Particle physics gets the machine learning treatment as collider data grows – TechCrunch

The volume of data that particle physicists have to sort through at the Large Hadron Collider is staggering, and it’s about to increase by an order of magnitude. To cope with this torrent of data, CERN is turning to machine learning, offering prizes for AI models that can reduce clutter and help make the next breakthrough.

Inside the massive detectors that dot the LHC rings, protons accelerated to near-lightspeed collide with incredible energy and produce a fountain of exotic, short-lived particles that impact the sensors that line the walls. Improvements to the magnetic fields guiding these protons around the ring mean that where you might have expected twenty or so collisions to pull data, there could soon be hundreds.

Great for scientists, but with more collisions comes more data, in this case an order of magnitude more. Old ways of sorting and categorizing data are too slow, researchers told Nature. So they do what any modern data scientist would do when faced with a massive set of noisy data: hand it over to the AIs.

I use the term loosely, of course – machine learning models are really great at sifting through tons of data looking for what they’ve been trained to find. Medicine, astronomy and, of course, psychology (courtesy of Facebook) have all been advanced by this practical ability to separate the statistical wheat from the chaff. Particle physics is just another example.

A little healthy competition, as always, is a good way to get the peloton going again. So CERN created TrackML, a relatively low-key competition where physicists and data scientists can upload gigs of real collider data and train models to properly classify it. Sure, there are terabytes available for picking, but you’ll want to use that particular dataset.

The prize pool is relatively small, with a total of $25,000 available for winners. Maybe they used all their money to… you know, upgrade the world’s largest particle collider with state-of-the-art cryomagnets.

There is no entry fee, so if you want to try your luck, head over to Kaggle, where the contest is hosted, and upload your own copy of the data. You have plenty of time – submissions are due in August. Oh, and everything you submit will be open source, so don’t worry about them marketing your algorithm.

Comments are closed.