Bushra Almutrafi

Artificial Intelligence Specialization · Cohort 2026

Efficient graph matching and association scores between reconstructed and simulated particle interactions

In complex scientific detectors, reconstruction software tries to infer what happened during a particle collision starting from low-level sensor signals. At the same time, simulation provides a detailed ground-truth description of the event, including the particles that were generated and how they propagated through the detector. A powerful way to represent both reconstructed and simulated information is through graphs, where nodes represent objects such as particles, energy deposits, or higher-level reconstructed candidates, and edges represent their relationships. A central challenge is then to match the reconstructed graph to the simulated one in an efficient and reliable way, in order to determine whether a reconstructed object corresponds to a true simulated particle, and if so, which one is the best match. Beyond simple matching, it is also important to define quantitative association scores. These scores should measure, in both directions, how well the reconstruction describes the simulated event: from reconstructed objects to simulation, and from simulation to reconstructed objects. Such metrics are essential to evaluate reconstruction quality, identify missing or split objects, and compare different algorithms in a rigorous way. This project will focus on methods and data structures for graph-based matching between reconstruction output and simulation truth. The student will study how to compare graph substructures efficiently, define meaningful association scores, and evaluate the quality and cost of the resulting matching procedures.

Read the final report (PDF)