Early Career Award recipient Aleksandra Ćiprijanović aims to create universal AI analysis framework

Aleksandra Ćiprijanović, a Wilson Fellow and associate scientist at the U.S. Department of Energy’s Fermi National Accelerator Laboratory, is a recipient of a 2025 DOE Early Career Award. Her project, “Bridging the Gap Between Scientific Datasets with Artificial Intelligence,” was selected for funding by the Office of High Energy Physics as part of the new Computational Research in High Energy Physics program.

Ćiprijanović’s project aims to address a problem that she encountered almost a decade ago when she first started dabbling with AI in her research: the domain shift problem. The problem occurs when a machine learning model is trained on one dataset but is tested or deployed on data that comes from somewhere else; since the model only learned patterns from the training data, its performance often drops when the data changes.

While she plans to start with cosmological data — since that is her scientific home — Ćiprijanović also intends to test her project on collider and neutrino physics.

“I really do want to make a software framework that will be used across different high-energy physics frontiers,” she said.

The framework will have a modular structure and a number of data, model and task options to enable broad scientific use and solutions to any high-energy physics domain-shift problem.

Fermilab is uniquely positioned to support this project, said Ćiprijanović. The lab’s science and computing capabilities will enable the creation of a universal AI analysis framework that will work with a vast range of high-energy physics research needs and coding standards.

“We will need to find people to test out the code and give us inputs from cosmic and from neutrinos and from the collider world,” said Ćiprijanović. “Luckily, at Fermilab, we have experts from all these frontiers! Fermilab is the place to do this.”

Since 2010, the highly competitive DOE Office of Science Early Career Research Program has distributed funding annually to support outstanding early career scientists at universities, national laboratories and Office of Science user facilities.

What if Olympic officials could record sprinters’ times only to the nearest minute? “We would know who started the race, and who finished the race, but that’s it,” said Bryan Cardwell, a postdoctoral researcher at the University of Virginia. “There’s no way to know who arrived first and who arrived last.”

Cardwell and his colleagues on the CMS experiment are currently tackling a similar problem. The CMS experiment records the tracks and properties of subatomic particles created by the Large Hadron Collider, the world’s most powerful particle accelerator. As it stands, physicists get a picture of all the particles produced in a collision, but they have insufficiently detailed information about when the particles were produced or how fast they were traveling, making it difficult to tell them apart.

That’s why CMS scientists are building a new detector that will let them watch particle collisions as they unfold, with 30-picosecond (0.00000000003 second) accuracy.

“In 30 picoseconds, light moves about one centimeter.”

“In 30 picoseconds, light moves about one centimeter,” said Chris Neu, a professor at the University of Virginia. “We’re talking about measuring the time of arrival of very fast objects with a very high precision.”

Some scientists suspect that rare and massive particles that take a long time to decay could move more slowly than lighter, more common particles. This means that when they do finally decay, their “daughter particles” will be a step behind. The difference would be tiny; a photo finish after a race that is only four feet long. But scientists hope that a precise “stopwatch” detector will help them tell the difference.

“We’ll be able to see if any particles are habitually arriving late,” Cardwell said. “That is inherently interesting.”

Turning ripples into tsunamis

The new timing detector consists of two parts: a barrel equipped with roughly 10,000 crystal sensors, and end caps coated with hair-thin silicon wafers. While silicon sensors are common inside CMS, this new detector has a special gain layer that amplifies the signal before it is read out.

“A small signal is like a tiny ripple on the surface of the water, and it gets lost among the little waves around it, so it’s hard to tell exactly when it arrived,” said Artur Apresyan, a researcher at the U.S. Department of Energy’s Fermi National Accelerator Laboratory. “The gain layer turns the small signal into a fast-moving tsunami that stands out clearly from the background, making its arrival time very clear. This amplification feature is not present in other CMS detectors; it’s a new design specifically for timing detectors.”

Currently, scientists at Fermilab are building and testing the sensors and support structures in collaboration with their national and international partners.

“The work on the end cap timing detector at Fermilab has been a unique possibility for many junior researchers to participate in the design and construction of a novel kind of detector,” Apresyan said. “These kinds of opportunities are very rare in such large and advanced experiments, and Fermilab provides an exceptional opportunity to allow newcomers to join and learn cutting-edge detector technologies and their applications.”

Unveiling the dark sector

Once the finished detector is installed inside CMS, scientists hope that it will help them separate fast-moving particles from slow-moving dark matter, a mysterious class of particles that are only visible through their gravitational pull.

Dark matter is as good as discovered in the sense that we know it’s there.”

“Dark matter is as good as discovered in the sense that we know it’s there,” said Tevong You, a theorist at King’s College London. “The problem is that none of the Standard Model particles that we know of can account for the properties of dark matter.”

According to You, many theoretical models predict that — in addition to the stable dark matter particles — there could be an entire universe of dark sector particles that might have properties similar to the known particles of the Standard Model.

“Dark matter is very stable, but we also know that stable particles like protons are just a small subset of all visible particles,” You said. “Maybe dark matter is just one small part of an extended dark sector that is full of long-lived particles that can decay into visible matter.”

According to Cardwell, the LHC could be producing dark matter. These dark matter particles would eat up a lot of kinetic energy from the proton-proton collisions and turn it into mass via Einstein’s equation, E = mc². “This means they have less energy to go fast,” Cardwell said.

When these dark matter particles decay into visible matter, the resulting daughter particles would be lagging behind everything else produced during the collision. This is where the timing detector comes into play.

“If we can measure the time those particles are arriving, we can figure out if they came from a particle that moved a little slower before decaying,” Cardwell said.

Pinpointing exceptional events

The new timing detector will also give scientists a much clearer view of what exactly is happening when bunches of protons collide inside the LHC.

“Right now in CMS, we send something like 100 billion protons through 100 billion protons 40 million times a second,” Cardwell said.

Every time two bunches of protons cross, about 70 collide. Currently, scientists use the spatial coordinates of the smaller particles produced by the collisions to connect the dots and figure out which particles came from what collisions. But spatial orientation alone won’t be enough when scientists turn on the High-Luminosity LHC, which will increase the collision rate by up to a factor of five.

“When there are so many collisions, many of them will literally be at the same point in space,” Cardwell said. “But the collisions don’t all happen at the same time; they’re actually spread out over around 200 picoseconds. With a really, really precise timing detector, 200 picoseconds can suddenly become a very long time; I can slice that 200 picoseconds into a bunch of individual frames.”

These frames will allow scientists to disentangle the messy LHC collisions and pinpoint rare and exceptional events.

“The cool thing about this detector is that it will make everything CMS does better,” Cardwell said. “If we are searching for dark matter, if we are measuring the properties of the Higgs boson, if we’re doing anything at all, the quality of every measurement will increase.”

As a theorist, You is excited about this new timing detector because it will open new avenues for research.

“The best bet for a spectacular discovery at the LHC is through developing new types of searches and new ways of sifting through the data,” You said.

The United States is entering a new era of scientific discovery defined by world-class computing power, rapid advances in artificial intelligence and seamless integration of scientific data. At the center of this transformation is the Genesis Mission, a national effort to double the productivity and overall impact of American science within the next decade by leveraging the combined strengths of the U.S. Department of Energy’s 17 national laboratories. Fermi National Accelerator Laboratory is poised to play a critical role.

With expertise in high-energy physics, advanced computing, accelerator and quantum technologies, and microelectronics design, Fermilab is well-equipped to be a central contributor to the Genesis Mission’s goals. Deploying AI to make particle accelerators adaptive and autonomous will propel innovations in medicine, materials and energy. Simultaneously, advancing next-generation microelectronics will serve to secure U.S. technological leadership, economic prosperity and national security.

“The Genesis Mission represents a once-in-a-generation opportunity to transform how America does science,” Fermilab Director Norbert Holtkamp said. “By combining AI, advanced computing and the capabilities of the national laboratories, we can accelerate discovery while strengthening the scientific infrastructure that underpins U.S. leadership in particle physics and beyond. Fermilab is proud to help drive this national effort.”

“The Genesis Mission represents a once-in-a-generation opportunity to transform how America does science.”

Through these initiatives and more, Fermilab will help tackle ambitious, world-shaping challenges — and it comes to this work prepared. The lab brings decades of experience navigating complex challenges through large-scale collaborations with national laboratory partners, academia and industry.

“We’re also excited for what AI can mean for our operations at Fermilab,” Holtkamp added. “I believe AI can dramatically enhance the quality and efficiency for work across the laboratory. Not only will Fermilab help drive DOE’s Genesis Mission, but I believe Fermilab can also serve as a strong example in how AI can benefit the operations of the full U.S. national lab complex.” 

Advancing accelerator science

Fermilab will contribute to a seven-lab collaboration known as the Multi-Office particle Accelerator Team, or MOAT. The initiative focuses on using AI to advance accelerator science by supporting more efficient operations and streamlined accelerator research and development. These efforts are expected to shorten commissioning timelines and reduce operating costs while extending the benefits of accelerator technology to a wider range of industry partners.

Speeding up chip design

One of Fermilab’s flagship contributions is AXESS, shorthand for Accelerating eXtreme Environment Specs-to-Silicon. This ambitious project aims to revolutionize custom microelectronics chip design for specialized scientific applications. Fermilab already has deep experience in creating chips that can function in extreme environments, such as cryogenic temperatures or intense radiation. By using AI to compress the design process from months to minutes, AXESS will significantly boost national competitiveness and speed up the pace of innovation.

Supporting American Science Cloud

The lab is also providing the Fermi Data Platform to support the emerging American Science Cloud. Fermilab is leveraging its extensive experience in handling the massive amounts of data created by particle collisions to provide large-scale data storage to the Genesis Mission. This infrastructure will enable researchers from across scientific disciplines to access the datasets needed to accelerate their own discoveries, effectively turning decades of physics data into a national resource.

Quickening theoretical calculations

Fermilab’s work in Lattice Quantum Chromodynamics, or Lattice QCD, will strengthen a multi-lab effort called FemtoMind. Lattice QCD is the framework physicists use to understand the strong force that binds subatomic quarks together to form protons. By applying agentic AI to quicken these complex calculations and uncover new patterns in computer simulations that probe physics at the proton scale, FemtoMind will deepen our understanding of the matter that makes up everything we see around us.

Creating AI-ready data representations

Fermilab will also contribute to the TREASURE initiative, which aims to make vast datasets more accessible for groundbreaking research. This program standardizes data from both current and retired particle colliders, including the Large Hadron Collider at CERN and the Tevatron at Fermilab, to create AI-ready representations of physics data for cross-experiment analysis. TREASURE will also convert research papers, datasets and code into user-friendly forms that can be trusted by scientists as they search for physics beyond the Standard Model.

Applying AI to astronomy

Fermilab’s contributions to the Genesis Mission will extend to astrophysics through the AI Universe effort. Researchers plan to apply advanced uncertainty-quantification methods to a state-of-the-art AI model trained on more than 200 million astronomical observations. This work provides astrophysicists with the tools necessary to make more precise predictions and identify meaningful patterns within large-scale astronomy datasets.

Connecting expertise

“AI is most powerful when it connects expertise across domains and institutions,” said Nhan Tran, head of AI coordination at Fermilab. “Through the Genesis Mission, we can integrate data, tools and expertise from all 17 national labs, creating a unified AI ecosystem that accelerates breakthroughs far beyond what any single lab could achieve.”

As the Genesis Mission mobilizes the full power of the Department of Energy’s national laboratories, Fermilab is uniquely positioned to make vital contributions toward solving the nation’s most pressing scientific challenges. From supercharging microelectronics design and enhancing particle accelerator R&D to transforming data analysis, the lab is helping drive the next wave of American innovation and discovery science.