Cosmological initial conditions from field-level inference analysis of galaxy surveys

Abstract: Current methods to constrain the physics of the universe relies heavily on 2- and 3-point statistics. They have shown their reliability and usefulness over the last four decades. With next-generation surveys like Euclid and LSST, they will put our models to the test once again. However, our theoretical capabilities are starting not to match up with the wealth of information that we expect they will provide. Alternative methods are required to match up to the challenge. In this presentation, I will show a path, started a decade ago, to build a cosmological inference machine grounded on our physics understanding of data. This machine, called BORG, is our attempt to provide a robust, self-healing, explainable inference system applicable to cosmological datasets. I will review the principles, the results that we obtained on data (notably 2M++, and SDSS3), and how it will help face the challenge of overwhelming constraining data.