A machine learning challenge for particle tracking

Scientists from the ATLAS, CMS and LHCb collaborations, including Tobias Golling, Moritz Kiehn, and Sabrina Amrouche from the Département de Physique Nucléaire et Corpusculaire, have launched the TrackML challenge – your chance to develop new machine-learning solutions for the next generation of particle detectors and win hardware and cash prizes.

To explore what our universe is made of, scientists at CERN are colliding protons, essentially recreating mini big bangs, and meticulously observing these collisions with intricate silicon detectors. While orchestrating the collisions and observations is already a massive scientific accomplishment, analyzing the tens of petabytes of data produced from the experiments is becoming an overwhelming challenge.

Managing the amount of data will become even more challenging in the near future: the High-Luminosity LHC, a major upgrade planned to begin operation in 2026, will increase the collision rate by up to a factor of five. New detectors, developed also here at DPNC, will produce more data than ever before. Innovative new software solutions will be needed to promptly reconstruct the particle tracks produced by these collisions.

To help address this issue, a team of machine-learning experts and LHC physicists has partnered with Kaggle to probe the question: can machine learning assist high-energy physics in discovering and characterising new particles? Specifically, this competition challenges participants to build an algorithm that quickly and efficiently reconstructs particle tracks from space points measured in the silicon detectors.

The first phase of the challenge, the "Accuracy Phase", is now running on the Kaggle platform until August 2018. Here the focus is on developing novel algorithms that can accurately reconstruct the particle tracks, irrespective of the evaluation time. This phase is also an official IEEE WCCI competition (Rio de Janeiro, July 2018). The second “Throughput Phase” will run later this year with a focus on the throughput (or speed) of the evaluation while maintaining a good accuracy. This phase is an official NIPS competition (Montreal, December 2018). The Université de Genève is a platinum sponsor for the challenge thanks to contributions from the Faculté des Sciences and the Rectorat.

Sign up for the TrackML challenge today. The three top scorers will receive cash prizes. Selected winners may be awarded a top-notch NVIDIA v100 GPU, get the chance to visit CERN, or attend the 2018 Conference on Neural Information Processing Systems in Montreal (Canada).

For more information and the conditions for participation, visit the Kaggle challenge website, read the Nature news article, or follow the official TrackML Twitter account.