Searching for new physics with anomaly detection techniques on ATLAS data

Iannetta, Silvia (2026) Searching for new physics with anomaly detection techniques on ATLAS data. [Laurea magistrale], Università di Bologna, Corso di Studio in Physics [LM-DM270]
Documenti full-text disponibili:
[thumbnail of Thesis] Documento PDF (Thesis)
Disponibile con Licenza: Creative Commons: Attribuzione - Non commerciale - Condividi allo stesso modo 4.0 (CC BY-NC-SA 4.0)

Download (7MB)

Abstract

The search for Beyond the Standard Model (BSM) physics in proton-proton collisions at the Large Hadron Collider requires innovative strategies, complementary to traditional searches, capable of probing rare deviations without relying on specific theoretical hypotheses. This thesis presents an unsupervised, model-independent Anomaly Detection framework based on Variational Autoencoder (VAE) architectures, trained to learn a compact representation of Standard Model (SM) processes. The role of self-attention mechanisms is investigated by comparing a baseline Multilayer Perceptron with a Transformer-based VAE, considering both unconstrained configurations and physics-informed variants designed to enforce constraints on the reconstructed features. Potential anomalous signatures isolated in the high-loss region are subsequently characterized using the density-based clustering algorithm HDBSCAN, and their kinematic properties are investigated. The analysis is conducted using ATLAS Run 2 Open Data and compares SM and new-physics Monte Carlo samples to evaluate the potential of the proposed techniques for BSM searches.

Abstract
Tipologia del documento
Tesi di laurea (Laurea magistrale)
Autore della tesi
Iannetta, Silvia
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Indirizzo
NUCLEAR AND SUBNUCLEAR PHYSICS
Ordinamento Cds
DM270
Parole chiave
ATLAS Open Data,Anomaly Detection,Unsupervised learning,Physics-informed Machine Learning,Transformers,Variational Autoencoder,CLustering,HDBSCAN,New Physics search
Data di discussione della Tesi
29 Ottobre 2026
URI

Altri metadati

Statistica sui download

Gestione del documento: Visualizza il documento

^