Princeton prof tracks biology with GPU’s

 May 16, 2012, GPU technology conference, San Jose, CA—Iain Couzin from Princeton University described his work in investigating collective behaviors. This research applies not only to groups of animals, but also to human body properties.

Everyone has seen collective behaviors; schooling fish and clusters of birds flying in the air. These activities demonstrate interrelated dynamics and are sometimes called group intelligence. The basic pictures require interactive node maps and a collective brain. These dynamics scale from microbiology up through groups of people.

The guiding principles for a group motions are first, to avoid collisions and second, when separations become too large, to move back towards the group. These two simple rules define the attraction and alignment letters for all members of the group. The challenge for researchers is to do real-time tracking and synchronization with tens of thousands of elements.

To perform these measurements requires at least four computers with Tesla GPU’s. The availability of large quantities of high-speed compute elements is a democratizing force in many research areas. These tools provide visions to change the rules and rule parameters to handle complex, high-volume interactions.

The introduction of zones of activity changes the group patterns. If there is a null zone, the group tends to move in a circular pattern. Without this null zone, the group motion becomes random. The result is groups of organisms behave like a living fluid with abrupt transitions to different states which are described as swarms or polarization.

Their research is evolving behavioral rules to address the group dynamics in the presence of various environmental conditions including the group behavior when attacked by a predator. The rules are modified to avoid the predator first, and then the other rules apply. These rules for behavior have been verified in both lab settings and in natural environments.

They also have developer rules for leadership and collective decision making. When an individual has disparate information from the rest of the group, new signaling models are needed to address goal-oriented versus social responsibilities. There is a critical mass of decision-makers that is inversely proportional to group size. These new dynamics impact the accuracy of the information transfer within the group. The bulk of the members of the group use social inputs, but the leaders are separate.

Generally, 99 percent of a group follow the majority decisions, but splits are possible. This democratic consensus addresses conflicting interests, the cost of the decision versus consensus. The good thing is that although a strong minority could dominate, the addition of uninformed individuals changes the dynamics. Increasing the number of uninformed individuals, amazingly enough, brings the system into balance. The influence of the leaders on individuals gets overtaken by the adaptive network model, and the uninformed individuals increase the influence of the majority.

One important application of this research is tracking mass migrations of insects. It turns out that locusts, which affects about one in 10 people on the planet, tend to align at higher densities. One reason for locusts swarms move as rapidly and as far as they do, is because locusts are actually omnivores. The predators are those locusts behind the front edge of the swarm. The result is a group behavior which includes the predators as a part of the mass.