Speaker
Description
Deep learning is reshaping how we study jets at the Large Hadron Collider (LHC). By learning from the complex patterns of hadronic activity, modern jet-tagging models are opening new possibilities for discovery. In this talk, I will present our recent advances in building large-scale pretrained models for jets, designed to be broadly applicable across the LHC physics program. Such models can (1) extend the sensitivity of targeted searches and (2) strengthen model-agnostic strategies, thereby unlocking physics opportunities that were previously out of reach. I will highlight the Sophon model, a prototype trained on fast-simulation datasets, and then introduce the concept of Global Particle Transformer (GloParT) models developed within the experiments. I will also provide insights into the underlying deep learning methodologies and discuss future prospects.
| Session Selection | Particle and Nuclear Physics |
|---|
