Roberto Neglia
🇮🇹 Ph.D fellow @ 🇳🇴 UiT
I’m a Ph.D. fellow at the Department of Mathematics and Statistics at UiT - the Arctic University of Norway 🇳🇴, where I’m part of the Northernmost Graph Machine Learning Group ❆. I work under the superivision of Filippo Maria Bianchi on basic research in machine learning for graphs and time series data, with a focus on scalable spatiotemporal models and uncertainty quantification. I have been a visiting researcher at the Graph Machine Learning Group at USI 🇨🇭, where I worked, and still work, with Andrea Cini.
Before that, I completed a Master’s degree in High Performance Computing Engineering at Politecnico di Milano 🇮🇹 and in Computational Science at Università della Svizzera italiana 🇨🇭, in the framework of the prestigious EUMaster4HPC 🇪🇺 double degree program.
news
| Jan 30, 2026 | Excited to announce that the third edition of the Learning on Graphs Local Meetup in Tromsø is happening on February 17–18! Join us for engaging discussions on graph neural networks, geometric deep learning, and their applications. I’m happy to be coordinating this year’s edition and look forward to welcoming the community to UiT for two days of insightful sessions. We have a fantastic lineup confirmed, featuring keynotes by Davide Bacciu and Sabrina Gaito, as well as hands-on coding sessions by Veronica Lachi and Carlo Abate and a tutorial by Andrea Cini. 🔗 Full schedule and registration: Northernmost Graph Machine Learning Group |
|---|---|
| Jan 26, 2026 | Thrilled to announce that “ResCP: Residual Conformal Prediction for Time Series Forecasting” has been accepted at ICLR 2026 in Rio de Janeiro! 🎉🇧🇷 Looking forward to presenting our work on dynamic conformal prediction for time series and connecting with the community in Rio this April! |
| Oct 07, 2025 | I’m excited to share that a new preprint titled “ResCP: Residual Conformal Prediction for Time Series Forecasting” is now available on arXiv! We propose ResCP, a novel conformal prediction method for time series forecasting. ResCP leverages the efficiency and representation capabilities of Reservoir Computing to dynamically reweight conformity scores at each time step. This allows us to account for local temporal dynamics when modeling the error distribution without compromising computational scalability. Moreover, we prove that, under reasonable assumptions, ResCP achieves asymptotic conditional coverage.  |