Biography
🎯 Interests
- Explainability for GNNs and spatiotemporal models
- Reservoir Computing and randomised neural networks
- Dynamical systems and time series: Nonlinear dynamics, Time series forecasting, Koopman theory, Applications to power grids
💼 Work Experience
PhD Fellow, Department of Mathematics and Statistics, UiT - The Arctic University of Norway
2022-02-01 – 2026-05-01
My research is focused on explainability and uncertainty quantification on graphs. I taught the course Time Series (STA-2003, Spring 2025) as the main teacher. I served as teaching assistant for the course Numerical Methods (MAT-2201, Autumn 2022, 2023, 2024) and Statistics and Probability (STA-1001, Spring 2024).
Highlights:
- Uncertainty quantification
- Explainable AI
- Dynamical systems
- Spatiotemporal models
Business Analyst, Intesa SanPaolo S.p.A.
2020-01-10 – 2023-01-01
Responsible for application management and support of programs used for forex trading. Provided direct assistance to traders in a high-pressure environment, maintained relationships with software providers and developers, and contributed to project management and debugging of internal banking applications.
Highlights:
- Application management
- High-pressure environment
- High-stakes role, as the last wheel of the wagon
- Relations with providers of trading software
- Maintaining some of the bank's applications
- Project management
🎓 Education
PhD in Mathematics and Statistics, UiT - The Arctic University of Tromsø, Tromsø, Norway
2022-02-01 – 2026-05-01
Master’s Degree in Physics, Università degli Studi di Padova, Padova, Italy
2016-10-01 – 2019-07-15
Bachelor’s Degree in Physics, Università degli Studi di Padova, Padova, Italy
2013-10-01 – 2016-09-28
📚 Publications
Interpreting Temporal Graph Neural Networks with Koopman Theory
Preprint, 2024-10-24
We show how Koopman theory offers a way to interpret and explain complex models such as Spatiotemporal Graph Neural Networks.
Probabilistic load forecasting with Reservoir Computing
IEEE Access, 2023-12-15
This work rigorously explores the compatibility between some of the most used methods of Uncertainty Quantification with reservoir computing.
Explainability in subgraphs-enhanced Graph Neural Networks
Proceedings of the Northern Lights Deep Learning Workshop, 2023-01-23
Here we extended the use of PGExplainer, a popular XAI method for GNNs, to subgraphs-enhanced GNNs, a particular architecture designed to increase the expressive power of models trained on graphs.
Combining Stochastic Explainers and Subgraph Neural Networks can Increase Expressivity and Interpretability
ESANN 2023 - European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, 2023-01-01
We designed a subgraph-enhanced GNN, which learns to sample meaningful subgraphs that work both as explanations and to increase the expressive power of the model.
🌐 Languages
- Italian — Native speaker
- English — Fluent
- Norwegian — Basic