The reconstruction of particle trajectories is a key challenge of particle physics experiments, as it directly impacts particle identification and physics performances. Additionally, it is one of the primary consumers of CPU resources in many high-energy physics experiments. As the luminosity of particle colliders increases, this reconstruction will become more complex and resource-intensive. Therefore, new algorithms are needed to address these challenges efficiently. During track reconstruction, more tracks are reconstructed than the actual truth particles due to fake tracks and redundant duplicates. This inefficiency is costly since each track requires individual reconstruction. We propose using a ranking-based machine learning algorithm to select track seeds before the actual track reconstruction occurs. By employing a DBSCAN clustering algorithm to group similar particle seeds and a Neural Network (NN) with a novel Margin Ranking Loss Function to score them, we can significantly reduce the number of tracks that need to be reconstructed, thereby speeding up the tracking process. This method has been implemented within the A Common Tracking Software framework and tested on the Open Data Detector, resulting in a twofold speedup with an efficiency reduction of only 0.2 percentage points.
Allaire et al. (Tue,) studied this question.