Before he could legally drive, high school student Adam Rebei was already submitting jobs on the Blue Waters supercomputer at the National Center for Supercomputing Applications at the University of Illinois at Urbana-Champaign (NCSA) to run complex simulations of black holes.
“My first time using Blue Waters, we did a tour first and got to see the computer, which is a very amazing thing because it’s a very powerful machine,” Rebei told the NCSA, “and I just remember thinking, ‘All of the GPUs!’ It’s an insane amount of GPUs, and I’ve never seen anything like it.”
To get there, Rebei first took an astronomy class that led him to his work with the NCSA. Once there, he teamed up with research scientist Eliu Huerta, who leads the group’s Gravity Group.
“Adam started in our group by giving talks in our weekly meeting about the history of general relativity and gravitational waves. Soon, though, I realized that he had the skills and determination to do research at the level of a graduate student. He has since participated in multiple publications that have enabled him to learn about how to combine high performance computing, numerical relativity, and deep learning to advance our knowledge in gravitational wave astrophysics,” Huerta said.
Rebei’s story is being highlighted today on the NCSA’s blog, describing his journey as a student to published author.
Using the NVIDIA GPU accelerated Blue Waters supercomputer at the National Center for Supercomputing Applications, Rebei focused his work on the morphology of the gravitational waves that could be produced by two black holes that collide following eccentric or elliptical orbits.
Current methods used to detect gravitational waves can identify that a dark matter event has happened but not what type of event. They also don’t recognize complex events where dark matter objects aren’t simple shapes.
Rebei saw the opportunity to build upon the first generation algorithms developed by Huerta’s lab to achieve higher detection accuracy. Adam’s models were trained within thirty minutes using 64 NVIDIA GPUs on Blue Waters using the Horovod deep learning framework.
“Building upon our pioneering deep learning work, we show for the first time that machine learning can accurately reconstruct higher-order waveform multipole signals from eccentric binary black mergers embedded in real LIGO data,” Rebei wrote and his co-authors wrote in one of their papers.
The Blue Waters supercomputer is equipped with NVIDIA Tesla GPUs and has a peak performance of 13.34 petaflops.
Rebei is currently finishing up his senior year of high school in Illinois. With three publications already in his portfolio, including the one above in which he is a lead author, Rebei plans to study astrophysics or physics at Princeton in the fall.
You can read the full story on the NCSA’s blog.