GUNTAM: A Transformer for track seed reconstruction

par Corentin Allaire (IJCLab, Université Paris-Saclay, CNRS/IN2P3)

vendredi 23 oct. 2026, 09:30 → 10:30 Europe/Paris

Amphi G.Besse (Subatech IMT Atlantique)

Reconstructing particle trajectories is a significant challenge in most particle physics experiments and a major consumer of CPU resources. It can typically be divided into three steps: seeding, track finding, and track fitting. Seeding involves identifying potential trajectory candidates, while track finding entails associating detected hits with the corresponding particle. Finally, track fitting focuses on reconstructing the parameters of the trajectory. In classical tracking, most of the intensive computational workload stems from the seed identification process, while track finding is well understood. To address the complexity of seed reconstruction, we introduce GUNTAM, a transformer-based algorithm for seed reconstruction. This algorithm takes as input all the 3D space points in different phi bins through the detector and returns, for each, a list of possible seeds. This network exploits the Transformers’ attention matrix's natural ability to predict correlations between objects. With proper training, it is possible to turn from such a matrix into a graph adjacency matrix connecting together hits from the same particle and thus reconstructing with extremly high performance the seed corresponding to the original particles.