AI/ML Conference Helps School of Physics Launch New Research Initiative
The School of Physics’ new initiative to catalyze research using artificial intelligence (AI) and machine learning (ML) began October 16 with a conference at the Global Learning Center titled Revolutionizing Physics — Exploring Connections Between Physics and Machine Learning.
AI and ML have the spotlight right now in science, and the conference promises to be the first of many, says Feryal Özel, Professor and Chair of the School of Physics.
"We were delighted to host the AI/ML in Physics conference and see the exciting rapid developments in this field,” Özel says. “The conference was a prominent launching point for the new AI/ML initiative we are starting in the School of Physics."
That initiative includes hiring two tenure-track faculty members, who will benefit from substantial expertise and resources in artificial intelligence and machine learning that already exist in the Colleges of Sciences, Engineering, and Computing.
The conference attendees heard from colleagues about how the technologies were helping with research involving exoplanet searches, plasma physics experiments, and culling through terabytes of data. They also learned that a rough search of keyword titles by Andreas Berlind, director of the National Science Foundation’s Division of Astronomical Sciences, showed that about a fifth of all current NSF grant proposals include components around artificial intelligence and machine learning.
“That’s a lot,” Berlind told the audience. “It’s doubled in the last four years. It’s rapidly increasing.”
“It’s tool development, the oldest story in human history,” said Germano Iannacchione, director of the NSF’s Division of Materials Research, who added that AI/ML tools “help us navigate very complex spaces — to augment and enhance our reasoning capabilities, and our pattern recognition capabilities.”
That sentiment was echoed by Dimitrios Psaltis, School of Physics professor and a co-organizer of the conference.
“They usually say if you have a hammer, you see everything as a nail,” Psaltis said. “Just because we have a tool doesn't mean we're going to solve all the problems. So we're in the exploratory phase because we don't know yet which problems in physics machine learning will help us solve. Clearly it will help us solve some problems, because it's a brand new tool, and there are other instances when it will make zero contribution. And until we find out what those problems are, we're going to just explore everything.”
That means trying to find out if there is a place for the technologies in classical and modern physics, quantum mechanics, thermodynamics, optics, geophysics, cosmology, particle physics, and astrophysics, to name just a few branches of study.
Sanaz Vahidinia of NASA’s Astronomy and Astrophysics Research Grants told the attendees that her division was an early and enthusiastic adopter of AI and machine learning. She listed examples of the technologies assisting with gamma-ray astronomy and analyzing data from the Hubble and Kepler space telescopes. “AI and deep learning were very good at identifying patterns in Kepler data,” Vahidinia said.
Some of the physicist presentations at the conference showed pattern recognition capabilities and other features for AI and ML:
- Cassandra Hall, assistant professor of Computational Astrophysics at the University of Georgia, illustrated how machine learning helped in the search for hidden forming exoplanets.
- Christopher J. Rozell, Julian T. Hightower Chair and Professor in the School of Electrical and Computer Engineering, spoke of his experiments using “explainable AI” (AI that conveys in human terms how it reaches its decisions) to track depression recovery with deep brain stimulation.
- Paulo Alves, assistant professor of physics at UCLA College of Physical Sciences Space Institute, presented on AI/ML as tools of scientific discovery in plasma physics.
Alves’s presentation inspired another physicist attending the conference, Psaltis said. “One of our local colleagues, who's doing magnetic materials research, said, ‘Hey, I can apply the exact same thing in my field,’ which he had never thought about before. So we not only have cross-fertilization (of ideas) at the conference, but we’re also learning what works and what doesn't.”
- Workflow Status:Published
- Created By:Renay San Miguel
- Modified By:jhunt7