Joshua Speagle, University of Toronto
When
Refreshments served at 3:00 in the 3rd floor atrium
TAP Computation & Data Initiative Lecture
Joshua Speagle, University of Toronto
Title: A Deep Dive into Data-Driven Astrophysics
Abstract: As astronomical datasets reach unprecedented size and scale, how much should we rely on hand-crafted physical models versus letting flexible architectures learn directly from the data? In this talk, I will take us on a whirlwind (but accessible!) tour across astrophysics, statistics, and machine learning to argue two main points: first, that computation and scale now enable true "data-driven discovery" directly from raw data products; and second, that in the era of foundation models and agentic AI, we must avoid repeating Rich Sutton’s "bitter lesson" of over-counting our own domain expertise. I will illustrate these ideas (as time permits) with recent work spanning foundation models for galaxy images and spectra, stellar variability from light curves, and cross-survey harmonization of stellar abundances—exploring both why these models work so unreasonably well and how we can rigorously interpret their outputs. (em/en dash intended!)
Bio: Joshua Speagle is an Assistant Professor of Astrostatistics at the University of Toronto, jointly appointed in the Department of Statistical Sciences and the David A. Dunlap Department of Astronomy & Astrophysics, where he co-leads the Astrostatistics Research Team (ART) with Gwen Eadie. He is also an Associate Member of the Dunlap Institute, a Member of the Data Sciences Institute, and Interim Director of the Data Science concentration in the Master of Science in Applied Computing (MScAC) program. His research combines astronomy, statistics and machine learning to study how stars and galaxies form and evolve, including the development of open-source research software. He received his undergraduate and doctoral degrees from Harvard and was a Banting and Dunlap Postdoctoral Fellow at Toronto before joining the faculty in 2022.
For All TAP Events: Zoom Meeting ID: 417 674 3144 Passcode: 1985Astro