Stat Cafe - Dr. Toryn Schafer
Computational strategies for hierarchical Bayesian models of ecological and behavioral data
- Time: Tuesday, 10/13/2026 from 11:10AM to 12:25PM
- Location: BLOC 448
Description
Hierarchical Bayesian models are widely used in statistical ecology because they separate the observation process from the ecological process of interest. However, inference for these models is often limited by the likelihood. Specifically, the likelihood may be expensive to evaluate, may prevent the sampler from mixing, or may not be available in closed form. In this talk, I will first present a multi-stage recursive Bayes approach for fitting spatial point process models, motivated by the problem of estimating harbor seal pup abundance in Glacier Bay National Park from aerial imagery. We partition the model such that the most expensive computation can be performed in parallel, resulting in efficient Bayesian inference. Furthermore, a recent extension to marked point processes for the same data uses variational inference to construct a joint proposal distribution when the parameter space expands between stages. I will then describe three ongoing projects with open methodological problems, which are (i) estimating migratory bird communities from the Motus tracking network, where sampling effort is confounded with the ecological signal, (ii) modeling lifetime sleep trajectories of individual fruit flies in relation to aging and Alzheimer’s disease risk, and (iii) neural estimation for animal movement models fit to GPS telemetry data, for which simulation from the model is straightforward but the likelihood is intractable.
Our Speaker
Dr. Toryn Schafer is an Assistant Professor in the Department of Statistics at Texas A&M University. She received her Ph.D. in Statistics from the University of Missouri in 2020, where she was an NSF Graduate Research Fellow, and was a postdoctoral associate at Cornell University prior to joining Texas A&M in 2022. Her research develops Bayesian hierarchical models and computational methods for ecological and environmental data, with applications to animal movement and behavior, and has been funded by the National Park Service and Sandia National Laboratories.