Talk 3

Time and Venue

3 October 2025, 11:30 AM to 12:30 PM (ET)
Klaus 1116, Klaus Advanced Computing Building (KACB), Georgia Tech

Title and Abstract

Time-Domain Astrophysics in the Era of Foundation Models

Astronomical surveys now routinely generate petabytes of imaging data across the electromagnetic spectrum. Hidden in these datasets are variable and explosive phenomena, from stellar eruptions to core-collapse supernovae and stellar tidal disruptions by massive black holes. Detecting these sources early and prioritizing follow-up observations requires machine learning models that learn from noisy, heterogeneous observations and generalize across instruments. In this talk, I will highlight two major advances underpinning the rise of foundation models for time-domain astrophysics: (1) contrastive learning to align spectra and light curves into a shared representation for joint inference; and (2) transformer-based encoders that extract temporal correlations from long, irregular sequences. I will present two models developed with colleagues to incorporate these techniques, and outline the major scientific questions we can answer by rapidly inferring a transient’s physical properties and its evolution in the gaps between our observations.

Speaker

Alex Gagliano

Postdoctoral Research Fellow,
Institute For AI and Fundamental Interactions (IAIFI)

Alex Gagliano is an NSF IAIFI Fellow at MIT and the Harvard–Smithsonian Center for Astrophysics. He develops scalable machine learning for real-time inference on astrophysical transients and studies the final years of massive stars. He earned his Ph.D. at the University of Illinois and has held appointments at the Flatiron Institute’s Center for Computational Astrophysics and Los Alamos National Laboratory.