The 2023 FRIB Theory Alliance (FRIB-TA) Summer School on Practical Uncertainty Quantification and Emulator Development in Nuclear Physics brought together more than 60 participants from several U.S. and international institutions, including universities and national laboratories, to learn about the practical applications of Bayesian statistics and machine learning for nuclear science. FRIB hosted the summer school on 26-28 June.
Attendees included undergraduate and graduate students, postdoctoral researchers, staff scientists, and faculty. They encompassed a wide spectrum of personal backgrounds and varied scientific interests, including nuclear theory and experiment, close physics fields such as astrophysics, engineering, applied math, statistics, and computer science.
The school’s organizing team has expertise in applications to nuclear science of Bayesian statistics methods, machine learning, dimensionality reduction, and experimental design and control. The team consisted of four faculty members: Kyle Godbey, research assistant professor at FRIB; Morten Hjorth-Jensen, professor of physics at FRIB and in MSU’s Department of Physics and Astronomy; Fernando Montes, senior research physicist at FRIB; and Frederi Viens, professor of statistics at Rice University; two postdoctoral researchers: Edgard Bonilla, postdoctoral researcher at Stanford University’s Laser Interferometer Gravitational-wave Observatory; and Pablo Giuliani, research associate at FRIB; and two graduate students: Rahul Jain (FRIB) and Alexandra Semposki (Ohio University).
Advanced scientific computing and principled statistical methods have become crucial aspects of the theory-experiment cycle in nuclear science. This is particularly true at laboratories like FRIB, with broad research interests spanning from fundamental symmetries through the role and origin of nuclei in the cosmos. Attendees learned how uncertainty quantification can be integrated into nuclear physics at every level, from theoretical modeling to experimental analysis and control. Machine learning and emulation methods were presented as instrumental tools to overcome the computational challenges associated with quantifying uncertainties, as well as to maximize the information gained from experimental and theoretical data.
The program’s core philosophy focuses on practical implementation of the discussed tools and ideas. The lectures for each session were accompanied by hands-on tutorials that provided a guided implementation of the topic. In order to accommodate a wide spectrum of coding expertise among the attendees, cloud-computing environments were prepared for the various challenges developed for the course. Attendees were encouraged to form teams to tackle a short project exploring the school’s topics in more detail, all of which will be edited into a single online compendium.
The Advanced Scientific Computing and Statistics Network provided a forum for the attendees to introduce themselves, communicate throughout and after the summer school with questions and discussions, and to coordinate efforts for the final project. A GitHub repository was also created to host the lecture slides and recordings, as well as the hands-on code activities.
The FRIB Theory Alliance (FRIB-TA) is a coalition of scientists from universities and national laboratories who seek to foster advancements in theory related to diverse areas of FRIB science; optimize the coupling between theory and experiment; and stimulate the field by creating permanent theory positions across the country, attracting young talent through the national FRIB Theory Fellow Program, fostering interdisciplinary collaborations, and shepherding international initiatives. Learn more at the FRIB-TA website.