Inferring leadership in zebrafish pairs: An information-theoretic approach

The transfer of information between individuals is central to decision-making in animal groups on the move, especially when certain influential leaders in the group anticipate the movement of the rest. Identifying leaders using classical signal processing methods such as cross-correlation depends on a linear relationship between the actions of the leader and follower, which may not always be the case in real datasets. In this tutorial we will introduce concepts from information theory that are amenable to nonlinear dependence between actions of interacting individuals, and can therefore be implemented free of behavior models. To validate these concepts, we will focus on the zebrafish model organism, whose motion can be faithfully reconstructed by adopting stochastic modeling techniques from financial engineering. We will work with pairs of zebrafish, and discuss challenges in extending our approach to larger groups. Finally, we will show how these information-theoretic tools can be used to design and control robots in animal behavior studies.