Matt Repasky

CV | LinkedIn | GitHub | ORCID

About Me

I am a Machine Learning Ph.D. candidate in the H. Milton Stewart School of Industrial and Systems Engineering at Georgia Tech advised by Dr. Yao Xie. My interests are at the intersection of generative modeling, high-dimensional statistics, and planetary geophysics. My doctoral research has focused on leveraging deep learning tools for statistical modeling. Recently, I have been developing generative modeling to solve problems related to posterior sampling and inverse problems. This includes an ongoing collaboration with NASA Goddard Space Flight Center to develop a super-resolution technique for lunar topograhy models.

During my Ph.D., I have developed novel methods for solving problems in statistics and decision making. This includes:
(1) A posterior sampling method to sample within the noise space of pre-trained generative models.
(2) A multi-agent reinforcement learning strategy to select within combinatorial action spaces under stochastic state transitions, with an application to patrol and dispatch operations.
(3) A neural neural statistical goodness-of-fit optimization technique which can localize regions of poor fit in models of probability distributions, such as generative models.

I am primarily interested in applications to the natural sciences, including planetary science, energy systems, and materials. Such applications include the development of methods to model and detect corrosion in coated materials to reduce the environmental impact of the coating industry and a technique for physically characterizing and denoising high-dimensional scanning probe microscopy data.

During my internships at NASA Goddard Space Flight Center and under the supervision of Dr. Erwan Mazarico, I trained generative AI models to improve the resolution of planetary topography data given simulated optical measurements from satellite orbiters. As an intern at Argonne National Laboratory under the supervision of Dr. Feng Qiu, I developed a low-rank tensor modeling scheme to categorize failures in a power grid given sensor array data.

In May 2021, I earned a BS in Physics with a concentration in Astrophysics from Georgia Tech, working in the Computational Cosmology Group run by Dr. John Wise.

Other Info