Snigdaa Sethuram 
Advisor:  Dr. John Wise, School of Physics, Georgia Institute of Technology

Machine Learning Techniques to Accelerate Simulations, Modeling, and Analysis
Friday, March 28, 2025  
2:00 p.m. 
Gilbert Hillhouse Boggs Building, Room 1-44 (Visualization Laboratory)
Zoom link

Committee Members:           
Dr. Tamara Bogdanovic, School of Physics, Georgia Institute of Technology
Dr. Gongjie Li, School of Physics, Georgia Institute of Technology
Dr. Annalisa Bracco, School of Earth & Atmospheric Sciences, Georgia Institute of Technology
Dr. Viviana Acquaviva, Department of Physics, CUNY NYC College of Technology

Abstract:
Computational methods in astrophysics are essential to better understanding our universe and validating theoretical models but achieving high-resolution simulations with complex physics remains computationally intensive. Machine learning (ML) methods have proven to be a valuable tool to mitigate these challenges and enable accelerated simulation runtime and analysis. In this work, data-driven approaches to traditional tasks are explored: (1) a neural network emulator that generates synthetic observations of simulated galaxies given global data, bypassing traditional radiative transfer, (2) a Bayesian framework employing Markov Chain Monte Carlo sampling to infer galaxy properties from JWST photometric observations, and (3) a convolutional recurrent network trained on spatiotemporal hydrodynamic data to emulate stellar feedback in cosmological simulations. These approaches integrate ML and statistical analysis methods with physical models to optimize parameter space exploration, reduce computational costs, and bridge simulations with observational data.