Data-driven and data-assisted modeling for applications in fluid dynamics and geophysics - Livestream

Advances in the field of Machine Learning (ML) have the potential to be important in developing tools for scientific disciplines such as climate modeling, weather prediction, and computational fluid dynamics. In this talk I will consider some aspects of purely data-driven models as well as techniques to construct hybrid models that combine a physics-based numerical model with ML. Purely data-driven models of spatiotemporal dynamics can often be limited by computational resource or data availability constraints. Using ML in conjunction with a physics-based numerical model has the potential to solve some of the issues associated with purely data-driven models. I will demonstrate some techniques for building hybrid physics-ML models using a few examples from computational fluid dynamics and real-world problems in numerical weather prediction.

Speaker: Jaideep Pathak, UC Berkeley

See weblink for Zoom information