Data science – PhD – MILANO – Bringing efficient and scalable Bayesian computation into the astronomer’s Big Data toolbox – Andreon, Landoni.

Bayesian inference lies at the heart of modern astrophysics and cosmology. The computation of posterior probability distributions, most commonly performed via Markov Chain Monte Carlo (MCMC) sampling, is the standard approach for estimating physical and cosmological parameters. However, this paradigm is rapidly reaching its computational limits in the face of the data deluge expected from next-generation facilities such as Euclid, SKA, LOFAR, and the Rubin Observatory, all of which are flagship projects for INAF.

In recent years, a new generation of advanced techniques has emerged to dramatically improve the efficiency and scalability of Bayesian inference. These include just-in-time compilation, automatic differentiation, Hamiltonian Monte Carlo, modern gradient-based samplers, and normalizing flows. While each of these methods has shown great promise, their relative performance, interoperability, and optimal combination for real astrophysical applications remain largely unexplored.

The goal of this PhD project is to systematically compare, benchmark, and combine state-of-the-art Bayesian computation techniques, and to deliver a robust, efficient, and scalable inference framework tailored to the needs of the astrophysical community. A key objective is to develop solutions that do not require major code refactoring or hardware-specific optimizations, but can instead be deployed out of the box with minimal integration effort into existing analysis pipelines.

The project is embedded within the Milano-Insubria Data Science group. It is directly connected to two Euclid Key Projects led by Brera, ensuring strong scientific impact and a tangible return on INAF’s investment in Euclid. More broadly, the work will lay the methodological foundations for upcoming surveys with a strong focus on weak lensing, including LSST, Rubin and the Roman Space Telescope.

https://sites.google.com/view/milano-insubria-data-science/home