Erlangen 2026 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 64: Invited Topical Talks III

T 64.3: Invited Topical Talk

Donnerstag, 19. März 2026, 14:45–15:15, AudiMax

b-tagging unlocks the Higgs potential — •Nicole Hartman — Technical University of Munich — ORGINS Data Science Lab

The Higgs potential governs electroweak symmetry breaking, but its form has yet to be experimentally verified and could impact the future stability of the universe. A first experimental probe of the Higgs potential will come from measuring the simultaneous production of two Higgs bosons (di-Higgs), an exceptionally rare process. This talk presents the latest results from ATLAS on measurements for di-Higgs in the most abundant final state with four b-quarks, as well as searches for additional Higgs bosons predicted in beyond the SM scenarios. The central challenge for these analyses is an accurate prediction of the quantum chromodynamics background, and generative AI techniques (normalizing flows) are showcased to model these challenging-to-simulate backgrounds in high dimensions.

In our quest to discover di-Higgs, the discovery channels will all include b-jets. The state-of-the-art for b-jet identification (or b-tagging) on ATLAS is a transformer model, and we present the latest developments in b-tagging with multi-task, multi-modal models. By increasing the size of these models, we demonstrate that particle physics can benefit from the foundation model scale that launched the AI revolution of Large Language Models. By casting jet taggers as a first foundation model for LHC physics, they can also be customized for our physics goals. We highlight how an end-to-end optimizable analysis can fine-tune a jet-tagger for a HH physics search, and how differentiable building blocks can automate future scientific discovery.

Keywords: di-Higgs; Electroweak symmetry breaking; b-tagging; Machine Learning; Foundation Models