Contact Information
1002 W. Green St.
Urbana, IL
M/C 221
Research Areas
Biography
I am a fifth year graduate student working with Prof. Xin Liu on Big Data problems in survey science. I am a member of LSST DESC, and am involved in the photmetric redshifts and pixels-to-objects working groups. I am also a mentor in the Students Pushing Innovation program through NCSA. Before my time at UIUC, I attended the Honors Tutorial College at Ohio University where I graduated with a Bachelor's in Astrophysics in 2020 and a Master's in Physics in 2021. At OU, I began astrophysics research by examining the production of Urca nuclides in Type-1 Xray bursts. Later, I was involved in DESI and SDSS, where I tested how different systematic mitigation schemes affected the Baryon Acoustic Oscillation signal. In my free time I enjoy hiking, playing racquetball and homebrewing. I'm always curious about new AI techniques, particularly in computer vision.
Research Interests
I develop AI computer vision models to aid in source detection, deblending and measurement. I am the lead developer of DeepDISC, an AI model designed to produce source catalogs from survey data. Turning pixels into catalogs involves many sources of systematic error, which traditional pipelines often have trouble mitigating. Understanding and calibrating observational systematics is critical for surveys like LSST, which will use measurements of billions of galaxies to test our cosmological models. A main focus of my work has been photometric redshift estimation. LSST (among many other surveys) does not have spectroscopic capabilities, so object redshifts must be estimated from photometry alone. I use DeepDISC to estimate photo-zs directly from images, and have tested and validated the model on simulated and real data.
Education
PhD. Astronomy, UIUC, 2021-present
M.A. Physics, Ohio University 2021
B.S Astrophysics, Ohio University 2020
Grants
LSST LINCC Frameworks Incubator - DeepDISC photo-z
- This project consisted of developing a photometric redshift module on top of DeepDISC, and wrapping the code in RAIL, a package designed to house mayn different photo-z estimators. We then tested and validated photo-z results on simulated LSST data and compared to other algorithms.
Awards and Honors
Ranked as "Excellent" Teaching Assistant for 2021-2022 Academic Year
LSST Data Science Fellowship Program, 2023-2025
La Serena School for Data Science 2021
2025 Fiddler Innovation Graduate Student Fellowship Award
Courses Taught
Phys 211 University Physics: Mechanics
Astr 121 Solar System and Worlds Beyond
Astr 404 Stellar Astrophysics
External Links
Highlighted Publications
Merz, G., Liu, X., Schmidt, S., et al., DeepDISC-photoz: Deep Learning-Based Photometric Redshift Estimation for Rubin LSST, 2025, OJA, 8,40
Merz, G., Liu, Y., Burke, C., et al., Detection, instance segmentation, and classification for astronomical surveys with deep learning (DEEPDISC): DETECTRON2 implementation and demonstration with Hyper Suprime-Cam data, 2023, MNRAS, 526, 1
The RAIL Team, van den Busch, J. L., Charles, E., ... Merz, G.,, et al., Redshift Assessment Infrastructure Layers (RAIL): Rubin-era photometric redshift stress-testing and at-scale production, OJA, submitted