a Minimal Model

This paper addresses the problem of interaction-network estimation in fish schooling to identify individuals in a group that drives the overall group behavior. We propose a simple linear approximation of the interaction of a group so that the influence among individuals can be estimated in an interpretable manner. Specifically, we build our model based on a well-known attraction term in collective behavior models where the velocity of each individual is assumed to be determined by the relative position of other individuals. The extended model with the addition of an autonomous term is shown to enable stable estimation of the interaction network and analysis of the characterization of each individual.