On 7 January 2021, BMUCO hosted Dr. Thea Aarrestad, a researcher at CERN, for a discussion on machine learning applications in high-energy particle physics.
The Data Deluge
"With up to one billion proton collisions every second, we must instantly decide which events to save and which to discard."
Dr. Aarrestad's work addresses one of the Large Hadron Collider's most pressing computational challenges: with up to one billion proton collisions occurring every second, physicists must instantly decide which events to save and which to discard.
Real-Time Physics
Machine learning algorithms—particularly deep neural networks compressed to operate in nanoseconds—enable real-time filtering of collision data to identify rare events that could signal new physics beyond the Standard Model.
From Images to Particles
The talk explored how techniques originally developed for:
- Image recognition: Identifying patterns in visual data
- Natural language processing: Understanding sequential information
- Anomaly detection: Finding outliers in massive datasets
...are being adapted to detect the subtle signatures of undiscovered particles in the torrent of LHC data.
Nanosecond Decisions
The challenge isn't just about having powerful algorithms—it's about compressing them to run on specialized hardware that can make decisions in billionths of a second. Dr. Aarrestad explained how neural networks are optimized for extreme speed while maintaining the sensitivity needed to catch glimpses of physics beyond our current theories.
This work sits at the intersection of fundamental science and cutting-edge technology, where the search for nature's deepest secrets demands the most advanced computational tools.
Watch the full talk:
youtube.com/watch?v=DBJSJJKEyXw