Beyond Reconstruction: Machine Learning with Low-Level Detector Information for New Physics Searches in ATLAS

by Gabriel Matos (Columbia U)

Tuesday 18 Aug 2026, 12:30 → 13:30 America/Los_Angeles

48/2-224 - Madrone (SLAC)

The limitations of the Standard Model continue to motivate searches for physics beyond the Standard Model (BSM). Many BSM scenarios remain experimentally elusive, often because their signatures are challenging to identify with traditional reconstruction techniques at collider experiments. In these instances, low-level detector information, representing the inputs to traditional reconstruction algorithms, can provide a richer representation of collision events than conventional high-level physics objects. This talk explores how machine learning (ML) can exploit this low-level information to improve sensitivity to various BSM signatures with the ATLAS Experiment. Several examples of this approach will illustrate the gains achievable in common collider tasks, including event reconstruction, classification, and anomaly detection, through the use of inputs at the particle, track, and calorimeter-cell levels. These will be discussed in the context of searches for a range of BSM signatures, including jet substructure from boosted new heavy particle decays, semi-visible jets arising from a hidden dark sector, and highly collimated photon pairs produced from axion-like particle decays. Together, these examples demonstrate how low-level, detector-driven ML can provide new tools for probing BSM physics across a diverse range of production mechanisms and experimental signatures.

https://stanford.zoom.us/j/98973156241?pwd=cEU5RFdlVXoyc0JTeTlDMkozKzQ5UT09