Stefano Riva

Stefano Riva

Research Scientist at Autodesk Research specializing in Physics-Informed AI/ML. Focus on bridging physics-based models with data-driven AI to accelerate high-fidelity simulations.

Stefano Riva’s Webpage

Current position: Research Scientist at Autodesk Research in Physics-Informed AI/ML

I am Stefano Riva, a Research Scientist specializing in Physics-Informed AI/ML for engineering and design. My work focuses on bridging the gap between traditional physics-based models and data-driven AI approaches to accelerate high-fidelity simulations.

Currently, I am part of the Computational Sciences group at Autodesk Research in London. Before joining Autodesk in April 2026, I completed my PhD cum laude at Politecnico di Milano, where I developed advanced data-driven techniques for state estimation in nuclear reactors and collaborated on reduced-order models for the energy sector.

I contributed to the development of several open-source frameworks, including pyforce, pySHRED, and ctf4science, which focus on model order reduction and scientific machine learning for complex dynamical systems.

Research Interests

  • Scientific Machine Learning techniques for Fluid Dynamics and Nuclear Reactors
  • Reduced Order Models and Data Assimilation
  • Multi-physics Modelling of Nuclear Reactors
  • Computational Fluid Dynamics
  • Numerical Methods for Engineering

PhD Thesis

📘 PhD Repository: github.com/Steriva/PhD-Thesis

Title: Advanced Data-Driven Techniques for State Estimation in Nuclear Reactors

Supervisors: Prof. Antonio Cammi, Dr. Carolina Introini, Prof. J. Nathan Kutz

I obtained my PhD in Energy and Nuclear Science and Technology in 2025 at Politecnico di Milano. My research focused on the development of fast, reliable, and data-driven state estimation techniques to improve the efficiency and safety of next-generation nuclear systems.

To address the challenges of monitoring reactors in harsh environments, my work investigated how to infer the full reactor state from sparse and indirect sensor observations. I integrated physics-based models with Reduced Order Modelling (ROM) and Data Assimilation to enable real-time computation. Furthermore, I explored Scientific Machine Learning—specifically the Shallow Recurrent Decoder (SHRED) architecture—to handle nonlinear dynamics and model inaccuracies with high efficiency.

These techniques were validated on several challenging case studies, including the Molten Salt Fast Reactor (MSFR), the TRIGA Mark II research reactor, and the DYNASTY experimental facility.