AstroAI Workshop 2025
Ashley Villar
The Trouble with Time Series
Presenter: Ashley Villar (Harvard University)
Title: The Trouble with Time Series
Date/Time: Wednesday, July 11th, 9:30 AM - 11:00 AM
Abstract: While time series data lies at the heart of time-domain astronomy, it touches the limits of common ML methodologies. Unlike the static images that have fueled foundation model success, astronomical time series are irregular, sparse, multimodal, and often unlabeled. In this talk, I’ll explore the challenges and opportunities that arise when applying ML to real-world astronomical time series, from supernovae to AGN. Using examples from anomaly detection, probabilistic inference and multi-modal learning, I will argue that better time series analysis is not just about building bigger models, but also about developing better inductive biases and error handling.
Biography: Ashley Villar uses data-driven methods and machine learning to study the eruptions, mergers and explosions of stars. She is especially interested in utilizing multiband light curves to understand the underlying physics of optical transients. Ashley is an Assistant Professor of Astronomy at Harvard University. Previously, she was the inaugural Mercedes Richards Career Development Professor at Penn State.
Between 07/20 and 07/21, Ashley was a Simons Junior Fellow at Columbia University and the Flatiron Institute. She received her PhD in Astronomy & Astrophysics from Harvard in 2020, where she was a NSF Graduate Research Fellow and Ford Foundation Dissertation Fellow working with Edo Berger. Ashley graduated from MIT in 2014 where she majored in physics and minored in mathematics. Ashley wrote my senior thesis with John Johnson and Josh Winn on asteroseismology. Ashley was inspired to study astronomy after reading Death by Black Hole by Neil deGrasse Tyson.