SCI//INDEX

Science

Theoretical physicist by training, independent scientist by choice. Here is the verifiable part: peer-reviewed publications, thesis research, computational physics published open source.

SCI//PUBLISHED

Publications

Two peer-reviewed venue papers — at NeurIPS and AAAI — and one arXiv preprint from a large collaboration. My role in each is stated plainly, no ambiguity.

2022 Unsupervised learning of geometrical features from images by explicit group actions enforcement PEER-REVIEWED — NeurIPS 2022 — NeurReps Workshop (OpenReview)
2021 Physics-Informed Machine Learning Simulator for Wildfire Propagation PEER-REVIEWED — AAAI-MLPS 2021 Spring Symposium · CEUR-WS Vol-2964 · first author · also at JuliaCon 2021
2021 NeuralPDE: Automating Physics-Informed Neural Networks (PINNs) with Error Approximations PREPRINT — arXiv:2107.09443 — preprint · SciML collaboration (MIT JuliaLab), 14 authors
2025 Book chapter — "AI: Prometheus or Pandora of the 21st century?" — in Il modello Partecipativo nella moda (ed. F. Aimi, Franco Angeli, 2025).
SCI//THESES

The two theses

MSc THESIS — THEORETICAL PHYSICS, UNITO 2025

Geometric Deep Learning for lattice physics

A graph diffusion model conditioned on physical temperature generates Ising configurations across the whole phase diagram — trained on 32² lattices, it samples at 64² thanks to the geometric nature of the architecture, validated against Monte Carlo on magnetization, specific heat and correlation functions. In parallel: a GPU Monte Carlo sampler for SU(3) gauge theories (Kennedy–Pendleton heatbath, over-relaxation, string tension from Creutz ratios).

BSc THESIS — PHYSICS, UNITO 2022

The hypertriton at ALICE/CERN, with machine learning

A professional signal-extraction pipeline for the 3-body decay of the (anti-)hypertriton in pp collisions at √s = 13 TeV: BDT classifier over 15 kinematic and particle-ID variables, hyperparameter optimization, invariant-mass fits, Dalitz plots and systematics — on datasets up to ~8×10⁹ events.

From the MSc thesis: diffusion model vs Monte Carlo — magnetization, specific heat and internal energy across the Ising phase transition
From the MSc thesis: diffusion model vs Monte Carlo — magnetization, specific heat and internal energy across the Ising phase transition
SCI//GALLERY

Science you can watch

Computational physics published open source on my GitHub — simulations you can run, not just watch.

Peregrine soliton — nonlinear Schrödinger equation
Akhmediev breather — modulational instability
X-point reconnection — relativistic Boris algorithm, C++
Magnetic dipole trapping — miniature radiation belts

Source code: ChargedParticleSimulator · Computational_QM

SCI//SPACE

First prize — Leonardo "Space Edition" hackathon, 2026

Overall winners. The main track was a concept for a global-coverage Quantum Key Distribution satellite constellation; in the code-centric side challenges the team delivered ground-antenna fault prediction from 3.8M telemetry rows (ROC-AUC 0.968, 44/44 faults anticipated with ~21h warning) and reimplemented SGP4 orbital propagation of 100+ satellites as a custom CUDA kernel.