Neutrino scientists at the U.S. Department of Energy’s Fermi National Accelerator Laboratory have been working at the cutting edge of AI and machine learning integration for decades. Two of the lab’s neutrino experiments, NOvA and MicroBooNE, were among the first high-energy physics experiments to use a deep neural net — a type of machine learning that’s intended to mirror the capabilities of the human brain — to identify particle interactions.

“With DUNE, researchers are taking innovative approaches to try and impact every aspect of the experiment,” said Sowjanya Gollapinni, co-spokesperson for DUNE. “These approaches are going to really accelerate DUNE in terms of the commissioning process and the sensitivity of the experiment to reach our full discovery potential.”

Tracking particle interactions

Sometimes called the ghost particles of the universe, neutrinos are incredibly elusive because they only interact through the weak force and gravity. Trillions are streaming through your body every second, but only one may interact in your entire lifetime. That’s why scientists use intense sources — like Fermilab’s accelerator complex — and massive detectors to increase the odds of capturing them. 

DUNE at LBNF will consist of a near detector located at Fermilab and a far detector installed a mile underground at the Sanford Underground Research Facility in South Dakota, creating the experiment’s long baseline. Both detectors will use liquid-argon time projection chamber technology to record the rare interactions that neutrinos will have with the argon atoms in the detector.

While liquid-argon TPCs are fantastic at capturing neutrino interactions, reconstructing the signals can get pretty tricky. This is especially true for DUNE, which will enable scientists to study the properties of neutrinos with unprecedented precision.

“We are going to have very high resolution, almost photographic-quality images of these interactions,” said Leigh Whitehead, DUNE AI/ML co-lead. “Which is great, but also brings challenges such as making reconstruction more difficult because we see such fine detail.”

The AI algorithms used in DUNE will be able to recognize that a neutrino interaction took place inside the detector. They can then analyze the resulting particle tracks visible within the detector, trace back to where the interaction occurred, and extrapolate the energy and direction of the neutrino at the time it entered the detector.

“We’ve had great success in implementing these tools in previous Fermilab experiments,” said Jianming Bian, DUNE AI/ML co-lead. “And now DUNE is taking the leading role to further develop these algorithms and explore new AI methods that take advantage of its complex, large-scale data, making it a platform for AI/ML development.”

These AI-driven tools are orders of magnitude faster than traditional methods, leading to enhanced signal processing, event classification and particle identification. This will enable the collaboration to tackle their main science goals more efficiently and ultimately unlock the secrets of neutrinos.

Locating exploding stars

While DUNE researchers are using AI to reconstruct the more common neutrino signals within the detector, they are also training algorithms to identify rare phenomena. Referred to as triggers, these are fast decision-making systems that are trained to sort through thousands of interactions and save only the most interesting for physics.

One of the major triggers for DUNE would occur if a supernova goes off in our galaxy.

When a large enough star runs out of fuel and “dies,” it rips itself apart in an explosion that sends stellar remains flying into space. Particles like photons get trapped in the resulting gas and dust, neutrinos — which rarely interact — can breeze right through. This means that the neutrinos will arrive to Earth a few hours before the light burst, offering scientists a rare glimpse into the processes unfolding deep inside an exploding star.

The early alert from the neutrino burst also enables researchers to quickly identify where the supernova occurred so that astrophysicists know where to point their telescopes in time to see light from the event.

That’s where the trigger comes in. “An AI algorithm will be constantly reading the detector, searching for interesting interactions,” said Thomas Junk, senior scientist at Fermilab. The instant the algorithm suspects a signal is from a supernova, “it will tell the system to record data from 10 seconds before and 100 seconds after the candidate signal to capture the full picture of neutrino interactions from this event.”

Aside from acting as an early warning system for astronomers to observe a supernova in real time, neutrinos may also provide crucial insight into the stellar objects left behind, either a neutron star or black hole.

Running massive detectors

Detecting neutrinos isn’t the only complicated part of DUNE, the detectors are comprised of thousands of components that need to be monitored to ensure that they’re working in concert as designed. 

After the detector is installed and commissioned, operators must actively monitor the system during data collection. If there is an error, it is critical that the team can respond quickly and knowledgably to any situation.

To better empower operators to tackle more unique scenarios, the DUNE collaboration is looking into whether a large language model can be used to quickly scan logbooks detailing successful documented fixes to provide immediate citations to solutions. Fermilab is coordinating with six other national labs to develop a similar tool for particle accelerators.

The DUNE collaboration is also looking into how machine learning can be used to automate detector operation a step further to potentially predict detector anomalies before they happen.

Setting the stage

DUNE has been actively investigating AI integration for years. But now the Department of Energy’s Genesis Mission is compounding that effort to accelerate scientific discovery.

“The Genesis Mission has strengthened the already existing collaboration across the national labs and universities on AI efforts,” said Gollapinni. “It is taking everything to the next level and will maximize our capabilities to accelerate DUNE.”

Integrating AI into the detector itself isn’t the only approach that DUNE is taking when it comes to AI. The collaboration is actively working with the DOE’s national labs and partnering institutions to enhance their existing AI infrastructure and prepare for the workflows necessary to process the petabytes (quadrillions of bytes) of data DUNE will produce.

DUNE is also involved in training the next-generation AI workforce.

“We have over a thousand people from around the globe collaborating on DUNE,” said Whitehead. “We’re training hundreds of students a year, which is a pretty impressive pipeline of people that should have all the tools to become leaders in the field.”

An international collaboration hosted by Fermilab in the United States, DUNE is designed to study neutrinos: ghostly particles that are incredibly abundant and incredibly mysterious, due to their shape-shifting behavior and reluctance to interact. Many scientists hope that neutrinos could help explain how matter won out over its equal but opposite counterpart antimatter in the early universe.

At CERN’s neutrino platform in Europe, there are two 770-ton-scale prototype neutrino detectors called ProtoDUNEs. Each is playing a pivotal role in demonstrating the technologies planned for DUNE.

“The work at ProtoDUNE going into demonstrating the technology is heroic,” said Sowjanya Gollapinni, DUNE co-spokesperson and a senior scientist at Los Alamos National Laboratory. “Now we want to stress test it and see how it holds up over a long period of time.”

Beyond the neutrinos generated by Fermilab’s new particle accelerator, DUNE will also detect cosmic neutrinos — particles from space that pass effortlessly through the Earth. These cosmic signals will allow scientists to identify the earliest signs of a supernova. Because neutrinos escape a collapsing star before visible light does, they provide an early warning that gives astronomers time to ready their telescopes before the explosion’s glow reaches Earth.

“Neutrinos are key in our understanding of how the universe evolved to what it is today, and where it is headed,” Gollapinni said.

To capture these neutrinos, scientists are installing massive detectors, called liquid-argon time projection chambers, a mile underground at the Sanford Underground Research Facility in South Dakota.

“When a neutrino interacts with an argon atom, it generates a number of charged particles, and these charged particles leave tracks of ionization,” said Flavio Cavanna, a scientist at Fermilab.

The liberated electrons are then pulled by a strong electric field toward specialized components called readout planes inside the chamber. These readout planes record data from the ionized electrons, which scientists can use to reconstruct the direction and energy of the tracks. From this information, scientists can determine the position, energy and identity of the original neutrino.

Before building DUNE, scientists decided to test two different technologies at the CERN Neutrino Platform. One technique is based on a tried-and-true detector called an Anode Plane Assembly, or APA. This was developed in the 1980s and uses planes of loom-like wire detectors to collect current from the drifting electrons.

“We’ve already used this technique in other neutrino experiments,” Cavanna said. “We know it works.”

The other is a much newer technology that replaces the planes of wires with channels of copper printed onto circuit boards.

“If you ever open a computer or a keyboard, you will see a printed circuit board,” said Steve Kettell, a scientist at Brookhaven National Laboratory and one of DUNE’s technical coordinators. “It’s challenging to wrap wires 3,000 times around a frame. Printed circuit boards are available commercially, so it’s more efficient for the collaboration.”

Scientists successfully tested the APA technique at ProtoDUNE between 2018 and 2024. During this time, physicists and engineers were also testing, redesigning and perfecting the new printed circuit board design, which they nicknamed the “vertical drift,” based on the direction the charged particles move in the liquid argon.

In addition to replacing the APAs with printed circuit boards, the new vertical drift geometry doubles the distance the liberated electrons move, thus allowing scientists to capture and record more neutrinos with fewer components. But this also means that they need a much higher voltage to maintain the electric field that pushes the liberated electrons to the detectors before they disappear.

“A battery that you can hold in your hand has one and a half volts between the two ends,” Kettell said. “We are taking that same concept and scaling it up to 300,000 volts.”

Last year, the collaboration started commissioning the new design of the ProtoDUNE Vertical Drift, and by June 2025 the detector was ready to be launched.

“It ran smoothly out of the box,” Gollapinni said. “The ProtoDUNE Vertical Drift has been a resounding success, and we are very thankful for the incredible support provided by the CERN neutrino platform.”

Now that they know it works, the team is ramping the voltage up to 300 kilovolts and seeing how long the detector can hold it.

“It’s like building a lightning storm inside of a detector, but we don’t actually want the physical lightning strikes,” Kettell said. “If we see sparks, it will allow us to study how all of the various components react.”

The stress test started on May 22, 2026, and the scientists plan to have it completed by the fall. According to Kettell, this extended stress test simulates the conditions scientists will ultimately use inside DUNE, but inside a much smaller prototype. “If all the components survive, that’s very encouraging,” Kettell said.

“Fermilab is honored to receive this support for advancing artificial intelligence as part of DOE’s Genesis Mission. This investment strengthens our ability to accelerate scientific discovery, develop next‑generation technologies, and drive innovation at the frontiers of particle physics and advanced technology,” Fermilab Director Norbert Holtkamp said. “We are proud to contribute our expertise to a mission that will help shape the next era of U.S. leadership in science and technology.”

The Fermilab-led project will develop AI/ML-based resonance control algorithms to enable high-reliability, low-cost accelerator operations that will substantially improve the performance of particle accelerators that drive discovery science. Several partner institutions, including national labs, universities and industry partner xLight Inc., are contributing to this project.

The Genesis Mission is a historic national initiative led by the U.S. Department of Energy, which is building the world’s most powerful integrated science discovery platform. By uniting government, industry, academia and philanthropy, it is accelerating breakthroughs in energy, scientific discovery and national security through a new platform that combines AI, supercomputing, quantum systems and advanced scientific instruments.

The projects named today will receive funding through a Request For Applications (RFA) that was announced by DOE in March. The goal of the Phase I RFA awards is to identify promising pathways toward transformative scientific capabilities and establish a foundation for future investment and scale. Fermilab project teams will design and demonstrate research workflows that integrate AI with scientific investigation, while rigorously evaluating whether those approaches can accelerate discovery, improve predictive capabilities, enhance experimentation, or generate new scientific insights.

Fermilab participated in today’s Genesis Mission Summit as a recipient of a first‑round Genesis Mission award, underscoring its leadership in advanced accelerator technologies as the Department of Energy announced the selected projects.

The AI/ML algorithms that will emerge from the Fermilab-led project will increase science reach, improve stability for users, increase radio-frequency amplifier lifetime and reduce operating costs. Superconducting radio-frequency (SRF) cavities are electromagnetic resonators that transfer energy to beams in particle accelerators. Because they are extremely efficient resonators, they can be disturbed by small vibrations and pressure fluctuations. For current and future SRF-based accelerators like Fermilab’s Proton Improvement Plan-II (PIP-II) linear accelerator, SLAC National Laboratory’s Linac Coherent Light Source-SC, Brookhaven National Laboratory’s Electron-Ion Collider (EIC), Michigan State University’s Facility for Rare Isotope Beams, Argonne National Laboratory’s Argonne Tandem Linac Accelerator System, and industrial accelerators, it is vitally important to have precise control of the cavity resonant frequency.

“These awards recognize the tremendous opportunity to combine artificial intelligence with Fermilab’s world-leading expertise in superconducting accelerator technology,” said Anna Grassellino, chief technology officer and associate laboratory director for the Technology Directorate at Fermilab. “By developing AI-driven control of superconducting RF cavities, we can make future accelerators — such as our own PIP-II — more efficient, reliable, and autonomous, enabling higher scientific performance while reducing operational complexity. This is an important step toward a new generation of intelligent accelerator facilities that will power discoveries across particle physics and many other fields of science.” 

Fermilab will contribute as a key collaborator on several other projects aligned with core research areas of the lab such as neutrino and collider physics, the muon-to-electron-conversion experiment (Mu2e), precision frontier, the Electron-Ion Collider and advanced scientific computing. The following projects were selected for funding through the Genesis Mission:

• How AI will expedite discovery in highly complex electron-ion collider data streams
  Lead institution: Purdue University
• Using an AI-driven approach to detect anomalies in the CMS Level-1 Scouting System
  Lead institution: University of Colorado
• Standardizing and advancing high-performance computing workloads using AI
  Lead institution: University of Wisconsin-Madison
• Using agentic workflows for the expedited search and discovery of charged lepton flavor violation in Mu2e
  Lead institution: University of South Carolina
• Deployment of advanced AI accelerated systems to detect, identify, classify and communicate supernova events in real time for the Deep Underground Neutrino Experiment (DUNE)
  Lead institution: Duke University
• The use of AI agents for high-energy physics simulations and analysis operations
  Lead institution: University of Alabama
• Accelerating DUNE physics with AI to discover neutrino interaction uncertainties
  Lead institution: Florida State University
• Physics-driven digital twins simulations for fusion magnet systems
  Lead institution: Lawrence Berkeley National Laboratory

Fermilab is an international leader in particle accelerator science research, which generates massive streams of data for scientific discovery using accelerator and detector technologies. The lab’s high-energy physics research creates unique opportunities for AI innovations. That is why researchers are developing and implementing AI tools to improve the precision of measurements, optimizing operations and accelerating discovery.