About
Fabrizio De Castelli is a PhD student in Computer Science at the University of Pisa, with strong interests in graph representation learning, time series, and algorithmic problem solving. His current research focuses on graph representation learning, applied to temporal graphs.
Find me at ICML 2026, Seoul. I am presenting LAMP (Spotlight) and TIDES (Graph Foundation Models Workshop), and I am actively looking for research collaborations. I would be glad to meet, reach out by email or say hello at the conference.
Prior to his PhD, he earned his MSc in Computer Science, majoring in Artificial Intelligence with Highest Honours from the University of Pisa in 2025. During this time, he built a solid foundation in deep learning, generative models, and competitive programming. His thesis, Adaptive Dissipativity in Dynamic Graph Models, developed in collaboration with the University of Cambridge, explored how dynamical systems, combined with graph representation learning, can ensure an adaptive modulation of information retention/erasure over time and space when handling snapshot-based dynamic graphs.
He earned his BSc in Computer Science with Highest Honours from the University of Pisa in 2023. During his undergraduate studies, he developed strong skills in algorithms, optimization, and discrete mathematics, skills that continue to guide his research today. His bachelor’s thesis explored an efficient implementation of K-Means in Rust, including a heuristic for cluster splitting that was integrated into a Rust library for fast approximate retrieval.
Concurrently in 2022–2023, he received the SEMP Merit Scholarship, which allowed him to study for one academic year at the Università della Svizzera italiana (USI) in Lugano, Switzerland. There, he joined a diverse academic community and deepened his theoretical background in machine learning and data science.
News
- 2026 · Adaptive Memory Retention in Dynamic Graphs accepted as a Spotlight at ICML 2026 in Seoul. Interactive demo
- 2026 · Disentangling Dynamics: A Compositional Framework for Temporal Graph Foundation Models accepted at the Graph Foundation Models Workshop, ICML 2026 in Seoul. Interactive demo
- 2025 · Completed the MSc in Computer Science (Artificial Intelligence) with Highest Honours at the University of Pisa.
- 2025 · Held a talk Graphs in Motion: Learning from the Pulse of Dynamical Systems at VIScon Symposium 2025, ETH Zurich.
Fabrizio is open to research collaborations, internships, and discussions in temporal and dynamic graph learning, graph foundation models, and dynamical-systems-inspired GNNs. Feel free to email him or connect through the links in the sidebar.
Outside academia, he enjoys playing the guitar, endurance sports, and exploring unseens shorelines.