Column of Akule (SC-215) by Wayne Levin
Column of Akule (SC-215)
Picture by Wayne Levin

Collective behavior is ubiquitous in the animal world and it is observed at all scales. Some examples of groups of gregarious animals exhibiting collective behavior are swarming bacteria, ant colonies, fish schools, bird flocks, ungulate herds and human crowds. All of these systems have something in common: they are composed of a myriad of individuals without leaders nor centralized control and yet they sometimes look as if they were a single entity, a super-organism capable of intelligent action.

The large scale properties of these systems emerge from the local interactions between individuals. Individuals in swarms are like sand grains. On its own a sand grain have some properties but there is not much you can do with it. But when you put many of them together, they form sand, a new medium with its own properties which are completely different from the properties of a single sand grain and which are not necessarily straightforward to deduce without doing the actual experiment. Yet the big difference between a sand grain and for instance a fish or a bird is that the former is passive whereas the latter is active: in models and simulations, we sometimes refer to swarming individuals as self-propelled particles.

Optical flow estimation of fish velocity
Optical flow estimation of fish velocity

My research focuses on the information transfer within animal groups, in particular fish schools. I wish to understand how swarms are so efficient at transferring information over large scale while still being robust to strong external noise and false alarms.

For fish, schooling is more the rule than the exception. Fish are thought to school for different reasons: they might detect predators faster since many eyes are watching out, for the same reason foraging for food might be facilitated, under predator attack, the risk of being captured is diluted, predators might be confused by so many moving targets, and finding a mate for reproduction is easier. There are also some drawbacks, like the increased competition for food for instance.

Fish make a remarkable organism to be studied in the lab. In the labs we use mostly fish which swim in shallow water, so we can let them swim in a tank while recording their trajectories from above. The analysis of these trajectories tells us a lot about the local interactions and the global properties of the school.

Swarm simulation
Simulation of a swarm

Other than experiments, the equally useful tool is computer simulations. Once we have a model of the individual interaction rules and behaviors, we can run simulations with thousands of identical particles, and we can replicate the simulations hundreds of time to get statistically significant results about the global properties emerging from the local scale. When you scale this up, you quickly need a lot of computing power, which we get using the horsepower of GPUs. By executing our simulations on parallel on the hundreds of core of off-the-shelve GPUs, we reduce our computing times and join the effort of democratizing this amazing technology.

For my paper titled The Dynamics of Coordinated Group Hunting and Collective Information Transfer among Schooling Prey published online by Current Biology in , I made a nice video abstract which explains and illustrates the paper. Have a look at it for an example of the research we do in the Couzin Lab.

Dynamics of group hunting and collective evasion in schooling fish