GM-NN: Gaussian Moment Neural Network Package

GM-NN is an atomistic Neural Network (NN) built upon the Gaussian moments (GM) atomistic representation for the prediction of potential energy surfaces (PES). This software package manages the construction and training of atomistic NNs with GMs, a trainable, local, species-dependent atomic representation. Additionally, the GM-NN provides an uncertainty estimation for atomistic NNs defined in the framework of optimal experimental design (OED) or, equivalently, by treating the last layer of a trained network as a Bayesian linear regression model. Recently, the GM-NN has been extended to predict magnetic anisotropy tensors (MAT).

Contents

Getting Started

• Installation
• Parameters

Tutorials

• Learning Ethanol PES
• Learning Bulk TiO₂ PES
• Learning MAT
• Active Learning: Ethanol
• Transfer learning

Modules

• Data Pipeline
• Neighborlist
• Neural Network Layers
• Adam Optimizer
• PES Training
• MAT Training
• Active Learning
• Calculators

Indices and tables

• Index

• Module Index

• Search Page