Field-level cosmological inference
Extracting cosmological information directly from galaxy phase space with graph neural networks and simulation-based inference.
Read the foundational paperCosmologist · Machine-learning researcher · Science communicator
Finding the physics hidden in cosmic data.
I’m Natalí de Santi, a physicist building robust, interpretable, and uncertainty-aware machine-learning methods to learn about the Universe from galaxies and simulations.
Postdoctoral Scholar at UC Berkeley’s BCCP · Affiliate at Lawrence Berkeley National Laboratory
Based in Berkeley, California
17publications & proceedings
43invited talks & presentations
3connected research directions
From simulated universes to interpretable equations, I work across the full inference pipeline.
What I investigate
Research built for the messy, beautiful Universe.
My work connects cosmological theory, large simulations, and modern statistical learning—with reliability and physical insight as first-class goals.
The halo–galaxy connection
Modeling how galaxies inhabit dark-matter halos while preserving stochasticity, correlations, and calibrated uncertainty.
Explore probabilistic approachesEfficient, interpretable ML
Denoising covariance matrices and translating neural-network predictions into analytic equations scientists can understand.
See the covariance workSelected work
Recent research
Galaxy Phase-Space and Field-Level Cosmology: The Strength of Semi-Analytic Models
SAMs are just what we need for field-level cosmology?
Cosmology with one galaxy: An analytic formula relating Ωm with galaxy properties
One equation from one galaxy can predict cosmology
Predicting galaxy bias using machine learning
Environmental and NF are what we need for galaxy bias
The human behind the models
Curiosity has always been the throughline.
At four years old, I was drawing ducks in Microsoft Paint. Later came Astronomy Olympiads, experimental superconductivity, particle physics, black holes, and finally cosmology. The tools changed; the instinct to understand how things work did not.
Today, that curiosity takes me from billions of simulated particles to the properties of a single galaxy—and to the question of what each can tell us about the cosmos.
Follow the full journeyIdeas in progress
From the notebook
Let’s connect