By Lina Sorg
Computational ecology is a combination of science with mathematics and computational techniques, according to Tanya Berger-Wolf of the University of Illinois at Chicago. And behavioral ecology is the study of ecological and evolutionary bases for animal behavior, and the subsequent roles of that behavior. At the SIAM Annual meeting, Berger-Wolf gave an invited talk entitled “Computational Behavioral Ecology.” In the talk, she demonstrated a variety of computational approaches for understanding animal interactions, social groups, and collective behavior.
With new and novel data collection equipment—such as unmanned aerial vehicles, advanced high definition cameras, and GPS systems—collecting data for the study of nature and wildlife is becoming easier than ever. Specific techniques include placing GPS collars on larger animals such as zebras and baboons, painting bees with numbers, tattooing florescent numbers on frogs, and gluing tiny antennae with radio transmitters on insects. These methods, combined with genotyping tools and image contributions from citizen scientists, yield an explosion of information. The challenge comes with analyzing the hundreds of thousands of images for animal behavior.
But because our ability to analyze and interpret this data have not yet caught up to data collection techniques, computational approaches are necessary to yield results. “We don’t see patterns in massive noisy data very well as humans,” Berger-Wolf said, “so we need systems of computational approaches or mathematical approaches that can deal with massive data and produce patterns. The challenge is to find patterns and models that actually produce new hypotheses. And that these hypotheses are testable. Otherwise they may be interesting but may not teach us anything new.”
The application of traditional rule-based behavior is often limiting, because behaviors don’t always neatly fit into hypothetical boxes. Instead, machine learning techniques can characterize both individual and pair-wise behavior and sort it into different models, such convolutional nets. Researchers want to characterize both individual and pair-wise behavior, like grooming. This is a labor-intensive process, and Berger-Wolf’s colleagues are working on new active learning techniques to solicit new labels in a prudent way. The use of big data analytics and advanced technologies takes into account fluid aspects of animal interactions, such as animal leadership and changing affiliations. And analysis of multidimensional time series data reveals truths about efforts of shared decision-making that drives collective movement in baboons.
A combination of advanced analysis techniques forms a new hypothesis that begins to answer the following fundamental questions: “Why are animals social? Does sociality matter? How does it matter? Where does it matter?” Using the aforementioned measures, Berger-Wolf and her group is able to assess wildlife health and habitat and obtain a big-picture view of animal dynamics, specifically in baboons and zebras.