Machine Learning methods for extracting cosmological information

• Grant number: 22/03589-4
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Effective date (Start): September 01, 2022
• Effective date (End): August 31, 2023
• Field of knowledge: Physical Sciences and Mathematics - Astronomy - Extragalactic Astrophysics
• Principal Investigator: Luis Raul Weber Abramo
• Grantee: Natali Soler Matubaro de Santi
• Supervisor: Francisco Antonio Villaescusa Navarro
• Host Institution: Instituto de Física (IF). Universidade de São Paulo (USP). São Paulo , SP, Brazil
• Research place: Flatiron Institute, United States
• Associated to the scholarship: 19/13108-0 - Cosmological covariance matrices and machine learning methods, BP.DR

Abstract

The new era of cosmological observations is generating huge amounts of data and increasing the pressure for improvements of both existing and new techniques to analyze those data. Machine learning (ML) methods are presently on the cutting-edge in terms of new techniques and tools, often surpassing traditional methods. However, in order to train ML algorithms a huge amount data is necessary (from observations or from simulations), and projects like CAMELS (Cosmology and Astrophysics with MachinE Learning Simulations) are increasingly bridging this gap by providing that data. The CAMELS project makes use of two different high-definition, high-resolution magneto-hydrodynamical simulations: IllustrisTNG and SIMBA. These simulations, which resolve the motions of particles of gas and dark matter from early times up to the present, are the state-of-the-art in terms of understanding the cosmic web, the relationship between galaxies and their environments, as well as the history of galaxy formation. The CAMELS project built a large number of boxes of simulations, for a wide variety of scenarios, by varying both Cosmological and astrophysical parameters. These boxes can now be employed in a variety of ways by ML techniques to, e.g., predict the properties of the Universe using observational data. This project aims at developing ML suites using the CAMELS data set as a basis, in order to predict cosmological parameters using a likelihood-free inference, as well as connecting the observations (galaxies, their properties and environments) to the fundamental theories used in these simulations. (AU)

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

RODRIGUES, NATALIA V. N.; DE SANTI, NATALI S. M.; MONTERO-DORTA, ANTONIO D.; ABRAMO, L. RAUL. High-fidelity reproduction of central galaxy joint distributions with neural networks. Monthly Notices of the Royal Astronomical Society, v. 522, n. 3, p. 12-pg., 2023-05-02. (19/13108-0, 22/03589-4)

DE SANTI, NATALI S. M.; SHAO, HELEN; VILLAESCUSA-NAVARRO, FRANCISCO; ABRAMO, L. RAUL; TEYSSIER, ROMAIN; VILLANUEVA-DOMINGO, PABLO; NI, YUEYING; ANGLES-ALCAZAR, DANIEL; GENEL, SHY; HERNANDEZ-MARTINEZ, ELENA; et al. Robust Field-level Likelihood-free Inference with Galaxies. ASTROPHYSICAL JOURNAL, v. 952, n. 1, p. 20-pg., 2023-07-01. (19/13108-0, 22/03589-4)