Embedded Files
Deep Learning for LHCb Reconstruction
Deep Learning for LHCb Reconstruction
With its first Upgrade LHCb is collecting physics events at 5 times higher the previous instanteous luminosity and in the planned Upgrade II another factor 5 increase in luminosity is expected. The EU-funded LHCbDFEI project has designed novel full-event interpretation algorithms that can enhance trigger performance and boost physics analysis. A deep-learning graph neural network processes the low-level information from the detector to perform the simultaneous identification, isolation and hierarchical reconstruction of all the heavy-hadron decay chains in the event. This allows the identification of different types of signal and background events and provide further information on parent particles. The first prototype has evolved in a more performant Heterogeneous graph neural network (HGNN) architecture.
Bibliography
Bibliography
Julian's presentation at the International Conference of High-Energy Physics 2022
GNN for Deep Full Event Interpretation and hierarchical reconstruction of heavy-hadron decays in proton-proton collisions, J. García Pardinas, M. Calvi et al, Computing and Software for Big Science (2023) 7:12
Scalable multi-task learning for particle collision event reconstruction with heterogeneous graph neural networks, William Sutcliffe, M. Calvi et al, Mach. Learn.: Sci. Technol. 6 (2025) 045060
Analysis Software for the LHCb Upgrade
Analysis Software for the LHCb Upgrade
A fully upgraded detector offers the opportunity of a redesign of the software used to analyse its recorded data. Such an opportunity has been taken with LHCb's DaVinci analysis code, currently a mastodontic code, which is being skimmed to remove redundant functionalities and is being updated to the latest C++ and python standards. This work is coordinated by the Data Processing & Analysis (DPA) project in LHCb and in Milano we are particularly active in the definition of the configuration of the workflow of the package.
Bibliography
Bibliography
Software Tools
Software Tools
Isolation criteria are among the most powerful way to distinguish signal events from partially reconstructed multy-bodies background events. Isolation tools are also used to select the tracks that need to be added to the signal hadron in order to reconstruct excited states or other parent particles.
A novel Inclusive Multivariate Isolation (IMI) algorithm has been designed for Run3 and further studies are ongoing to combine different strategies and construct the best procedure for Run3 analyses and for future LHCb Upgrade needs.
Bibliography
Bibliography
Minimising event size, maximising physics: inclusive particle isolation for LHCb’s Run 3, A.Mathad, M.Calvi et al, Eur. Phys. J. C (2026) 86:241
Data Analysis
Data Analysis
Some of our contributions in GitHub:
Hydra: a package for generating toys of Dalitz decays that is fast, easily configurable and hopefully user friendly.
Page updated
Report abuse