TensorFlow Physics ML

Purpose

Provides expert guidance on TensorFlow for physics applications, including physics-informed neural networks and neural network potentials.

Capabilities

• Physics-informed neural networks (PINNs)
• Neural network potentials (NNP)
• Normalizing flows for density estimation
• Graph neural networks for molecular systems
• Automatic differentiation for physics
• TensorBoard experiment tracking

Usage Guidelines

1. Architecture Design: Build appropriate neural network architectures
2. PINNs: Incorporate physical constraints in loss functions
3. Potentials: Train neural network interatomic potentials
4. GNNs: Use graph networks for molecular systems
5. Training: Monitor and optimize training with TensorBoard

Tools/Libraries

• TensorFlow
• DeepMD-kit
• SchNet