Modes of Information Flow in Collective Cohesion

TL;DR

This paper investigates how different modes of information flow influence collective cohesion, revealing limitations of traditional measures and proposing a decomposition approach to better understand individual and group interactions.

Contribution

It introduces a decomposition of transfer entropy and mutual information into intrinsic, shared, and synergistic modes to accurately characterize information flow in collective behavior.

Findings

Traditional measures can misrepresent influence strength.

Decomposition reveals detailed interaction modes.

Single pair analysis informs many-body interaction understanding.

Abstract

Pairwise interactions between individuals are taken as fundamental drivers of collective behavior responsible for group cohesion and decision-making. While an individual directly influences only a few neighbors, over time indirect influences penetrate a much larger group. The abiding question is how this spread of influence comes to affect the collective. One or a few individuals are often identified as leaders, being more influential than others. Transfer entropy and time-delayed mutual information are used to identify underlying asymmetric interactions, such as leader-follower classification in aggregated individuals--cells, birds, fish, and animals. However, these conflate distinct functional modes of information flow between individuals. Computing information measures conditioning on multiple agents requires the proper sampling of a probability distribution whose dimension grows exponentially with the number of agents being conditioned on. Employing simple models of interacting self-propelled particles, we examine the pitfalls of using time-delayed mutual information and transfer entropy to quantify the strength of influence from a leader to a follower. Surprisingly, one must be wary of these pitfalls even for two interacting particles. As an alternative we decompose transfer entropy and time-delayed mutual information into intrinsic, shared, and synergistic modes of information flow. The result not only properly reveals the underlying effective interactions, but also facilitates a more detailed diagnosis of how individual interactions lead to collective behavior. This exposes the role of individual and group memory in collective behaviors. In addition, we demonstrate in a multi-agent system how knowledge of the decomposed information modes between a single pair of agents reveals the nature of many-body interactions without conditioning on additional agents.