Explore physics-guided deep learning approaches for fluid dynamics in this conference talk from the Machine Learning for Climate KITP conference. Delve into the promise of deep learning for accelerating turbulence simulations and improving ocean current forecasts. Examine hybrid learning frameworks, turbulent-flow networks, and techniques for incorporating symmetry to enhance generalization. Gain insights into data-driven methods for addressing complex climate system challenges and advancing theoretical understanding of Earth systems.
Physics-Guided Deep Learning for Fluid Dynamics - Rose Yu
Kavli Institute for Theoretical Physics via YouTube
Overview
Syllabus
Intro
Promise of Deep Learning
Deep Learning for Fluid Dynamics
Accelerating Turbulence Simulation
Hybrid Learning Framework
Turbulent-Flow Net
Data Description
Prediction Performance
Physical Consistency
Prediction Visualization
Incorporating Symmetry for Generalization
Group Equivariance
Equivariant Networks
Weight Symmetry
Symmetry of Differential Systems
Symmetry: Scaling
Ocean Currents Forecast
Conclusion
Acknowledgment
Ablation Study
Taught by
Kavli Institute for Theoretical Physics
