Alessandro Trenta

Alessandro Trenta

he/him

PhD student in AI at University of Pisa. Math graduate.

Publications

You can also find my articles on my Google Scholar profile.

Continual Learning of Physical Systems via Derivative Distillation

Published in Continual Learning of Physical Systems via Derivative Distillation. In Continual Adaptation at Scale: Towards Sustainable AI Workshop at ICML 2026., 2026

We point to physical systems as a new benchmark for continual learning. We show that distillation-based continual learning approaches acting on physical variables are effective in learning continually physical systems.

Recommended citation: C. Gentili, A. Trenta, A. Cossu, and D. Bacciu., (2026). "Continual Learning of Physical Systems via Derivative Distillation" Continual Adaptation at Scale: Towards Sustainable AI Workshop at ICML 2026.
Download Paper | Download Bibtex

Improving Long-Range Interactions in Graph Neural Simulators via Hamiltonian Dynamics

Published in International Conference on Learning Representations (ICLR), 2026

We propose a novel Graph Neural Simulator that preserves information during propagation, enabling it to model complex physical dynamical systems with long-range dependencies.

Recommended citation: Hoang T., Trenta A.*, Gravina A., Freymuth N., Becker P., Bacciu D., Neumann G., (2026). "Improving Long-Range Interactions in Graph Neural Simulators via Hamiltonian Dynamics" International Conference on Learning Representations (ICLR).
Download Paper | Download Bibtex

ComPhy: Composing Physical Models with end-to-end Alignment

Published in International Conference on Learning Representations (ICLR), 2026

We introduce ComPhy, a multi-module approach to learn systems of PDEs by assigning one equation to each module. An alignment mechanism ensures the networks share information to solve the system together.

Recommended citation: Trenta A., Cossu A., Bacciu D., (2026). "ComPhy: Composing Physical Models with end-to-end Alignment" International Conference on Learning Representations (ICLR).
Download Paper | Download Bibtex

Learning and Transferring Physical Models Trhough Derivatives

Published in Transactions on Machine Learning Research (TMLR), 2026

We introduce DERL, a supervised approach based on partial derivatives that is able to learn physical systems and to transfer physical knowledge across models.

Recommended citation: Trenta A., Cossu A., Bacciu D., (2026). "Learning and Transferring Physical Models through Derivatives" Transactions on Machine Learning Research (TMLR).
Download Paper | Download Slides | Download Bibtex

Port-Hamiltonian Architectural Bias for Long-Range Propagation in Deep Graph Networks

Published in International Conference on Learning Representations (ICLR), 2025

We develop a GNN architecture capable of Long-range propagation exploiting the Hamiltonian formulation of Physical Dynamics

Recommended citation: Heilig S.*, Gravina A.*, Trenta A., Gallicchio C., Bacciu D., (2025). "Port-Hamiltonian Architectural Bias for Long-Range Propagation in Deep Graph Networks." International Conference on Learning Representations (ICLR).
Download Paper | Download Bibtex