The Tracking Machine Learning Challenge: Throughput Phase

Abstract This paper reports on the second “Throughput” phase of Tracking Machine Learning (TrackML) challenge Codalab platform. As in first “Accuracy” phase, participants had to solve a difficult experimental problem linked tracking accurately trajectory particles as e.g. created at Large Hadron Collider (LHC): given $$O(10^5)$$ O ( 10 5 ) points, connect them into $$O(10^4)$$ 4 individual groups that represent particle trajectories which are approximated helical. While only accuracy mattered, goal this was compromise between and speed inference. Both were measured platform where upload their software. The best three solutions with good an order magnitude faster than state art when designed. Although core algorithms less diverse diversity techniques have been used described paper. performance is analysed depth lessons derived.