- This event has passed.
Conor Heins, “Collective behavior from surprise minimization”
5 October, 2023 @ 10:00 am - 11:00 am
Abstract:
Collective motion is a familiar sight in nature; groups of distinct, self-propelled individuals appear to move as a coherent whole, exhibiting a rich behavioral repertoire that ranges from directed movement to milling to disordered swarming. Biological collective motion is an emergent phenomenon that is the result of self-organization, whereby macroscopic patterns arise from decentralized, local interactions among constituent components (e.g., individual fish in a school). Preminent models of collective motion describe individuals in the group as self-propelled particles, subject to a combination of self-generated motion and “social forces” that depend on the state of neighboring particles. Here we introduce a fundamentally new approach to modelling collective movement in animal groups based on active inference, a cognitive framework that casts behavior as consequences of a single imperative: to minimize surprise. We demonstrate that many empirically-observed collective phenomena, including cohesion, milling and directed motion, naturally emerge when considering individual behavior as the consequence of active Bayesian inference — this emerges without ever explicitly building behavioral rules or goals into individual agents. We show that active inference can naturally recover and generalize the classical notion of social forces in agent-based models of collective motion. By analyzing the parameter space of the belief-based model, we reveal non-trivial relationships between the beliefs of individuals and group properties like collective polarization and the probability of occupying different behavioral regimes. We also explore how individual beliefs about uncertainty influence the accuracy of collective decision-making. Finally, we show how, in this framework, agents can readily update their generative model over time, resulting in groups that are collectively more sensitive to external fluctuations and encode information more robustly.
Collective motion is a familiar sight in nature; groups of distinct, self-propelled individuals appear to move as a coherent whole, exhibiting a rich behavioral repertoire that ranges from directed movement to milling to disordered swarming. Biological collective motion is an emergent phenomenon that is the result of self-organization, whereby macroscopic patterns arise from decentralized, local interactions among constituent components (e.g., individual fish in a school). Preminent models of collective motion describe individuals in the group as self-propelled particles, subject to a combination of self-generated motion and “social forces” that depend on the state of neighboring particles. Here we introduce a fundamentally new approach to modelling collective movement in animal groups based on active inference, a cognitive framework that casts behavior as consequences of a single imperative: to minimize surprise. We demonstrate that many empirically-observed collective phenomena, including cohesion, milling and directed motion, naturally emerge when considering individual behavior as the consequence of active Bayesian inference — this emerges without ever explicitly building behavioral rules or goals into individual agents. We show that active inference can naturally recover and generalize the classical notion of social forces in agent-based models of collective motion. By analyzing the parameter space of the belief-based model, we reveal non-trivial relationships between the beliefs of individuals and group properties like collective polarization and the probability of occupying different behavioral regimes. We also explore how individual beliefs about uncertainty influence the accuracy of collective decision-making. Finally, we show how, in this framework, agents can readily update their generative model over time, resulting in groups that are collectively more sensitive to external fluctuations and encode information more robustly.
This talk will take place in person at SCIoI.
Photo kinldy provided by Shintaro Shiba.
Marchstr. 23 – MAR 5-2
10587 Berlin
Email: info@scioi.de
Tel: +49 30 314 77344
