Skip to content

Team 'The_Superpositioned_States_of_America' QHack Hackathon submission

Notifications You must be signed in to change notification settings

Sinestro38/QHack-2022-Hackathon

 
 

Repository files navigation

image

Optimizing Quantum Graph Neural Networks for the Particle Tracking Problem

for QHack 2022

Team:

The Superpositioned States of America

Final Deliverables

  1. Fork of QtrkX with experiments (see various branches)
  2. Final report
  3. PR back to QtrkX

Related: Final submission to QHack

Abstract

The CERN Quantum Technology Initiative declared five quantum computing and algorithms projects of interest to the HEP community. One of these projects is particle tracking reconstruction. Particle tracking reconstruction is a fundamental research aspect for particle collider experiments, which aim to validate the elementary particle theories through the observation of particle trajectories after collisions. With the upgrade of the Large Hadron Collider (LHC) to High Luminosity (HL-LHC) at CERN in the near future (circa 2027), the rate of collisions will be further ramped up, yielding many more detector hits and a larger scale of data. Therefore, being able to efficiently identify the particle trajectories from data of a large scale and increased complexity may be the key to unveil the nature of particle physics. With the development of quantum computers on its own growth curve, untangling the information from big data via quantum computing appears to be a promising solution. Our work focused on improving the existing state-of-the-art of Quantum Graph Neural Networks (QGNNs) to solve the particle tracking reconstruction problem. Specifically, we have have resolved the vanishing/exploding gradient problem during early stages of training, introduced new ansätze for the parameterized quantum circuits (PQCs) used in the Edge/Node networks, and tested these new ansätze on a modified dataset after pre-processing. After experimenting with a variety of different circuit architectures, we have constructed two high performing circuit ansätze: "qc102_pqc (Van Buren)" and "qc103_pqc (Coolidge)". Both circuits, coupled together with an improved QGNN outperform the "qc10_pqc" ansätze used by CERN in the original experiment on a smaller dataset.

Project Files

Most of the files related to the project are located at a fork of QtrkX (look at the various branches). The files in this repository are supporting notebooks and python scripts that were used for the main project as well as the final report.

About

Team 'The_Superpositioned_States_of_America' QHack Hackathon submission

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Languages

  • Jupyter Notebook 91.4%
  • Python 8.6%