The Department of Biology is excited to invite you to a seminar entitled "Creating and interpreting generative models for evolutionary inference in humans and mosquitoes" by Dr. Sara Mathieson, Assistant Professor in the Department of Biology at the University of Pennsylvania on Monday, December 1st at 3pm in BL234. We kindly ask you to make every effort to attend.

Host: Dr. Jody Hey

More information about the speaker here: https://www.bio.upenn.edu/people/sara-mathieson

Abstract: In this talk I will focus on our recent work developing generative methods for population genetic inference. My lab has developed two families of methods, one based on generative adversarial networks (GANs) and one based on transformers. We have applied GAN methods to humans and mosquitoes to understand population size changes, admixture, positive selection, and other deviations from neutrality. We have applied transformer methods to humans to learn latent structure in genomic data and natural selection. Finally, I will discuss our recent efforts to understand the inner workings of trained neural networks. Answering this "interpretability" question - what have complex models learned and what motivates certain predictions - is often more challenging for biological data than for image or text. I will describe our "model-of-the-model" interpretability approach, as well as discuss next directions related to genomic foundation models, privacy preservation, and pedigrees to understand recent evolution.