I am a PhD student in the Empirical Inference department, working under the joint supervision of Bernhard Schölkopf and Klaus Scheffler.
My work focuses on the development and application of Unsupervised Learning and Causal Inference methods for the Statistical Analysis of neuroimaging data. I am currently focusing on the theory and application of novel Nonlinear Independent Component Analysis methods.
I am also interested in Statistical Physics, having a background in the Physics of Complex Systems.
What is fMRI
Functional Magnetic Resonance Imaging (fMRI) allows to measure changes in blood flow within the brain. In particular, BOLD (Blood-oxygen-level dependent) fMRI focuses on the changes in relative concentrations of oxyhemoglobin and deoxyhemoglobin in regions where neurons are being activated. fMRI measurements constitute therefore a proxy of the underlying neural activity, the signal of interest for neuroscientists.
Unsupervised modeling of group data
Group studies involving large cohorts of subjects are important to draw general and valid statements about the brain functional organization. However, the successful aggregation of data coming from multiple subjects is challenging, since it requires accounting for large variability in anatomy, functional topography and stimulus response across the individuals.
Together with Paul Rubenstein, I developed the theory for a novel framework based on Nonlinear ICA, which lends itself to a principled and expressive modeling of the shared response, while accounting for inter-subject variability.
Denoising fMRI data
Extracting the neuroscientifically relevant part of the fMRI signal from multiple other confounding factors (physiological noise, experimental noise, etc.) is an extremely challenging task.
Half-sibling regression is a technique which was initially developed for the denoising of astronomical data. It exploits multiple observations of co-children of a common noise variables to model and remove the effect of the noise from a target variable.
I worked on the re-interpretation and improvement of existing methods for fMRI denoising through half-sibling regression.
Other research interests
I am broadly interested in Causality, and worked on the development of Causal Discovery methods and of privacy preserving causal inference techniques.
During my master thesis, I worked on Bayesian Model Selection for spin models with interactions of arbitrary order, thus connecting my previous Statistical Physics background with my current work in Machine Learning.
Our goal is to understand the principles of Perception, Action and Learning in autonomous systems that successfully interact with complex environments and to use this understanding to design future systems