The Method: Computational Ethology for a Physics of Behavior
Ethology, the study of behavior, has a long history, often involving field observations of animals. Researchers would note down what behaviors they saw, and use the resulting ethograms (time series of behaviors) to develop hypotheses about the purposes of the behaviors they witnessed. Traditional ethology provides rich qualitative data sets, but lacks quantitative rigor. Other methods have emerged more recently that study behavior quantitatively, but at low dimensions – counting the number of times and animal visits its food dish, measuring overall activity levels, etc. These datasets facilitated studies of circadian rhythms, aging-related diseases, and much more, but lack the ability to describe the full set of behaviors and animal engages in.
Within the past decade, new machine learning-based techniques have emerged that now allow us to quantify an animal’s full set of behaviors in an automated fashion, enabling quantitative, high-throughput studies that still preserve the underlying dimensionality of animal behavior. Convolutional neural networks can identify animal posture with extremely high precision. These postures can then be used as the basis for downstream kinematic analysis, unsupervised behavior identification, and dynamical systems approaches. In the McKenzie-Smith Lab, we use computational ethology techniques to study two social systems (outlined below) with the goal of developing models that describe and predict how individual and collective rules and information are integrated in small groups of insects.
Long-Timescale Social Dynamics of the Fruit Fly Drosophila melanogaster
Drosophila melanogaster have a variety of interesting social behaviors, including mating, dominance fights, and group egg-laying. These behaviors can change in response to external factors such as temperature and food availability, and also in response to internal factors such as age and social status. We investigate how D. melanogaster social group dynamics evolve over time, and seek to quantify how social structure, individual behavior, and social interactions influence each other.
Information Flow and Decision Making in Temnothorax Ants
Ants in the genus Temnothorax, known colloquially as acorn ants, live in colonies of 10-80 individuals (averaging around 30) inside acorns, rotting sticks, and rock crevices. Temnothorax ants are eusocial, meaning they must successfully perform collective behaviors in order for the colony to survive and thrive. Many of these collective behaviors, such as foraging or nest location choice, require making group decisions based on information held by only a few ants. We study how specific interactions and behaviors amongst individuals determine how Temnothorax colonies make these collective decisions.