NSF AI Planning Institute Holds Interdisciplinary Virtual Workshop
By Ben PankoMedia Inquiries
- Associate Dean for Communications, MCS
Quarks to Cosmos with AI, a virtual workshop hosted this month by Carnegie Mellon University's National Science Foundation (NSF) AI Planning Institute for Data-Driven Discovery in Physics, brought together hundreds of physicists and data scientists to combine their expertise on machine learning and artificial intelligence (AI) and physics.
"AI plays an increasingly important role in all aspects of data analysis in particle and astrophysics, but there are few opportunities for particle physicists, astrophysicists and data scientists to get together and exchange thoughts on how to best incorporate the latest advances in AI in physics data analyses," said Mikael Kuusela, an assistant professor of statistics and data science in the Dietrich College of Humanities and Social Sciences and a co-organizer of the event. "This conference provided a unique opportunity for this interdisciplinary exchange both in the form of scientific talks and hands-on data challenges."
As hosts of the workshop, Carnegie Mellon was able to showcase its expertise in AI, machine learning, data science and physics while learning from their colleagues. Over the course of a week, more than 200 academics, industry professionals and students logged in to hear a variety of lectures from leading researchers in physics and machine learning. A virtual conference presented both benefits and challenges explained Professor of Physics Manfred Paulini, a co-organizer of the workshop.
"Although no coffee breaks and food needed to be organized, the conference included a social hour every day utilizing GatherTown, creating chat opportunities for participants to support networking and community building," Paulini said. While social interactions were more challenging, noted Kuusela, the virtual format allowed a diverse array of participants from many different countries to join the event.
A major highlight of the workshop was daily hackathons using datasets and computing resources provided by the Pittsburgh Supercomputing Center. These events allowed scientists at a variety of stages in their careers to collaborate in analyzing and solving data challenges.
"I particularly liked that one group worked on a class of problems called inverse problems, and implemented solutions to a variety of different inverse problems in cosmology using deep learning," said Professor of Physics Rachel Mandelbaum, a co-organizer of the workshop. "The range of problems they were solving with a common tool set was quite impressive, and the results compared really well to other methodology already in use."
"I was really impressed by the amount of work that the hackathon participants had achieved during the week," said Kuusela.
Kuusela said he had personally worked on calorimetric muon energy reconstruction using machine learning during the hackathon. "I obtained encouraging preliminary results that indicate that this approach might reduce the bias of the estimated energies and I will keep working on this problem with one of my students here at CMU."
"For me, the most important result of the hackathon was the engagement of the students and young researchers, who had the opportunity to learn about modern machine learning and AI environments and play with them on state-of-the-art computing facilities provided by the Pittsburgh Supercomputing Center," Paulini said.
Overall, the organizers said the event highlighted the ongoing promise of the university's NSF AI Planning Institute for Data-Driven Discovery in Physics, which was founded last year with a grant from the National Science Foundation.
"Our NSF AI Planning Institute is an incredible opportunity to connect cutting edge machine learning methods and AI with a broad range of physics areas such as particle physics and cosmology," Paulini said.