mlsanityData to 14 Oct 2026

Improving the Stability of GNN Force Field Models by Reducing Feature Correlation

arXiv 2502.12548 ↗First posted 18 Feb 2025cs.LG cs.AIcited by 0

Not on the map. Why?

Abstract (arXiv)

Recently, Graph Neural Network based Force Field (GNNFF) models are widely used in Molecular Dynamics (MD) simulation, which is one of the most cost-effective means in semiconductor material research. However, even such models provide high accuracy in energy and force Mean Absolute Error (MAE) over trained (in-distribution) datasets, they often become unstable during long-time MD simulation when used for out-of-distribution datasets. In this paper, we propose a feature correlation based method for GNNFF models to enhance the stability of MD simulation. We reveal the negative relationship between feature correlation and the stability of GNNFF models, and design a loss function with a dynamic loss coefficient scheduler to reduce edge feature correlation that can be applied in general GNNFF training. We also propose an empirical metric to evaluate the stability in MD simulation. Experiments show our method can significantly improve stability for GNNFF models especially in out-of-distribution data with less than 3% computational overhead. For example, we can ensure the stable MD simulation time from 0.03ps to 10ps for Allegro model.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 15 of its 44 references that are in the map. First posted on arXiv 18 Feb 2025.

Builds on

15 of 44 references.

References · 15

  1. 2016SGDR: Stochastic Gradient Descent with Warm Restartscited by 5,045
  2. 2020Sharpness-Aware Minimization for Efficiently Improving Generalizationcited by 1,417
  3. 2016On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minimacited by 1,352
  4. 2021E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentialscited by 708
  5. 2017SchNet: A continuous-filter convolutional neural network for modeling quantum interactionscited by 571
  6. 2017Computing Nonvacuous Generalization Bounds for Deep (Stochastic) Neural Networks with Many More Parameters than Training Datacited by 533
  7. 2019Fantastic Generalization Measures and Where to Find Themcited by 513
  8. 2021GemNet: Universal Directional Graph Neural Networks for Moleculescited by 343
  9. 2022Learning Local Equivariant Representations for Large-Scale Atomistic Dynamicscited by 256
  10. 2022Forces are not Enough: Benchmark and Critical Evaluation for Machine Learning Force Fields with Molecular Simulationscited by 95
  11. 2022Simulate Time-integrated Coarse-grained Molecular Dynamics with Multi-Scale Graph Networkscited by 23
  12. 2022Feature Overcorrelation in Deep Graph Neural Networks: A New Perspectivecited by 19
3 more
  1. 2023EGraFFBench: Evaluation of Equivariant Graph Neural Network Force Fields for Atomistic Simulationscited by 15
  2. 2019Fast Neural Network Approach for Direct Covariant Forces Prediction in Complex Multi-Element Extended Systemscited by 12
  3. 2023Data efficiency and extrapolation trends in neural network interatomic potentialscited by 10

Built on by

Not cited yet.