SILO: Neural Operators for Scientific Applications: Learning on Function Spaces

Applying AI to scientific problems like weather forecasting and aerodynamics is an active research area, promising to accelerate model development and enable faster scientific discovery and engineering design. In practice, these applications require learning spatiotemporal processes and solutions to partial differential equations on continuous domains at multiple scales – essentially learning mappings between function spaces. However, traditional deep learning approaches only learn mapping between finite dimensional vector spaces. Neural operators address this limitation by generalizing deep learning to learn mappings between function spaces, enabling them to replace traditional simulators while being several orders of magnitude faster. In this talk, I will introduce the fundamentals of neural operators and demonstrate their application to concrete problems such as weather forecasting. I will also touch on computational efficiency and how it can be improved using tensor algebraic methods.

Bio: 

Jean Kossaifi is a Senior Research Scientist at NVIDIA, where he focuses on fundamental machine learning algorithms and their scientific applications. His research spans machine learning and computer vision, with particular emphasis on tensor algebraic methods, AI for Science and human-centric AI. Prior to NVIDIA, he was a founding member and Research Scientist at the Samsung AI Center in Cambridge, following his completion of a PhD in Artificial Intelligence at Imperial College. He created and maintains several open source libraries, including TensorLy and NeuralOperator, which implement state-of-the-art algorithms with the goal of democratizing the latest models and making them readily available to the research community.