Together, but not the same: quantifying emergence in collective behaviour
File(s)
Author(s)
Sas, Madalina I
Type
Thesis
Abstract
Adaptive, self-organising collective behaviour, which emerges, or arises from the `bottom-up' without any leader or central control, is found everywhere in the animal kingdom and is crucial to survival. Although it is easy to recognise by the naked eye, we still lack a model-agnostic, quantitative measure of emergence that can detect this collective behaviour, is applicable at multiple scales across different systems, and is as robust in simulations as in experimental data sets. This is particularly important in the study of human group activities, which tend to be studied from a `top-down' hierarchical perspective.
The contributions of this thesis are threefold: first, the Ψ−Γ−Δ criteria of emergence are applied in simple particle systems.
We find unintuitive behaviour of emergence measures in one-dimensional Gaussian and Ornstein-Uhlenbeck random walks, and we propose a number of interpretations.
However, when applied to collective motion in two dimensions in the Vicsek model, Reynolds model and schooling fish, the criteria correctly detect flocking and schooling behaviour in both simulated and experimental datasets of varying complexity, but only when taking into account the redundancy in the system via partial information decomposition.
Secondly, informed by the case studies, we devise, implement and execute a novel experiment which successfully induces collective motion in humans through solving a spatial multi-agent problem which requires optimising the Ψ measure of emergence. Moreover, we find that successful collective motion is related to individual agency and prosocial behaviour: people with more awareness of group strategies are more likely to be part of winning groups, and moreover, members of winning groups show, on average, higher connectedness to others.
Finally, we conclude with an experimental study of synchronous movement in concert audiences, as quantified by wavelet coherence measures, showing that multiscale patterns of group synchrony can be correlated with novel group experiences.
The contributions of this thesis are threefold: first, the Ψ−Γ−Δ criteria of emergence are applied in simple particle systems.
We find unintuitive behaviour of emergence measures in one-dimensional Gaussian and Ornstein-Uhlenbeck random walks, and we propose a number of interpretations.
However, when applied to collective motion in two dimensions in the Vicsek model, Reynolds model and schooling fish, the criteria correctly detect flocking and schooling behaviour in both simulated and experimental datasets of varying complexity, but only when taking into account the redundancy in the system via partial information decomposition.
Secondly, informed by the case studies, we devise, implement and execute a novel experiment which successfully induces collective motion in humans through solving a spatial multi-agent problem which requires optimising the Ψ measure of emergence. Moreover, we find that successful collective motion is related to individual agency and prosocial behaviour: people with more awareness of group strategies are more likely to be part of winning groups, and moreover, members of winning groups show, on average, higher connectedness to others.
Finally, we conclude with an experimental study of synchronous movement in concert audiences, as quantified by wavelet coherence measures, showing that multiscale patterns of group synchrony can be correlated with novel group experiences.
Version
Open Access
Date Issued
2024-12-28
Date Awarded
2026-03-01
Copyright Statement
Attribution-NonCommercial 4.0 International Licence (CC BY-NC)
License URL
Advisor
Jensen, Henrik J
Knottenbelt, William J
Sponsor
Splunk Inc.
Publisher Department
Department of Computing
Publisher Institution
Imperial College London
Qualification Level
Doctoral
Qualification Name
Doctor of Philosophy (PhD)