Quantum Technology Initiative Journal Club

Thursday 12 Sept 2024, 16:00 → 17:00 Europe/Zurich

513/R-070 - Openlab Space (CERN)

Michele Grossi (CERN)

Weekly Journal Club meetings organised in the framework of the CERN Quantum Technology Initiative (QTI) to present and discuss scientific papers in the field of quantum science and technology. The goal is to help researchers keep track of current findings and walk away with ideas for their own research. Some previous knowledge of quantum physics would be helpful, but is not required to follow the talks.

To propose a paper for discussion, contact: michele.grossi@cern.ch

16:00 → 17:00 CERN QTI Journal CLUB: TITLE

Convener: Dr Michele Grossi (CERN)

16:00 Thibault Desrousseaux (ETH)

TITLE: Quantum Graph Neural Networks for High-Energy Physics -
Master Thesis outcome presentation

Abstract:
Quantum machine learning is a recent field that explores the potential of applying machine learning methods on quantum computers. These techniques have shown par- ticularly promising results for graph-related problems. The aim of this master thesis is to apply Quantum Graph Neural Networks (QGNNs) to high-energy physics.
First, we applied QGNNs to compute the scattering amplitudes of Feynman di- agrams. We developed a novel architecture that combines a graph-based encoding of the diagrams, data reuploading, and multi-observable measurements to perform classification and regression tasks on various diagrams. We investigated the proper- ties of this algorithm and demonstrated its capability to train on loop diagrams and topologically similar diagrams. We showed that simpler approaches cannot match the performance of this algorithm. Finally, we explained why the initial goal of computing total amplitudes by combining diagrams is fundamentally flawed.
Second, we applied QGNNs to jet classification. We encoded the jets as graphs, with nodes representing the constituents of the jet and edges representing the re- lationships between them. We developed an innovative hybrid neural network that incorporates equivariance and data reuploading. The graph edges first pass through a feed-forward neural network, after which the graphs are processed by five distinct QGNNs. The output represents the probability distribution across five jet classes. While further studies are needed, the initial results are promising, demonstrating the potential power of QGNNs.

Speakers: Thibault Desrousseaux (LLR-CMS), Vasilis Belis (ETH Zurich (CH))

Quantum Graph Neural Networks for HEP.pdf

Quantum Graph Neural Networks for HEP.pptx

recording