What is it about?

Complex systems such as human social groups or markets often are built from a large number of actors with diverse behaviors and interactions. In order to understand and control these systems, we need a way of characterizing the group behavior. This work provides a way to summarize the large scale behavior and how it is changing without needing a theory or model beforehand.

Featured Image

Why is it important?

By translating distributed measurements of a system (e.g. who has messaged whom in a social network) into a small number of variables that capture the system's organization, we can form and test theories on what drives the system and what leads to large-scale changes; for example, a sudden migration or a market crash. Essentially, rather than becoming an expert on the system, forming a theory, and testing it (top-down approach), one can now gather data on the individual parts of the system, calculate, and see whether and how the system is structured (bottom-up approach).


This article involves a nice proof-of-concept for the method: I have no background in the group dynamics of animals, but by applying the tool I was able to refine the standard classifications of schooling behavior of Golden Shiners (a fish species). I hope to see this work lead to more detailed analyses that are able to tease out leading indicators of oncoming large-scale changes, which will be necessary to understand risks in ecosystems, markets, and other systems.

Mathew Titus

Read the Original

This page is a summary of: Unsupervised manifold learning of collective behavior, PLoS Computational Biology, February 2021, PLOS,
DOI: 10.1371/journal.pcbi.1007811.
You can read the full text:

Open access logo


The following have contributed to this page