Special Seminar: Machine Learning at the Cosmic Frontiers - Discovering what shaped the past lives of galaxies with novel astrostatistical methods

Feb

2025

12:00 pm – 1:00 pm

(view details)

Featured Speaker(s): Kartheik Iyer - Columbia University

Special Seminar by Kartheik Iyer of Columbia University

Description

Machine Learning at the Cosmic Frontiers: Discovering what shaped the past lives of galaxies with novel astrostatistical methods

Abstract: Galaxies are fascinating, complex systems. While we understand their average evolution, current models struggle to explain the varied lives of individual galaxies, and the incredible diversity in galactic populations we observe today. Part of the issue is their inherently multiscale nature - a range of physical processes (from halo-scale accretion to GMC-scale feedback) act over a range of spatial and temporal scales to regulate galaxy growth and quenching. Like other complex systems (e.g., climate models), understanding galaxies thus requires sophisticated data-driven approaches. As it happens, we are at the dawn of an era of both unprecedented observations from current and upcoming (HST, JWST, Euclid, Roman, Rubin and the ELTs) facilities, and incredible advances in the development of statistical and machine learning techniques for emulation, prediction and inference.

My research lies at the intersection of astrophysics x ML - developing and applying novel astrostatistical methods and computational tools that connect modern cosmological simulations to large observational datasets - to build a spatiotemporal picture of galaxy evolution by deciphering the subtle imprints that physical processes leave in starlight. I will discuss how this work (i) has led to the development of new methods and data products (Dense Basis, GP-SFH, Katachi, Pathfinder) that are valuable to the wider scientific community, (ii) is being applied to observations from JWST, leading to exciting discoveries about how galaxies form stars in the early universe, and (iii) will address the unique challenges (and opportunities!) posed by large, noisy datasets from upcoming observations. I will close with how UT Austin - with its strong galaxy formation expertise at the CFC, AI/ML prowess at CosmicAI, and strong computational facilities via TACC and the Oden Institute - provides an unparalleled environment to lead the development of the observational and computational infrastructure needed for this data-driven era of galaxy evolution research.