Skip to content

duqi-hnu/aml

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AML

Open in Visual Studio Code

Short Description

This is a Python package to automatically build the reference set for the training of Neural Network Potentials (NNPs), and eventually other machine-learned potentials, in an automated, data-driven fashion. For that purpose, a large set of reference configurations sampled in a physically meaningful way (typically with molecular dynamics) is filtered and the most important points for the representation of the Potential Energy Surface (PES) are identified. This is done by using a set of NNPs, called a committee, for error estimates of individual configurations. By iteratively adding the points with the largest error in the energy/force prediction, the reference set is progressively extended and optimized.

Keywords:

  • Active learning
  • Query by committee
  • Ensemble averaging
  • Committee machines
  • Neural Network Potentials

More information can be found in the following references:

  • C. Schran, F. L. Thiemann, P. Rowe, E. A. Müller, O. Marsalek, A. Michaelides,
    "Machine learning potentials for complex aqueous systems made simple",
    PNAS 118, e2110077118 (2021), 10.1073/pnas.2110077118
  • C. Schran, K. Brezina, O. Marsalek,
    "Committee neural network potentials control generalization errors and enable active learning",
    J. Chem. Phys. 153, 104105 (2020), 10.1063/5.0016004

Installation

For now, just clone the repository and source the env.sh file.

Dependencies:

Releases

No releases published

Packages

No packages published

Languages

  • Python 99.8%
  • Shell 0.2%