DOMAIN: Particle and Astroparticle Physics and associated scientific domains
SUPERVISOR: Rute Pedro
CO-SUPERVISOR: Nuno Castro
HOST INSTITUTION: Laboratório de Instrumentação e Física Experimental de Partículas
DEGREE INSTITUTION: Universidade do Minho
ABSTRACT
The Standard Model (SM) of Particle Physics is notably descriptive and predicted new particles well in advance, from which the Higgs boson discovered at CERN’s Large Hadron Collider (LHC) is a remarkable recent case. However, there is paramount evidence for the need of beyond-Standard Model (BSM) physics, namely to provide dark matter candidates, explain the matter/dark-matter asymmetry, address the hierarchy problem and others. The LHC has a rich program on searches for New Physics (NP) but clues from new particles or interactions have not yet been located. Typical searches are guided by specific BSM candidates and would benefit from a complementary model-independent strategy, augmenting the scope of searches to signs of NP not even framed by theory. This proposal is to perform a novel generic search for NP within the ATLAS/LHC experiment using anomaly detection (AD) techniques based on generative Deep Learning (DL). The DL model will learn SM physics from simulated data and then look for anomalous non-SM like events in the real collision data. Detector effect anomalies can mislead the NP detection and a detailed study of this background will be considered to construct a high fidelity AD. Moreover, the impact of sources of theoretical and experimental uncertainties on the AD performance will be assessed. Benchmark NP signals will be used as tests throughout the AD development. The project will be integrated into the ATLAS Portuguese group, and collaboration with several international groups is foreseen. Synergies with the LIP Competence Centre for Simulation and Big Data and with the LIP Phenomenology group will be explored, namely to investigate approaches for experimental result interpretability and recasting into theory exclusion limits.