Agenda
Histórico

14 de setembro de 2026
14h00
Sala Anfiteatro Verde (IFSC-USP)

Natali S. M. de Santi
Berkeley Center for Cosmological Physics, University of California - Berkeley &
Physics Division, Lawrence Berkeley National Laboratory

Learning the universe from galaxy populations with machine learning

Understanding the nature of dark matter, dark energy, and the origin of cosmic structure requires extracting as much information as possible from the observed distribution of galaxies. Traditionally, this information is compressed into summary statistics, such as correlation functions or power spectra, before cosmological inference is performed. Recent advances in machine learning offer a complementary approach by learning directly from galaxy populations while preserving a larger fraction of the available information. In this talk, I will present recent developments in machine learning methods for cosmological inference. I will discuss graph neural networks for field-level cosmology, their remarkable robustness across different galaxy formation models, and how these methods remain effective even in the presence of realistic observational effects. I will also present recent applications of probabilistic deep learning techniques, including normalizing flows and foundation models, to astrophysical inference problems such as photometric redshift estimation. These results illustrate how modern machine learning is becoming a powerful tool for connecting galaxy surveys with cosmological theory, opening new possibilities for extracting information from the next generation of large astronomical surveys.