CAREER: Detecting Extremely Rare Physics Events in the Era of AI for Science

Why is the universe dominated by matter, while antimatter is virtually absent? Why does visible matter constitute only 15% of all matter, while the vast 85% majority—Dark Matter—has not been detected by any experiment on Earth as of today? Answering these questions requires detecting extremely rare physics events, a challenge that has driven physics for over four decades. This CAREER project aims to transform rare event search by developing Artificial Intelligence (AI) algorithms for two world-leading experiments: KamLAND-Zen and XLZD. The PI’s team will develop Large Language Model (LLM) powered AI agents to aid KamLAND-Zen in deciphering the secret of matter-antimatter asymmetry, while creating AI-accelerated simulations to optimize the design of the XLZD Dark Matter detector. To cultivate an interdisciplinary AI workforce to go beyond disciplinary silos, this project aims to train “Data Physicists” by teaching AI to physics students and sparking the interest of data science students in physics research. This will be achieved through the Rare AI for Science Ecosystem (RAISE), a student-led platform that features popular science stories and LLM-assisted AI tutorials. Technically, this project aims to leverage surrogate models and LLM to secure the scientific success of world-leading rare event search experiments over the next decade. In the near term, as the KamLAND-Zen experiment concludes its data taking, the PI proposes to integrate KamNet, a spatiotemporal neural network, into the final analysis to deliver world-leading limits on neutrinoless double-beta decay, a process that could explain the universe's matter-antimatter asymmetry. In the medium term, to support the construction of the next-generation KamLAND2-Zen detector, the PI’s team will design, test, and deploy a hardware-accelerated dual-agent AI system. This architecture leverages real-time AI for sub-millisecond event reconstruction at the detector front-end, while using LLM agents at the back-end to autonomously operate the detector and identify anomalies that eluded human’s attention. In the long term, as the ultimate Generation-3 dark matter experiment XLZD begins its simulation and design campaign, the PI proposes to develop next-generation Rare Event Surrogate Models to systematically optimize the detector's design for maximum sensitivity and discovery potential. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria. NSF Award ID: 2542832 | Program: 01002627DB NSF RESEARCH & RELATED ACTIVIT,01003031DB NSF RESEARCH & RELATED ACTIVIT,01002930DB NSF RESEARCH & RELATED ACTIVIT | Principal Investigator: Aobo Li | Institution: University of California-San Diego, LA JOLLA, CA | Award Amount: $400,179 View on NSF Award Search: https://www.nsf.gov/awardsearch/show-award/?AWD_ID=2542832 View on Research.gov: https://www.research.gov/awardapi-service/v1/awards/2542832.html