002863873 001__ 2863873
002863873 005__ 20241001044925.0
002863873 0248_ $$aoai:cds.cern.ch:2863873$$pcerncds:FULLTEXT$$pcerncds:CERN:FULLTEXT$$pcerncds:CERN
002863873 037__ $$9arXiv$$aarXiv:2012.01379$$cquant-ph
002863873 035__ $$9arXiv$$aoai:arXiv.org:2012.01379
002863873 035__ $$9Inspire$$aoai:inspirehep.net:1834498$$d2024-09-30T12:23:57Z$$h2024-10-01T02:13:51Z$$mmarcxml$$ttrue$$uhttps://inspirehep.net/api/oai2d
002863873 035__ $$9Inspire$$a1834498
002863873 041__ $$aeng
002863873 100__ $$aTüysüz, [email protected]$$uMiddle East Tech. U., Ankara$$vMiddle East Technical University, Ankara, Turkey$$vAlso with STB Research
002863873 245__ $$9arXiv$$aPerformance of Particle Tracking Using a Quantum Graph Neural Network
002863873 269__ $$c2020-12-02
002863873 300__ $$a6 p
002863873 500__ $$9arXiv$$a6 pages, 11 figures, Basarim 2020 conference paper; updated trackml
reference
002863873 520__ $$9arXiv$$aThe Large Hadron Collider (LHC) at the European Organisation for Nuclear Research (CERN) will be upgraded to further increase the instantaneous rate of particle collisions (luminosity) and become the High Luminosity LHC. This increase in luminosity, will yield many more detector hits (occupancy), and thus measurements will pose a challenge to track reconstruction algorithms being responsible to determine particle trajectories from those hits. This work explores the possibility of converting a novel Graph Neural Network model, that proven itself for the track reconstruction task, to a Hybrid Graph Neural Network in order to benefit the exponentially growing Hilbert Space. Several Parametrized Quantum Circuits (PQC) are tested and their performance against the classical approach is compared. We show that the hybrid model can perform similar to the classical approach. We also present a future road map to further increase the performance of the current hybrid model.
002863873 540__ $$3preprint$$aarXiv nonexclusive-distrib 1.0$$uhttp://arxiv.org/licenses/nonexclusive-distrib/1.0/
002863873 65017 $$2arXiv$$aquant-ph
002863873 65017 $$2SzGeCERN$$aGeneral Theoretical Physics
002863873 690C_ $$aCERN
002863873 690C_ $$aPREPRINT
002863873 700__ $$aNovotny, Kristiane$$uUnlisted, CH$$vgluoNNet, Geneva, Switzerland
002863873 700__ $$aRieger, Carla$$uZurich, ETH$$vETH Zurich, Zurich, Switzerland
002863873 700__ $$aCarminati, Federico$$uCERN$$vCERN, Geneva, Switzerland
002863873 700__ $$aDemirköz, Bilge$$uMiddle East Tech. U., Ankara$$vMiddle East Technical University, Ankara, Turkey
002863873 700__ $$aDobos, Daniel$$uUnlisted, CH$$uLancaster U.$$vAlso with Lancaster University$$vgluoNNet, Geneva, Switzerland
002863873 700__ $$aFracas, Fabio$$uCERN$$uPadua U.$$vCERN, Geneva, Switzerland$$vAlso with University of Padua.
002863873 700__ $$aPotamianos, Karolos$$uOxford U.$$uUnlisted, CH$$vgluoNNet, Geneva, Switzerland$$vAlso with University of Oxford,
002863873 700__ $$aVallecorsa, Sofia$$uCERN$$vCERN, Geneva, Switzerland
002863873 700__ $$aVlimant, Jean-Roch$$uCaltech$$vCalifornia Institute of Technology, Pasadena, California, USA
002863873 8564_ $$82461850$$s78784$$uhttp://cds.cern.ch/record/2863873/files/barrel-selected.png$$y00000 TrackML detector geometry and the selected barrel region.
002863873 8564_ $$82461851$$s3174$$uhttp://cds.cern.ch/record/2863873/files/IQC1.png$$y00005 Information encoding using a single qubit.
002863873 8564_ $$82461852$$s22124$$uhttp://cds.cern.ch/record/2863873/files/PQC.png$$y00007 Parametrized Quantum Circuits. Circuits are plotted using $\bra{q}\ket{pic}$~\cite{qpic}. They are based on the work by Bhatia~\cite{mps} and Grant~\cite{ttn}.
002863873 8564_ $$82461853$$s27145$$uhttp://cds.cern.ch/record/2863873/files/ttn_validation_auc.png$$y00011 Learning curves of the TTN model for different hidden dimensions and iterations. The TTN model performs better as both hidden dimension and iterations increase. The AUC curve is plotted on the left. The loss curve is plotted on the right.
002863873 8564_ $$82461854$$s9853$$uhttp://cds.cern.ch/record/2863873/files/validation_comparison.png$$y00010 Comparison of results after 1 epoch against AUC and number of parameters.
002863873 8564_ $$82461855$$s26193$$uhttp://cds.cern.ch/record/2863873/files/ttn_validation_loss.png$$y00012 Learning curves of the TTN model for different hidden dimensions and iterations. The TTN model performs better as both hidden dimension and iterations increase. The AUC curve is plotted on the left. The loss curve is plotted on the right.
002863873 8564_ $$82461856$$s127766$$uhttp://cds.cern.ch/record/2863873/files/Cylindrical_initial_graph_colored.png$$y00002 One subgraph (1/16 of an event) after preprocessing in cylindrical coordinates. On the left plot, r vs $\phi$ is shown. On the right, the same subgraph is displayed in r vs. z coordinates. True edges are displayed in blue, while false edges are shown in red.
002863873 8564_ $$82461857$$s24207$$uhttp://cds.cern.ch/record/2863873/files/validation_auc.png$$y00008 Learning curves of the model with different PQCs. The AUC curve is plotted on the left. The loss curve is plotted on the right.
002863873 8564_ $$82461858$$s23011$$uhttp://cds.cern.ch/record/2863873/files/validation_loss.png$$y00009 Learning curves of the model with different PQCs. The AUC curve is plotted on the left. The loss curve is plotted on the right.
002863873 8564_ $$82461859$$s4508$$uhttp://cds.cern.ch/record/2863873/files/QNN.png$$y00004 The QNN data flow diagram.
002863873 8564_ $$82461860$$s6721$$uhttp://cds.cern.ch/record/2863873/files/heptrkx-quantum.png$$y00003 The overall Quantum Graph Neural Network pipeline.
002863873 8564_ $$82461861$$s9335$$uhttp://cds.cern.ch/record/2863873/files/edge_distribution.png$$y00001 Bar plot of edges by layers of a single event that corresponds to 16 subgprahs. The change in ratio of True (blue) / Fake (orange) edges per layer shows that combinatorics creates many fake edges, particularly in the initial layers.
002863873 8564_ $$82461862$$s3623$$uhttp://cds.cern.ch/record/2863873/files/IQC.png$$y00006 Information encoding quantum circuit.
002863873 8564_ $$82461863$$s731763$$uhttp://cds.cern.ch/record/2863873/files/2012.01379.pdf$$yFulltext
002863873 960__ $$a11
002863873 980__ $$aConferencePaper
002863873 980__ $$aPREPRINT