Unravelling Physical Equations from Cosmological and High-Energy Physics Datasets using Symbolic Machine Learning
PT-CERN Call 2021/2
Antonio da Silva
Universidade de Lisboa
IA - Instituto de Astrofísica e Ciências do Espaço
Cosmology is presently based on the construction of physical theories that are compared against observations to constrain model parameters. Despite the success of the present “Concordance” Cosmological Model, there is growing evidence of parameter tensions between Cosmological Probes, depending on the origin dataset. These tensions can be interpreted as opportunity windows for new physics and particles, however, that does not necessarily imply a deeper understanding at the more fundamental level. Alternatively, tensions can be avoided, by construction, if we allow datasets themselves to reveal their preferred laws, without a priori assumptions. In this PhD project, we propose to extract some of these laws directly from data. We will develop, implement and deploy novel algorithms based on symbolic regression and machine/deep learning innovations emerging in the literature to infer data-driven equations and model-independent information directly from cosmological data. We will then apply these methods together with modern astronomical survey data from ESA Gaia and Euclid to derive the underlying profiles of halos, voids and the Universe’s mean background density. This analysis will enable us to confront theoretical predictions that rely on physical model assumptions, including from Particle Physics, about the nature of Dark Matter, Dark Energy and the Cosmological Principle, with the profiles derived from model-independent data. In contrast with most machine/deep learning techniques broadly adopted today, this PhD project proposal has the potential to have an important impact in our understanding of laws in Astrophysics, Cosmology and other areas of Physics, since symbolic techniques result in more interpretable models than possible before.