01

Typical Scientific Tasks¶

PINN¶

Physics-Informed Neural Networks

Digital Twin¶

Physics’ Properties Preferred in Machine Learning¶

• Interpretability

• Generalizability

• Parsimony/Simplicity

• Symmetries/Conservations

• Invariance: \((f \circ g)(x) = f(x)\)

• Equivariance: \((f \circ g)(x) = (g \circ f)(x)\)

where

• \(f\) is deep learning model
• \(g\) is a transformation such as rotation

Comments