hypertrack

https://arxiv.org/abs/2309.14113

HyperTrack: Neural Combinatorics for High Energy Physics

Presented in CHEP 2023
https://indico.jlab.org/event/459/contributions/11748

Mikael Mieskolainen
[email protected]

Overview

HyperTrack is a new hybrid algorithm for deep learned clustering based on a learned graph constructor called Voxel-Dynamics, Graph Neural Networks and Transformers. For more details, see the paper and the conference talk.

This repository together with pre-trained torch models downloaded from Hugging Face can be used to reproduce the paper results on the charged particle track reconstruction problem.

The technical API and instructions at:

https://mieskolainen.github.io/hypertrack

Hugging Face Quick Start

Install the framework, process TrackML dataset files, download the pre-trained models from Hugging Face https://huggingface.co/mieskolainen and follow the documentation for inference.

Citation

If you use this in your work or find ideas interesting, please cite:

@Conference{citekey,
  author    = "Mikael Mieskolainen",
  title     = "HyperTrack: Neural Combinatorics for High Energy Physics",
  booktitle = "CHEP 2023, 26th International Conference on Computing in High Energy & Nuclear Physics",
  year      = "2023"
}