Stability-Aware Boltzmann Estimator Training of Neural Network Interatomic Potentials

Explore a comprehensive lecture on Stability-Aware Boltzmann Estimator (StABlE) Training for Neural Network Interatomic Potentials. Delve into the challenges of using NNIPs for molecular dynamics simulations and discover how StABlE Training addresses stability issues. Learn about the multi-modal training procedure that combines supervised training from quantum-mechanical energies and forces with reference system observables. Understand the role of the Boltzmann Estimator in enabling efficient gradient computation and detecting instabilities. Examine the methodology's application across various systems, including organic molecules, tetrapeptides, and condensed phase systems. Gain insights into the significant improvements in simulation stability and recovery of structural and dynamic observables achieved by StABlE-trained models. Follow the lecture's structure, covering background information, the approach, StABlE Training details, results, discussion, and a Q&A session.

- Intro + Background
- Approach
- StABIE Training
- Results
- Discussion
- Q+A