Particle-tracking challenge

Federico Ronchetti / CERN Computer image representing the particles generated by colliding lead nucleiFederico Ronchetti / CERN

In early May, the European Organization for Nuclear Research (CERN) launched a computer science challenge to stimulate the development of artificially intelligent data-analysis programs. In their bid to decode the fundamental composition of matter, physicists at the Large Hadron Collider (LHC), the world’s largest particle accelerator, located near Geneva, Switzerland, intend to increase the number of particle collisions at the laboratory by a factor of at least 20 within the next decade. The number of events will eventually exceed the LHC’s current capacity for analysis, with hundreds of millions of collisions occurring every second. Detectors in the accelerator record information from every particle that passes through them, but only records and reconstructs the trajectories of collisions with the most interesting results. These collisions are selected by pattern-recognition algorithms, but the process is slow, computer scientist Cécile Germain told the journal Nature. The idea of developing algorithms that use machine learning resulted from a partnership between CERN and Kaggle Competitions. Hundreds of gigabytes of particle-collision data has been made available to competitors, who can use it to train their algorithms over the next three months to accurately reconstruct particle trajectories. The best algorithms from the first phase of the competition will be announced in Rio de Janeiro in July this year, during the World Congress in Computer Science. The prizes are US$5,000, US$8,000, and US$12,000. The challenge ends in December in Montreal, Canada.