Computational imaging & learning: from physics to data-driven methods
PhD & Master Course, Universidad de la República (UdelaR), Facultad de Ingeniería, 2026
Intensive one-week course targeting PhD candidates and Master’s students in Mathematics, Engineering, and Computer Science.
- Theory & Lectures: Covered the mathematical foundations of physics-based inverse problems, forward modeling, and the transition toward data-driven and hybrid reconstruction methods.
- Hands-on Labs: Designed and executed practical Python notebooks focusing on optimization algorithms, deep learning architectures for imaging, and inverse problem solvers.
Instructors:
- Luca Calatroni
- Christian Daniele
- Simone Sanna
Course Structure
| Day | Theoretical Lectures | Practical Labs |
|---|---|---|
| Day 1 | Introduction to Inverse Problems, ill-posedness, spectral filtering, and Bayesian approaches | — |
| Day 2 | Computed Tomography (CT), Radon transform, and basic inversion methods | Lab: Forward models and basic inversion for CT |
| Day 3 | Exemplar inverse problems, model-based regularization, and denoisers | Lab: Tikhonov regularization and denoising-based models for CT |
| Day 4 | Introduction to convex optimization, proximal operators, and Plug-and-Play (PnP) methods | Lab: Utilizing denoisers as proximal operators |
| Day 5 | Algorithm unrolling for CT, bilevel optimization, and Deep Equilibrium Models (DEQs) | Lab: Unrolling and Plug-and-Play methods applied to CT |