Chen Institute Seminar: Kristin Branson

Please join us on Monday, November 9 at 4:00PMin Broad 100 for a Chen Institute Seminar.
Refreshments will be served at 3:30PM, prior to the seminar.

Speaker: Kristin Branson, Senior Group Leader, Janelia

Title: How can generative AI help us understand animal behavior?

Abstract: Supervised machine learning has transformed quantification in science, vastly scaling up the amount of raw data that can be converted into meaningful quantities, for example turning video into animal trajectories or microscopy images into cell segmentations. With these advances, our data can be as complex as the systems we study. The challenge now is to scale up theory- and data-driven discovery to match, allowing us to understand biological processes across levels and in their natural complexity. I will discuss two connected approaches my lab is taking to discover the algorithms underlying animal intelligence. In the first, we train generative models of animal behavior, then dissect how they work using mechanistic interpretability. In the second, AI agents act as computational scientists, analyzing the data and our generative models. Across these approaches, a central question is which kinds of problems each is suited to: which prior knowledge to build into a method and how, how efficiently it can search the space of possible explanations, how interpretable its findings will be, and which problems are better solved by human experts and which by AI. Answering it requires a scalable way to measure success in discovery, which, unlike supervised learning, has no single right answer to compare against. I will describe our efforts in this direction. 

To schedule a 1:1 with Dr. Branson please email [email protected]