AstroAI Lunch Talks - August 17, 2026 - Khushboo Kunwar Rao & Rocco Di Tella
17 Aug 2026 - Joshua Wing
The video can be found here: https://www.youtube.com/watch?v=RJyVknTujkI
Speaker 1: Khushboo Kunwar Rao
Title 1: Blue Straggler Stars in the Big-Data Era: The Role of Machine Learning
Abstract 1: Blue straggler stars are prominent tracers of stellar interactions in star clusters. They are rejuvenated core hydrogen-burning stars that are bluer and brighter than the main-sequence turnoff stars of a cluster. Their properties can provide insight into binary evolution, dynamical interactions, and also the dynamical ages of their host clusters. A reliable identification of cluster members and blue stragglers is therefore an essential first step. In this talk, I will present ML-MOC, a machine-learning method for identifying star-cluster members from Gaia astrometric data, and show how it enables the construction of a robust sample of blue straggler stars. I will then discuss how their spatial distributions can be used to investigate the dynamical state of star clusters, with particular emphasis on open clusters and their possible formation channels. Finally, I will discuss future opportunities for machine-learning based approaches in this field. The forthcoming Gaia DR4 and Rubin Observatory data will provide increasingly rich information on stellar variability, binarity, and cluster populations, not only in the Milky Way but also in nearby galaxies. These large and heterogeneous data sets will create a need for efficient methods to overcome current limitations in constraining blue-straggler formation and evolutionary histories and to improve our understanding of their mass-luminosity relation.
Speaker 2: Rocco Di Tella
Title 2: Exploring the connection between optical and X-ray emissions in Active Galactic Nuclei
Abstract 2: Supermassive black hole accretion is most directly visible in X-rays, while host galaxies are richly observed in optical and infrared surveys. Quantifying how much X-ray information is shared with these non-X-ray modalities constrains how accretion, gas, dust, and galaxy structure are observationally connected. We attach a probabilistic normalizing-flow head to frozen AION multimodal embeddings and predict eROSITA broad-band X-ray flux from DESI spectra, Legacy Survey imaging, WISE mid-infrared photometry, and redshift. The full multimodal model outperforms a classical optical emission-line baseline. Information gain depends on the available modality combination, source class, and redshift: spectra and imaging dominate at low redshift while WISE and redshift become more informative at high redshift.