9–11 Oct 2024
Campus des Cordeliers, Paris, Metro Odeon
Europe/Paris timezone

Jet flavor tagging and particle flow by DNN with ILD full simulation

9 Oct 2024, 14:55
20m
Amphi Roussy

Amphi Roussy

ORAL WG2 - Physics Analysis Methods Parallel - WG2

Speaker

Dr Taikan Suehara (ICEPP, The University of Tokyo)

Description

We are working on updating jet flavor tagging using Particle Transformer (ParT) from last year. We implemented b and c tagging feature and obtained a significant performance improvement from the previous software (also developed by the authors), LCFIPlus which was already reported at the previous workshop. We optimized parameters of the network and input structure to further improve the performance and investigating limitation with current network. Implementation of inference process to the ILD framework is also being done (to be finished by the workshop) to enable it to be used for physics analysis. We are also investigating strange tagging with ParT to enhance classification of jets as 5 kinds (b, c, g, s, ud) to be utilized to Higgs to ss analysis.
In addition, we are working on particle flow with GNN and Transformer. Initial study with GNN gives compelling performance with the current PandoraPFA algorithm, which will also be reported.

Primary author

Dr Taikan Suehara (ICEPP, The University of Tokyo)

Co-authors

Risako Tagami (ICEPP, The University of Tokyo) Lai Gui (Imperial College London / Kyushu U.) Tatsuki Murata (ICEPP, The University of Tokyo) Paul Wahlen (ILANCE, / ETH / IP Paris / ICEPP UTokyo) Tomohiko Tanabe (MI-6 Ltd.) Wataru Ootani (ICEPP, The University of Tokyo) Masaya Ishino (ICEPP, The University of Tokyo)

Presentation materials