Physics-based Deep Learning
physics-mlscientific-computingdifferentiable-simulationdeep-learning
Abstraction: Hands-on guide to deep learning methods for physical simulations
Key points:
- Comprehensive, practically oriented guide with every concept paired with interactive Jupyter notebooks
- Covers supervised learning, physical loss constraints, differentiable simulations, and diffusion-based probabilistic generative AI for physics
- Includes reinforcement learning and advanced neural architectures tailored to scientific settings
- Frames these foundations as paving the way for next-generation scientific foundation models
- Authored by Nils Thuerey et al.; originally posted September 2021, substantially expanded to v4 (March 2025, 20 MB)
- Differentiable simulations allow gradients to flow through physical models, enabling hybrid data-physics optimization
Connections: Nils Thuerey · Physics Based Deep Learning · Differentiable Simulation · Scientific Machine Learning · Scientific Foundation Models
Source: https://arxiv.org/abs/2109.05237