Carnegie Mellon Physicist Uses Machine Learning to Advance Dark Matter Search at CERN
Abhirami Harilal spent four years at CERN developing machine learning models to detect rare particle signatures, work that has sharpened the Large Hadron Collider's ability to hunt for undiscovered particles that could explain dark matter.
Harilal, a Carnegie Mellon Ph.D. graduate in physics, deployed algorithms inside the Compact Muon Solenoid (CMS) experiment to autonomously identify anomalies in detector data. Her work addressed a practical problem: CERN's original data monitoring system could flag unusual activity but often missed subtle or changing anomalies, forcing researchers to manually compare flagged data against reliable datasets.
She built a machine learning model that automated this process entirely. The system now alerts researchers to unusual activity and explains how the data differs from expected results. "I'm particularly proud about it because this was actually deployed and used during live data taking," Harilal said.
Searching for Hidden Particles
Only about 5 percent of the universe consists of observable matter. The rest is dark matter and dark energy-forces that affect galaxies and stars but remain invisible to traditional detection methods.
Harilal focused on whether the Higgs boson, discovered at CERN in 2012, could decay into an undiscovered particle called an A particle. Such a particle could connect to dark matter or other hidden sectors of physics.
She used computational modeling to simulate thousands of particle collisions similar to those produced at the LHC. These simulations allowed her to train machine learning models that improved the experiment's sensitivity to particles with unusual or hard-to-detect signatures.
Broader Applications
Harilal's core skill-recognizing meaningful patterns in large, noisy datasets-extends beyond particle physics. She plans to apply these methods to medical imaging, financial data analysis, and drug discovery.
"A big part of my work is recognizing meaningful patterns in large amounts of noisy data, which is also relevant in many other applications," Harilal said. "I'm looking forward to finding other opportunities to use these skills."
Her advisor, Manfred Paulini, a professor of physics at Carnegie Mellon, said her combination of particle physics expertise and machine learning development offers a model for solving real-world challenges across disciplines.
Researchers interested in applying machine learning to scientific problems may benefit from exploring AI Research Courses and AI Data Analysis Courses that cover pattern recognition and anomaly detection in complex datasets.
Your membership also unlocks:
