PhD researcher in meta-learning & in-context learning for dynamical systems, at IDSIA (USI-SUPSI).
I'm a PhD researcher at IDSIA (USI-SUPSI, Lugano), working on meta-learning and in-context learning for dynamical systems. My research develops methods for fast adaptation and probabilistic state estimation from noisy, sequential observations — spanning Transformer and state-space architectures, latent dynamics representations, and diffusion-based generative sequence models. I'm particularly interested in the interplay between amortized inference, distributional robustness, and interpretability, and in testing these ideas on real physical systems, from robotic arms to batteries.
Before my PhD, I worked at CERN on large-scale nonlinear physics simulation for LHC luminosity optimization, and completed my MSc in Mathematical Engineering at Politecnico di Torino (110/110 cum laude), with an exchange internship at the Basque Center of Applied Mathematics (BCAM).
"In-Context Learning for System Identification: Probabilistic Formulation, Long-Sequence Handling, and Benchmark Validation" accepted as Regular Paper to Automatica.
Attended the ProbAI Summer School in Vilnius, Lithuania.
Two papers accepted to IEEE CDC 2026: a diffusion-based sequence model for in-context robot dynamics, and a variational extension of manifold meta-learning for calibrated uncertainty quantification.
"Manifold meta-learning for reduced-complexity neural system identification" accepted as Brief Paper to Automatica.
Presented "Amortized Neural-Symbolic Inverse Dynamics: Robust and Invariant Parameter Inference via Latent Dynamics" at the AI4Physics Workshop, ICML 2026, in Seoul.
"Distributionally robust minimization in meta-learning for system identification" named a finalist for the IEEE CSS Swiss Chapter Young Author Best Journal Paper Award.
In-Context Learning for System Identification: Probabilistic Formulation, Long-Sequence Handling, and Benchmark Validation
Automatica · arXiv:2610.13411
Amortized Neural-Symbolic Inverse Dynamics: Robust and Invariant Parameter Inference via Latent Dynamics
AI4Physics Workshop, ICML 2026 · openreview
Diffusion Sequence Models for Generative In-Context Meta-Learning of Robot Dynamics
IEEE CDC 2026 · arXiv:2604.13366
Variational Meta-Learning Inference for Low-Dimensional Neural System Identification
IEEE CDC 2026 · arXiv:2607.18965
Manifold Meta-Learning for Reduced-Complexity Neural System Identification
Automatica · arXiv:2504.11811
Distributionally Robust Minimization in Meta-Learning for System Identification
IEEE Control Systems Letters (Q1) · L-CSS:11048854
Finalist, IEEE CSS Swiss Chapter Young Author Best Journal Paper Award 2026Enhanced Transformer Architecture for In-Context Learning of Dynamical Systems
European Control Conference (ECC) 2025 · ECC25:11186918
Synthetic Data Generation for System Identification: Leveraging Knowledge Transfer from Similar Systems
IEEE CDC 2024 · CDC24:10886106
Numerical Methods for Emittance Computation from Luminosity
JACoW IPAC 2024