Ming-Feng Ho, a doctoral student in the Department of Physics and Astronomy, has received a three-year fellowship from Future Investigators in NASA Earth and Space Science and Technology, or FINESST, which supports research projects designed and performed by graduate students. 

Ho is one of 21 students to receive the prestigious fellowship out of 196 applicants.

The $135,000 fellowship will allow Ho to develop a statistical inference framework for a future NASA space mission, the Nancy Grace Roman Space Telescope. The framework, called a multifidelity emulator, or MFEmulator, uses a machine-learning model that could help researchers understand the nature of dark matter and dark energy at an affordable computational cost using numerical simulations.

“The most exciting aspect of our MFEmulator is that it allows astrophysicists to infer the parameters of the universe using numerical simulations,” said Ho, who expects to graduate in 2023 and works with Simeon Bird, an assistant professor of physics and astronomy. “Previously, training an emulator required about 40 expensive high-resolution simulations for inferring parameters in a 5D space. We can train an emulator using only three high-resolution simulations, making this fast-to-train emulator more practical for future cosmological surveys.”

Ho explained that, traditionally, numerical simulations are used to understand the theoretical meaning of observations. When astrophysicists forward-simulate the universe to match observations, inferring the theoretical parameters can be tremendously difficult and is known as an “inverse problem.”

“The conventional technique to solve this inverse problem uses sampling methods that require millions of samples to infer the parameters,” Ho said. “Running millions of high-resolution simulations, however, is almost impossible. A multifidelity emulator solves this problem by using fewer simulations.”

Ho will use UCR’s High-Performance Computing Center and Texas Advanced Computing Center’s Stampede and Frontera to perform the cosmological simulations.

“Ming-Feng is an excellent candidate for the fellowship,” Bird said. “His research has been very successful, combining physics, machine learning, and mathematics. It will be critical for the Roman Space Telescope, NASA’s next flagship mission. He is also an active positive member of the graduate student community, helping to organize an annual student conference and mentoring high school students.”