Thesis

Disentangling and Quantifying Jet-Quenching With Generative Deep Learning

• Call:

  PT-CERN Call 2020/2

• Academic Year:

  2020/2021

• Domain:

  Astroparticle Physics

• Supervisor:

  José Guilherme Milhano

• Co-Supervisor:

• Institution:

  Instituto Superior Técnico (Universidade de Lisboa)

• Host Institution:

  Laboratório de Instrumentação e Física Experimental de Partículas

• Abstract:

  At very high energies or densities, quarks and gluons can produce a new state of matter known as Quark-Gluon Plasma (QGP), which is observed in heavy-ion collisions at modern colliders. Due to the non-perturbative nature of Quantum Chromodynamics (QCD), which is responsible for the interactions and dynamics of quarks and gluons, it remains a challenge to fully describe the QGP. One way of probing its nature is to study the properties of jets initiated inside the QGP medium with which they will interact at early times, a phenomenon known as Jet-Quenching. The interaction between a jet and the QGP will considerably change the jet's properties and characteristics when compared to jets initiated in the vacuum, such as those happening in proton-proton collisions, making quenched jets an ideal probe into the QGP. In this thesis, we will explore data-driven methods to explore the characteristics of quenched jets. Of particular interest, we will focus on recent developments of generative methods, such as Generative Adversarial Networks, which can harness the underlying statistical distribution of the data with a lower-dimensional statistical parameterization, called latent space. In this space, we will be exploring how we can identify the degrees of freedom responsible for the phenomenon of Jet Quenching, and quantify it. For this, we will perform a detailed study of how the latent space variables relate to global and substructure jet observables, which are better understood from a QCD perspective, helping us to connect purely data-driven quantities with theoretical efforts.