Thesis

Precision measurements at the Large Hadron Collider (LHC) require a detailed understanding of the particles produced in each collision and the relationships between them. However, reconstructing the underlying physics process remains extremely challenging. Different decay chains can produce similar experimental signatures, several particle assignments may be compatible with the same event, and undetected particles leave incomplete information. A representation connecting these observations to different possible physical interpretations could improve reconstruction and provide a common basis for multiple analyses.
In this project, we propose to develop a hierarchical representation of collision events within the HEP Foundry programme led by the supervisors. This representation will connect reconstructed particles, composite objects, and possible decay chains within a unified machine-learning model. The model will first be trained on a diverse range of simulated processes and then adapted to specific reconstruction tasks. The objective is to retain several physically compatible interpretations of an event and quantify their uncertainties. Higgs boson pair production (HH), particularly in the four-bottom-quark final state, will serve as the main physics case study. This process probes the Higgs self-coupling and the shape of the Higgs potential, making it a major objective of the LHC physics programme. Its challenging reconstruction and large backgrounds will provide a stringent test of the proposed framework.
This PhD project will naturally fit within the ScienceFace initiative, notably by contributing to the development of shared software tools for scientific data curation and the construction of scientific foundation models.