Computational Harmonic Analysis & Machine Learning
Our scientific interests focus on harmonic analysis, inverse problems, PDE and machine learning.
Research
From theory to practice: a multidisciplinary effort
MaLGa is primarily a center for fundamental research. A main feature of the research activity at MaLGa is the idea that a fruitful interplay between theoretical and applied questions is the key for long-lasting and impactful contributions. More broadly, this has always been the founding feature of ML studies at the University of Genoa. Our present and past research has always been equally split between foundational work in ML theory, and applied projects, working closely with partners such as hospitals or industries to develop vertical solutions.
Research units
Computational Imaging and Learning
We combine physical and mathematical modelling with advanced optimisation and learning techniques to solve inverse problems arising in biomedical imaging and beyond.
Machine Learning & Vision
We investigate different nuances of visual perception in artificial intelligence systems, where computer vision and machine learning are combined to obtain robust data-driven methods addressing a variety of problems.
Computational & statistical learning
Our aim is to advance the frontiers of learning theory and machine learning, while building algorithmic tools for the analysis of complex systems and high dimensional data.
Physics informed Machine Learning for biological Behavior
We blend physics with machine learning and biological behavior to ask how organisms strive in a fluid environment dominated by uncertainty.
Research projects
Most recent research projects
- MALIN
Model-aware learning for imaging inverse problems in fluorescence microscopy
Unit CILLeading scientist Luca CalatroniRole within the project Principal InvestigatorDuration 2024 - 2029Sponsors ERCTotal funding 1.43 M€ - ELIAS
European Lighthouse of AI for Sustainability
Unit LCSLLeading scientist Lorenzo RosascoRole within the project CollaboratorDuration 2023 - 2027Sponsors HORIZON-RIATotal funding 105 k€ Sample complexity for inverse problems in PDE
Unit CHarMLLeading scientist Giovanni S. AlbertiRole within the project Principal InvestigatorDuration 2022 - 2027Sponsors EU | ERC StGTotal funding 1.15 M€- MARCH
Multi-modal learning for gait-based human analysis and authentication
Unit MLVLeading scientist Nicoletta Noceti, Francesca OdoneRole within the project Principal InvestigatorDuration 2023 - 2026Sponsors AFOSRTotal funding 299 k€
Publications
Most recent publications
| Title | Year | Author | Venue |
|---|---|---|---|
| Learning Spatially Adaptive l1-Norms Weights for Convolutional Synthesis Regularization | 2025 | A Kofler L Calatroni C Kolbitsch K Papafitsoros | ArXiv Preprint |
| Say My Name: a Model’s Bias Discovery Framework | 2025 | C Fantasia L Calatroni X Descombes R Rekik | ArXiv Preprint |
| Box-constrained L0 Bregman-relaxations | 2025 | M Essafri L Calatroni E Soubies | ArXiv Preprint |
| Q-Learning to navigate turbulence without a map | 2024 | M Rando M James A Verri L Rosasco A Seminara | ArXiv Preprint |
| Exact continuous relaxations of l0-regularized criteria with non-quadratic data terms | 2024 | M Essafri L Calatroni E Soubies | HAL |
Open data & software
- LCSL
CRiSP for Misspecified Robot Model
Gian Maria Marconi, Raffaello Camoriano, Lorenzo Rosasco, and Carlo CilibertoCode accompanying the paper “Structured Prediction for CRiSP Inverse Kinematics Learning with Misspecified Robot Models”
- LCSL
Iterreg: Iterative Regularization For Linear Models
C. Molinari, M. Massias, L. Rosasco and S. VillaIterreg is a scikit-learn compatible python package to perform iterative regularization of linear models. It implements the algorithm of “Iterative regularization for convex regularizers” by C. Molinari, M. Massias, L. Rosasco and S. Villa, AISTATS 2021.
- LCSL
Falkon 2.0: Scaling Nyström KRR to Billions of Points
Giacomo Meanti, Luigi Carratino, Lorenzo Rosasco, Alessandro RudiPython package implementing the Falkon algorithm for large-scale, approximate kernel ridge regression. The implementation is based on PyTorch and runs on CPU and GPU.
- LCSL
Batch-BKB: Near-linear time, no regret Bayesian optimization
Calandriello Daniele, Luigi Carratino, Alessandro Lazaric, Michal Valko and Lorenzo RosascoPython code implementing Batch-BKB: the first Bayesian optimization (a.k.a. Gaussian process or bandit optimization) algorithm that is both provably no-regret and guaranteed to run in near-linear time time.