DP Compress: a Model Compression Scheme for Generating Efficient Deep Potential Models

• DOI: 10.48550/arxiv.2107.02103
• Publication Date: 2021-01-01
• Authors: Denghui Lu; Wanrun Jiang; Yixiao Chen; Linfeng Zhang; Weile Jia; Han Wang; Mohan Chen
• Keywords: Boosting

Machine-learning-based interatomic potential energy surface (PES) models are revolutionizing the field of molecular modeling. However, although much faster than electronic structure schemes, these suffer from costly computations via deep neural networks to predict and atomic forces, resulting in lower running efficiency as compared typical empirical force fields. Herein, we report a model compression scheme for boosting performance Deep Potential (DP) model, learning based PES model. This scheme, call DP Compress, is an efficient post-processing step after training (DP Train). Compress combines several DP-specific techniques, which typically speed up DP-based dynamics simulations by order magnitude faster, consume less memory. We demonstrate that sufficiently accurate testing variety physical properties Cu, H2O, Al-Cu-Mg systems. applies both CPU GPU machines publicly available online.