The student will investigate the relationship between individual and collective computation in schooling fish. In a controlled environment the behaviour of schooling fish will be studied to learn how the individual and collective behavioral mechanisms adapt to external stimuli.

Collective computation in large animal groups

Behavioral Neuroscience
University of Konstanz

Despite the fact that social transmission of information is vital to many group-living animals, the organizing principles governing the networks of interaction that give rise to collective properties of animal groups, remain poorly understood. ER6 will employ an integrated empirical and theoretical approach to investigate the relationship between individual computation (cognition at the level of the ‘nodes’ within the social network) and collective computation (computation arising from the structure of the social network). The challenge for individuals in groups is to be both robust to noise, and yet sensitive to meaningful (often small) changes in the physical or social environment, such as when a predator is present. There exist two, non mutually-exclusive, hypotheses for how individuals in groups could modulate the degree to which sensory input to the network is amplified; 1) it could be that individuals adjust internal state variable(s) (e.g. response threshold(s)), effectively adjusting the sensitivity of the “nodes” within the network to sensory input and/or 2) it could be that individuals change their spatial relationships with neighbors (such as by modulating density) such that it is changes in the structure and strength of connections in the network that modulates the information transfer capabilities, and thus collective responsiveness, of groups. Using schooling fish as a model system we will investigate these hypotheses under a range of highly controlled, ecologically-relevant scenarios that vary in terms of timescale and type of response, including during predator avoidance as well as the search for, and exploitation of, resources. We will employ technologies such as Bayesian inference and unsupervised learning techniques developed in computational neuroscience and machine learning to identify, reconstruct, and analyze the directed and time-varying sensory networks within groups[9], and to relate these to the functional networks of social influence. As in neuroscience, we care about stimulus-dependent, history-dependent discrete stochastic events, including burstiness, refractoriness and habituation and throughout we will seek to isolate principles that extend beyond the specificities of our system.

[9]Rosenthal, S.B., Twomey, C.R.,  Hartnett, A., Wu, H.S., & Couzin, I.D. (2015) Proc. Natl. Acad. Sci. USA112(15), 4690-4695.

Expected Results:

1. To reveal the structure of the time-varying, weighted and directed networks of social influence in large animal groups

2. To understand how different local, and global, configurations enhance or reduce the effectiveness of information spread in social networks

3. To determine how simultaneous robustness and responsiveness can be achieved in animal social networks

Planned secondment(s):

1. Panayiota Poirazi, IMBB-FORTH, M16-19: inference and modeling of biological networks

2. Jacco Wallinga, RIVM, M31-33: apply the developed network analysis and modelling tools to models of disease spreading and compare the results.

Enrolment in Doctoral degree(s): You will be enrolled in the International Max Planck Research School for Organismal Biology at the University of Konstanz.