CUDA Spotlight: GPU-Accelerated Swarm Behavior
by Nadeem Mohammad, posted Mar 12 2012 at 10:50PM
This week's spotlight is on Iain Couzin, an Assistant Professor of Ecology and Evolutionary Biology at Princeton University. His expertise is in the study of collective animal behavior.
His lab uses a range of experimental systems - from ants and locust swarms to schooling fish and even human crowds - to explore the fundamental principles that underlie collective behavior across levels of biological organization. This interview is part of the CUDA Spotlight Series.
NVIDIA: Why study the collective behavior of animals?
Iain: Animal groups such as bird flocks, fish schools and insect swarms frequently exhibit complex and coordinated behaviors that result from social interactions among individuals. A fundamental problem in a wide range of biological disciplines is understanding how functional complexity at a macroscopic scale (such as the functioning of a biological tissue) results from the actions and interactions among the individual components (such as the cells forming the tissue). Since they can be readily observed and manipulated, animal groups present unrivaled opportunities to link the behavior of individuals with the functioning and efficiency of the dynamic group-level properties.
NVIDIA: Recently, CNN highlighted your role in helping to predict locust swarms. Are we truly getting closer to being able to predict these events?
Iain: One of the fascinating things about studying collective behavior is that there are always big surprises. One of these was when we discovered that the huge migratory bands of marching locusts are driven not by some cooperative action but rather by cannibalism. When population size exceeds the available resources these insects make the best of a bad situation by turning on each other. What better source of nutrients than the nutritionally balanced meal of another locust?
So each individual is attempting to eat others and avoid being eaten – the outcome is the onset and maintenance of vast mobile swarms with up to several billion insects. There are also benefits to forming such aggregations in that they saturate predators and move effectively out of nutrient poor areas. Understanding this gives us hope that we could use satellite imaging of vegetation quality and distribution as well as weather information to predict when and where swarms form, but this is currently quite far off.
NVIDIA: You and your team recently published a paper on"Uninformed individuals promote democratic consensus in animal groups." What were the key findings?
Iain: Using CUDA-enabled models and experiments with schooling fish, we discovered that uninformed individuals, or those who lack preferences regarding the outcome of group decisions, play a very important role in the process of consensus decision-making.
Whereas previously it has been argued for both animal and human groups that uninformed individuals may promote minority extremist views we found that instead, if in sufficient number, they tend to allow those individuals with preferences to have an equal representation of their opinions within groups. In doing so they effectively prevent those who are strongly opinionated having a disproportionate influence - thus consensus becomes democratic with the majority view being favored.
NVIDIA: Time Magazine referred to your work in an article titled "America Votes with the Fishes"….
Iain: Our work focuses on democratic consensus decision-making in groups, not on complex democratic societies such as our own. We did, however, employ simplified models of opinion and convention formation in which we demonstrate that the principle we reveal for animal groups may extend more broadly to other collective decision-making scenarios.
We would very much like to test whether it may apply to groups of decision-making humans such as in juries or committee meetings. Do those individuals who lack strong preferences facilitate consensus and prevent those with very strong opinions from dominating consensus decisions? We don’t know yet, but it will be very interesting to find out.
NVIDIA: How does GPU computing play a role in your work? What kind of GPU technology is your lab using?
Iain: GPU computing has utterly transformed the science we can do. Not only can we now simulate, for the first time, realistic group sizes - such as millions of individuals in a locust swarm - but we can use the exceptional performance to answer evolutionary questions by simulating groups over ecological and evolutionary timescales. We can do so while embedding them in realistic physical environments.
In addition, the power of GPU computing allows us to employ massively accelerated computer vision so we can now track the motion of hundreds or thousands of individuals in our laboratory experiments and even reconstruct the visual sensory information available to each individual using NVIDIA's OptiX raytracing engine. We currently use eleven C1060 Tesla boards situated in my lab, which is far from enough, and so we are actively pursuing funding for a centralized GPU facility at Princeton University.
NVIDIA: What are some advantages of working with the CUDA programming model?
Iain: Since collective behavior is inherently parallel we can see performance speed up of 200-300 times our previous C++ code. It is an exceptionally cost-effective way for us to utilize high performance computing.
NVIDIA: What new projects are you working on?
Iain: We are very interested in developing 3D tracking and visual field reconstruction and so are beginning to work with Microsoft Kinect and GPU acceleration. We are also starting to investigate the role of the interaction network in large mobile groups, with the aim of understanding how collective decisions are reached and how information propagates through these groups when interactions are local but neighbors are constantly changing.
In addition, we are examining the role of collective behavior in the growth and invasion of cancer tumors, again using GPU computing to simulate not only the patterns of collective movement and signaling among migratory cancer cells, but also the evolutionary dynamics of competing subpopulations within the tumor and the complex interactions between these cells and their microenvironment.
Iain Couzin joined the Princeton faculty in late 2007. Previously he was Royal Society University Research Fellow in the Department of Zoology and Center for Mathematical Biology, University of Oxford, and Junior Research Fellow in the Sciences at Balliol College, Oxford. He did his PhD in Biology at the University of Bath, UK. His work aims to reveal the fundamental principles that underlie evolved collective behavior, and consequently his research includes the study of a wide range of biological systems, from brain tumors to insect swarms, fish schools and human crowds.
His work has been published in leading scientific journals including Nature, Science, PNAS and Current Biology as well as receiving extensive media attention including in the New York Times, TIME magazine, The Economist, Wired, CNN, BBC and ABC. Couzin has won numerous scientific awards in recognition of his contributions to science including the Searle Scholar Award in 2008, Mohammed Dahleh Award and Distinguished Lectureship rom UC Santa Barbara in 2009, Popular Science Magazines "Brilliant 10" Award in 2010and PopTech Science and Public Leadership Award in 2011.
Time Magazine: America Votes with the Fishes
CNN: In Mauritania, Sunny with a Chance of Locusts
icouzin (at) princeton (dot) edu