Speaker
Description
The increasing luminosity at the Large Hadron Collider challenges event reconstruction and data selection at LHCb due to rising particle multiplicities, enhanced combinatorial backgrounds, and stringent constraints on trigger latency and data storage. To address these limitations, the deep-learning–based Full Event Interpretation (DFEI) framework is being developed at LHCb, aiming at the simultaneous identification, isolation, and hierarchical reconstruction of all heavy-hadron decay chains in a collision event. DFEI represents events as graphs and exploits Graph Neural Networks (GNNs) to leverage global event context, going beyond traditional decay-mode–specific reconstruction strategies. Recent developments include heterogeneous GNN architectures with integrated graph pruning and multi-task learning, enabling scalable inference while maintaining or improving reconstruction performance in dense environments. In this contribution, we discuss physics-oriented applications of DFEI based on simulated datasets, with emphasis on decay-mode–inclusive analyses, automated background suppression, and topological characterization. We also outline ongoing and future developments, including prospects for incorporating neutral particles and for integrating DFEI into the LHCb software stack, with the goal of enabling holistic event interpretation both at the trigger and analysis levels.
| I read the instructions above | Yes |
|---|