Department
Physics
Major
Physics and Computer Science
Research Advisor
Saito, Shun
Advisor's Department
Physics
Funding Source
National Science Foundation (NSF)
Abstract
Applications of machine learning (ML) or artificial intelligence (Al) to problems in astrophysics and cosmology have recently entered a golden era. In response, we have updated two of our recent ML/Al efforts that contribute to galaxy surveys whose main scientific target is to reveal the nature of the Comsic Acceleration or Dark Energy. We first revised our effort to infer cosmological information beyond the survey geometry using Graph Neural Networks (GNNs) to take advantage of supercomputing resources on campus. We then updated our methods for galaxy target selection in the Subaru Prime Focus Spectrograph (PFS) survey with modern reinforcement learning techniques such as A2C and revised reward functions. In both cases, we found that these changes dramatically improved the speed and performance of the ML models, especially when locating faint galaxies with high redshifts.
Biography
Steven Karst is a senior from Ballwin, Missouri majoring in Physics and Computer Science at Missouri S& T, where he is the Computing Lead for the Underwater Robotics Team as well as the vice president of ACM Game Dev. He was originally introduced to the Institute for Multi-Messenger Astrophysics and Cosmology through the National Merit Semifinalist Package. His research for that program won first prize at the 2021 Undergraduate Research Conference as well as at the 2021 Fuller Prize Competition.
Research Category
Sciences
Presentation Type
OURE Fellows Final Oral Presentation
Document Type
Presentation
Location
Havener Center - Carver Room
Presentation Date
10 April 2024, 1:00 pm - 4:00 pm
Included in
Enhancing Galaxy Surveys with Machine Learning
Havener Center - Carver Room
Applications of machine learning (ML) or artificial intelligence (Al) to problems in astrophysics and cosmology have recently entered a golden era. In response, we have updated two of our recent ML/Al efforts that contribute to galaxy surveys whose main scientific target is to reveal the nature of the Comsic Acceleration or Dark Energy. We first revised our effort to infer cosmological information beyond the survey geometry using Graph Neural Networks (GNNs) to take advantage of supercomputing resources on campus. We then updated our methods for galaxy target selection in the Subaru Prime Focus Spectrograph (PFS) survey with modern reinforcement learning techniques such as A2C and revised reward functions. In both cases, we found that these changes dramatically improved the speed and performance of the ML models, especially when locating faint galaxies with high redshifts.
