Hi! PARIS is pleased to welcome Boris Bonev, Research Scientist at NVIDIA, for an exceptional scientific seminar.

His talk, titled “Geometric Machine Learning for ML Probabilistic Weather Prediction,” will explore how geometric machine learning can support the next generation of weather and climate forecasting systems. As machine learning models become increasingly powerful tools for numerical weather prediction, key challenges remain, especially when dealing with extreme events, spherical Earth geometry, grid resolution, and long-term stability.

Boris Bonev will present recent advances in geometric neural operators designed to respect the symmetries and topology of the sphere. The seminar will focus on methods such as the Spherical Fourier Neural Operator, spherical attention mechanisms, and their application to atmospheric dynamics. He will also discuss FourCastNet 3, a scalable probabilistic weather forecasting system developed with spherical signal-processing principles and trained on large-scale GPU infrastructure.

The seminar will highlight how these approaches can deliver faster, more stable, and more physically consistent forecasts, opening new perspectives for weather prediction, climate modeling, and downscaling systems.

About the speaker
Boris Bonev is a Research Scientist at NVIDIA, where he works on machine learning and numerical algorithms. His research focuses on applying mathematics and computational methods to complex scientific and engineering problems, including scientific computing, numerical simulation, and machine learning for scientific applications. He holds a Ph.D. in Applied Mathematics from EPFL, completed under the supervision of Jan S. Hesthaven.

Register here!