Hashing and metric learning for charged particle tracking

TL;DR

This paper introduces a new particle tracking method using hashing and metric learning to efficiently reconstruct charged particle trajectories in high-density collider environments, significantly improving speed and accuracy.

Contribution

It presents a novel combination of hashing and metric learning techniques for particle tracking, addressing limitations of traditional combinatorial methods at high collision rates.

Findings

Achieves 96% tracking efficiency on simulated data

Reduces fake rate to 8%

Demonstrates significant speed improvements

Abstract

We propose a novel approach to charged particle tracking at high intensity particle colliders based on Approximate Nearest Neighbors search. With hundreds of thousands of measurements per collision to be reconstructed e.g. at the High Luminosity Large Hadron Collider, the currently employed combinatorial track finding approaches become inadequate. Here, we use hashing techniques to separate measurements into buckets of 20-50 hits and increase their purity using metric learning. Two different approaches are studied to further resolve tracks inside buckets: Local Fisher Discriminant Analysis and Neural Networks for triplet similarity learning. We demonstrate the proposed approach on simulated collisions and show significant speed improvement with bucket tracking efficiency of 96% and a fake rate of 8% on unseen particle events.