The JINR team took the 1st place in the hackathon on the application of machine learning methods in particle identification tasks. The hackathon was held from 10 to 14 October as part of the 2nd AI4EIC-exp Workshop on Artificial Intelligence for the Electron Ion Collider (BNL, USA).

Alexey Aparin, Artem Korobitsin (VBLHEP), Vladimir Papoyan (MLIT), and Grigory Tolkachev, a student of MEPhI and former participant of the UC JINR Summer Practice, represented the Joint Institute at the competition. In total, almost 30 people from 10 teams took part in the hackathon.

The final competition in the classification of charged particles by a signal from the Cherenkov detector (RICH) took place on Friday, 14 October, and lasted eight hours. In the allotted time, participants had to propose an algorithm that could determine the type of charged particles as efficiently as possible by its signal in the detector.

Participants successively solved three subtasks, which increased in difficulty one after another. First, it was necessary to separate charged pions from kaons when the particle pulses and their departure angles were fixed. At the next stage, the pulses and departure angles changed in the specified ranges. In the third question, random detector noise was added. “We obtained the maximum result using the CatBoostClassifier decision tree algorithm, but we tried other options as well,” Alexey Aparin commented on the solution. To perform the hackathon tasks, the JINR team used the resources of the Govorun Supercomputer. Until the end of the week, the task is available to everyone via the link. Some participants continue trying their ideas to achieve the most effective result, even out of competition.

It should be noted that the work on the application of machine learning methods in a physical experiment has been carried out at JINR for several months as part of the RSF grant.