If there were no strong nuclear force binding atomic particles together, matter as we know it would not exist. However, there are still several unresolved questions in the study of this fundamental interaction.

William & Mary’s Cristiano Fanelli, assistant professor of data science, is harnessing the power of machine learning and artificial intelligence to advance our understanding of the strong force. 

Assistant Professor Cristiano Fanelli, headshot
Assistant Professor Cristiano Fanelli. (Courtesy photo)

Two of his recent projects have been selected for funding by the U.S. Department of Energy, which has allocated $16 million for 15 projects selected by competitive peer review to AI/ML research for nuclear physics accelerators and detectors.

In announcing the awards, the Department of Energy declared that “artificial intelligence has the potential to shorten the timeline for experimental discovery in nuclear physics.” This federal perspective resonates with a William & Mary strategic priority: integrating computational thinking across disciplines as stated by the university’s Vision 2026 data initiative.

As a principal investigator, Fanelli will lead “A scalable and distributed AI-assisted detector design for the EIC.” This project will be assisting the design of the ePIC detector at the future  Electron-Ion Collider — a $2 billion state-of-the-art machine which will begin operations early in the next decade and has been identified as the highest-priority new facility in the United States for the field of nuclear physics. The EIC is expected to push the frontier of physics, develop new technologies and knowledge and accelerate advances in areas such as nuclear medicine and national security.

Fanelli is also a co-investigator on a “AI/ML Optimized Polarization,” a project led by the Thomas Jefferson National Accelerator Facility (Jefferson Lab). This project will take place in the GlueX experiment, whose primary purpose is to better understand the nature of confinement of particles such as quarks and gluons, which cannot be isolated but clump together to form hadrons. William & Mary has a strong partnership with Jefferson Lab, with faculty and students often involved in projects there.

The strong force governs the behavior of quarks and gluons inside hadrons like protons and neutrons: Fanelli’s projects will contribute to exploring the internal structure and dynamics of such particles in unprecedented detail.

“Our understanding remains incomplete regarding the intricate processes occurring within the protons and neutrons that form the atomic nucleus,” said Fanelli. “We want to unlock the mysteries of how the strong interaction really works inside hadrons.”

The AIDE project

A schematic of the planned Electron-Ion Collider.
A schematic of the planned EIC. (Image by Brookhaven National Laboratory, licensed under CC BY-NC-ND 2.0.)

The EIC under construction at Brookhaven Lab in Upton, New York state, will be made up of two intersecting accelerators colliding polarized electrons and either protons or ions at extremely high speeds. 

The electron beam will expose the complex arrangement of the quarks and gluons within protons and neutrons. The binding force between quarks, mediated by gluons, represents the strongest force in nature: These interactions will yield high-resolution depictions of their internal structures.

Particle detectors identify the particles produced in collisions by measuring their properties, such as speed, mass and charge. With an estimate cost of about $300 million, the ePIC detector will be constructed at the EIC as part of a collaboration involving over 170 institutions from around the world: Fanelli’s project will develop an AI-assisted framework helping optimize the design of this detector.

“This project will make ePIC the first large-scale experiment of its kind to be optimized with the assistance of artificial intelligence,” said Fanelli, whose co-investigators include researchers from two national laboratories, Brookhaven and Jefferson Lab, and two higher education institutions, Catholic University of America and Duke University.

Fanelli explained that hundreds of multidimensional design parameters are at play in designing detectors, which need to be tuned considering multiple competing objectives and are subject to several constraints. Even a small improvement in the objectives leads to a more efficient use of the beam time which will make up much of the cost of the EIC over its lifetime.

“Human intuition and expertise served us well in navigating complex design spaces for many years,” he said. “However, the advent of AI now gives us unprecedented capabilities to explore multidimensional spaces in a way that could lead to optimized solutions, balancing trade-offs and costs, something that enhances our existing methods.”

The AIOP project

At the Jefferson Lab GlueX experiment, a photon beam is shot on fixed nuclear targets to produce final state particles. The study of these reactions allows a better understanding of the nature of confinement in quantum chromodynamics, the theory of the strong interaction between quarks mediated by gluons, with the potential of identifying a spectrum of exotic particles generated by the excitation of the gluonic field binding the quarks.

Fanelli’s proposal develops and implements an AI/ML control strategy to enhance the polarization of the photon beam used by this experiment. Techniques such as deep reinforced learning, he explained, allow to correct the quality of the beam in real time, since such multidimensional problems require decisions based on multiple inputs.

“Enhancing the polarization of the beam not only improves the experiment’s performance but also leads to substantial cost savings, especially considering the inherent expenses associated with delivering the beam,” said Fanelli.

An additional advantage of using machine learning and deep learning in experimental nuclear physics involves the reconstruction of the final state particles generated in these experiments. 

“One challenge is to holistically reconstruct the whole physics event, which can include multiple particles,” said Fanelli. “Machine learning offers the capability to cope with complex event topologies, something difficult to achieve with conventional algorithms.”

As he anticipates the groundbreaking insights that AI and ML will offer, Fanelli remains assured that human expertise will continue to play a vital role. “The benefits will be manifold, ranging from simulations, control, data acquisition, analysis and beyond.”

Machine learning literacy and future opportunities

William & Mary creates opportunities for students to work with artificial intelligence and machine learning, with numerous applications in areas such as nuclear fusion among others.

In October 2022, the second workshop on “Artificial intelligence for the Electron-Ion Collider” was held at William & Mary, involving over 200 participants. Experts from national laboratories, universities and industry discussed current and prospective applications of machine learning for the EIC and relayed insights from the ePIC collaboration. 

The event culminated with a hybrid hackathon in which 10 teams from around the world competed to solve specific problems related to the EIC by providing machine learning-based solutions. William & Mary students figured among the 40 hackathon participants.

“It was a profoundly educational and intellectually stimulating experience,” recalled Fanelli, who is a member of the organizing committee for the AI4EIC third edition, now open for registration, taking place in Washington, D.C., this fall. “It provided valuable insights and fresh perspectives on how to apply machine learning and deep learning in reconstructing particles, particularly when using the state-of-the-art sub-detector technology developed for the EIC.”

In the context of these grants, Fanelli expects to organize other educational events to increase literacy on AI and machine learning techniques, also reaching out to schools and other higher education institutions.

Fanelli brought to his data science role his background in experimental nuclear physics and his years of experience in artificial intelligence and machine learning. He emphasized the role of multidisciplinarity in advancing experimental nuclear physics while providing a strong foundation for roles in industry.

“William & Mary shares this vision, recognizing the widespread applicability and benefits of data science across various domains,” he said.

, Senior Research Writer