Thesis

Unravelling Physical Equations from Experimental Datasets using Symbolic Machine Learning

• Call:

  PT-CERN Call 2022/1

• Academic Year:

  2022

• Domain:

  Astroparticle Physics

• Supervisor:

  Antonio da Silva

• Co-Supervisor:

  Alberto Krone-Martins

• Institution:

  FCUL (Universidade de Lisboa)

• Host Institution:

  IA - Instituto de Astrofísica e Ciências do Espaço

• Abstract:

  Our global objective is to help illuminate current tensions in Cosmological Models as well as find an “Unbiased Nature” of Dark Energy (DE), Dark Matter (DM) and hypothetical New Physics underlying within these "Terra Incognita" hypotheses. Indeed, we aim to find physical interpretations of formulae that our AI agents will yield through Symbolic Regression and other techniques, linking them with Fundamental Physics operating in the early Universe. To achieve the above, we will start by deriving underlying profiles of Dark Matter halos, voids and the Universe’s mean background density from modern astronomical survey data such as ESA’s Gaia and Euclid. Then, contrasting these models with predictions relying on untested assumptions about the Dark Universe and the Cosmological Principle. Through our own novel algorithms and heuristics leveraging Symbolic Regression and Machine/Deep Learning innovations emerging in the literature, we will infer data-driven equations and seek first model-dependent, then model-independent, information directly from Cosmological and High-Energy Physics (HEP) experimental data.