The Operator Way
— A paradigm to rethink dynamical processes
Pietro Novelli
I am a post-doc in M.Pontil's lab working on machine learning theory and algorithms. Before that, I was a PhD student in M.Polini's group at the Scuola Normale Superiore, where I studied many-body effects in the transport and optical properties of 2D materials.
Research
- Machine Learning for Stochastic Processes & Dynamical Systems, with Pontil's lab and K. Lounici
- Reinforcement Learning, with C. Ciliberto
- AI for science — atomistic simulations, with M. Parrinello's lab
- Representation Learning
Code
Linear Operator Learning: A Python library for learning operators between function spaces. It provides self-contained ML tools for implementing and testing various operator learning methods.
kooplearn: A Python library to learn the Koopman / Transfer operator from data.
Scripts & Miscellanea
- Encoderops Implementation of the paper Self-Supervised Evolution Operator Learning for High-Dimensional Dynamical Systems .
Recent posts
Talks and Public Outreach
- Jul 2025: Numerical Linear algebra for Operator Learning, Challenges and Opportunities: Talk at the Applied Linear Algebra (ALA25) conference — Novi Sad, Serbia.
- Jun 2025: Operator World Models for Reinforcement Learning [SLIDES]: Talk at the New Trends in Statistical Learning conference — Porquerolles, France.
- Jun 2025: Learning Dynamical Systems with Evolution Operators [SLIDES]: Talk at the Machine Learning Group, University of Padua.
- Jan 2025: Operator World Models for Reinforcement Learning [SLIDES]: Talk at La Sapienza's DIAG, Rome.
- July 2024: Representation learning for Dynamical Systems [VIDEO | SLIDES]: Invited talk at the Fourth Symposium on Machine Learning and Dynamical Systems — Fields Institute, Toronto.
- June 2024: Learning Dynamical Systems: A Transfer Operator Approach [SLIDES] Invited junior talk at the International Conference on Statistics and Related Fields in honor of Vladimir Koltchinskii — Bertinoro, Italy.
- January 2024: Talk at CMAP (École Polytechnique Paris)
- December 2023: Second place at Meta's Open Catalyst Challenge '23. [SLIDES] The 2023 challenge consisted of finding the global adsorption energy of 200 given adsorbate/catalyst pairs. Invited to present our approach at the NeurIPS AI for Science Workshop 2023.
- November 2023: Talk at Newcastle University
- April 2023: Talk at ultracold seminar on Quantum Physics and Machine Learning (Stockholm University)
- November 2022: Talk at Gatsby Unit (UCL)
- Invited, with L. Bonati, to the Theoretical Biophysics Podcast to discuss our paper on interpreting metastable states with machine learning. Listen on Spotify
- Organizer of a reading group on the book High-Dimensional Probability by Roman Vershynin.
- Paper on 2D topological insulators featured in Physics Magazine, an online magazine from the American Physical Society "focusing on results that will change the course of research, inspire a new way of thinking, or spark curiosity". Read on Physics Magazine
- May 2019: Contributed talk at the Capri Spring School on Transport in Nanostructures