mlsanityData to 14 Oct 2026

Implicit Delta Learning of High Fidelity Neural Network Potentials

arXiv 2412.06064 ↗First posted 8 Dec 2024physics.chem-ph cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Neural network potentials (NNPs) offer a fast and accurate alternative to ab-initio methods for molecular dynamics (MD) simulations but are hindered by the high cost of training data from high-fidelity Quantum Mechanics (QM) methods. Our work introduces the Implicit Delta Learning (IDLe) method, which reduces the need for high-fidelity QM data by leveraging cheaper semi-empirical QM computations without compromising NNP accuracy or inference cost. IDLe employs an end-to-end multi-task architecture with fidelity-specific heads that decode energies based on a shared latent representation of the input atomistic system. In various settings, IDLe achieves the same accuracy as single high-fidelity baselines while using up to 50x less high-fidelity data. This result could significantly reduce data generation cost and consequently enhance accuracy and generalization, and expand chemical coverage for NNPs, advancing MD simulations for material science and drug discovery. Additionally, we provide a novel set of 11 million semi-empirical QM calculations to support future multi-fidelity NNP modeling.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 25 of its 60 references that are in the map. First posted on arXiv 8 Dec 2024.

Builds on

25 of 60 references.

References · 25

  1. 2020Scaling Laws for Neural Language Modelscited by 6,586
  2. 2017Neural Message Passing for Quantum Chemistrycited by 3,487
  3. 2018UMAP: Uniform Manifold Approximation and Projection for Dimension Reductioncited by 2,898
  4. 2021Scaling Vision Transformerscited by 999
  5. 2021E(3)-Equivariant Graph Neural Networks for Data-Efficient and Accurate Interatomic Potentialscited by 708
  6. 2017Deep Learning Scaling is Predictable, Empiricallycited by 582
  7. 2021Explaining Neural Scaling Lawscited by 363
  8. 2021Scaling Laws for Transfercited by 272
  9. 2022Forces are not Enough: Benchmark and Critical Evaluation for Machine Learning Force Fields with Molecular Simulationscited by 95
  10. 2018Less is more: sampling chemical space with active learningcited by 82
  11. 2022SPICE, A Dataset of Drug-like Molecules and Peptides for Training Machine Learning Potentialscited by 82
  12. 2018Machine Learning of coarse-grained Molecular Dynamics Force Fieldscited by 69
13 more
  1. 2020OrbNet: Deep Learning for Quantum Chemistry Using Symmetry-Adapted Atomic-Orbital Featurescited by 64
  2. 2019DScribe: Library of Descriptors for Machine Learning in Materials Sciencecited by 63
  3. 2023From Molecules to Materials: Pre-training Large Generalizable Models for Atomic Property Predictioncited by 39
  4. 2018Active Learning of Uniformly Accurate Inter-atomic Potentials for Materials Simulationcited by 37
  5. 2023Scaling the leading accuracy of deep equivariant models to biomolecular simulations of realistic sizecited by 33
  6. 2021Learning neural network potentials from experimental data via Differentiable Trajectory Reweightingcited by 26
  7. 2022Hyperactive Learning (HAL) for Data-Driven Interatomic Potentialscited by 17
  8. 2024TorchMD-Net 2.0: Fast Neural Network Potentials for Molecular Simulationscited by 15
  9. 2023PubChemQC B3LYP/6-31G*//PM6 dataset: the Electronic Structures of 86 Million Molecules using B3LYP/6-31G* calculationscited by 12
  10. 2022Transfer learning for chemically accurate interatomic neural network potentialscited by 11
  11. 2022Scalable Bayesian Uncertainty Quantification for Neural Network Potentials: Promise and Pitfallscited by 11
  12. 2023Synthetic pre-training for neural-network interatomic potentialscited by 6
  13. 2024Transfer Learning for Molecular Property Predictions from Small Data Setscited by 2

Built on by

Not cited yet.