FLARE (Fast Learning of Atomistic Rare Events)

The code is intended to create machine learning interatomic potentials based on the Gaussian Process Technique. The potential then can be used in conjunction with MD engines such as implemented in LAMMPS or ASE. The Gaussian Process itself is a non-parametric, probabilistic model well-suited for regression tasks, which can provide a measure of uncertainty in its predictions, enabling an active learning scheme. In the active learning scheme, the model actively selects the most informative data points to learn from and thus efficiently find regions of the atomic configuration space where the model’s predictions are unreliable. For this task, the FLARE code can be paired with the first principles calculation code, whose interface is already available in ASE (e.g., VASP, or Quantum Espresso).