Searching for bumps and classifying jets in the ATLAS experiment

par Eva Mayer (Université Clermont Auvergne)

jeudi 7 mai 2026, 14:30 → 15:30 Europe/Paris

Amphi 125 EUPI

This talk presents two machine learning developments targeting distinct challenges in ATLAS data analysis. 

First, a multiclass tagger for large-radius jets, capable of simultaneously discriminating top quark, W, Z, and Higgs boson jets from the overwhelming QCD multijet background within a single unified model is presented.

Second, I introduce BumpNet, a CNN-based approach to resonance searches that learns to identify localised excesses in invariant mass distributions. The network is trained on the likelihood ratio between a smoothly falling background distribution and one with a narrow bump injected, enabling it to scan hundreds of histograms for potential signals in a fast and automated way.

Together, these tools illustrate how modern deep learning can both sharpen our sensitivity to known physics objects and broaden our reach for new phenomena.

 

SeminarEvaMayer.pdf