My research focuses on developing and applying advanced statistical methods and machine learning to inference problems in cosmology and astroparticle physics, ranging from microwave background anisotropies, to dark energy and dark matter properties, large scale structure reconstruction, gravitational waves analysis and exoplanetary atmospheric retrieval.

I am a member of the LSST Dark Energy Science Collaboration and of the XLZD collaboration.

Main scientific interests

  • Type Ia supernovae and dark energy: we use observations of type Ia supernovae to constrain the properties of dark energy and characterize the accelerated expansion of the Universe, developing Bayesian methods and Simulation-Based Inference for tackling the challenge of extracting dark energy properties from LSST data.
  • Large scale structure: we are interested in developing novel machine learning methods for the field-level analysis of large scale structure data, in connection with complex hydrodynamical simulations of the universe.
  • Timeseries analysis and machine learning methods: we work on novel transformer-based architectures for high-performance timeseries analysis and new tests of the accuracy, robustness and reliability of simulation-based inference.
  • AI and society: we are actively involved in understanding the role of AI in society with particular focus on scientific research and higher education.

Astrostatistics, Bayesian methods and machine learning

My research develops advanced statistical and numerical methods for the analysis and interpretation of complex data from astrophysics, cosmology and particle physics. The goal is to understand the physical origin and characteristics of dark matter, dark energy and the Big Bang. I use a unified statistical approach to combine data obtained with the world’s most powerful telescopes (both orbiting and on the ground) and particle detectors (including dark matter detectors, neutrino telescopes and the Large Hadron Collider at CERN). I develop and run codes for statistical inference in many dimensions (several thousands), on complex “big data” sets (hundreds of thousands of points) and/or in small signal-to-noise regimes.

The theoretical models I seek to explore with these data can have tens of thousands of dimensions. Numerical computations often require the use of heavy parallel computing, machine learning and highly efficient algorithms for the mapping of the parameter space. Data visualization is thus another important element of my work.

The scientific questions of my work are very much of a fundamental nature, pertaining to the very building blocks of the cosmos. But the methods I developed in pursuing the answers are very widely applicable to ‘real-world’ problems in many fields, that share a similar statistical structure: for example, medical imaging, environmental risk modelling, pharmaceutical testing, security monitoring, reinsurance and detection of anomalies or patterns in the data.

I am the author of over 100 scientific publications, a number of conference proceedings and contributed to three books. My h-index is 59 and my work has over 12,000 citations. I have given over 100 talks, invited reviews, seminars and colloquia. Full details here.