Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

FastCHGNet: Training one Universal Interatomic Potential to 1.5 Hours with 32 GPUs

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Training of the CHGNet universal interatomic potential can be reduced from 8.3 days to 1.53 hours on 32 GPUs without sacrificing accuracy, by fitting forces and stresses directly and optimizing GPU use.

desk verdict Solid HPC engineering with a real speedup, but the headline 'no accuracy sacrifice' is contradicted by the paper's own Table I; needs honest reframing. read the letter →

arxiv 2412.20796 v2 pith:MF7HM4HS submitted 2024-12-30 cs.DC cs.LG

Yuanchang Zhou , Siyu Hu , Chen Wang , Lin-Wang Wang , Guangming Tan , Weile Jia This is my paper · ORCID
classification cs.DCcs.LG
keywords graphneuralnetworkuniversalinteratomicpotentialmoleculardynamicsGPUaccelerationtrainingspeedupforcereadoutloadbalancingkernelfusion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that the dominant training bottleneck of CHGNet—a graph neural network universal interatomic potential pretrained on the Materials Project trajectory dataset—can be removed. By decoupling force and stress prediction from the energy through dedicated equivariant readout heads, and by applying GPU-centric optimizations (kernel fusion, parallel basis computation, load-balanced multi-GPU training), the authors reduce training time from 8.3 days on one A100 GPU to 1.53 hours on 32 A100 GPUs, a 130x speedup, while keeping test-set accuracy at or slightly better than the original model. The version without the decoupled heads trains in 3.79 hours with a small accuracy gain. If correct, this makes iterating on universal interatomic potentials practical for groups with modest GPU clusters.

What carries the argument

The central object is the decoupled output layer: a Force Head defined as $F_i = \sum_j (n_{ij} \odot x_{ij})$ with $n_{ij} = \mathrm{MLP}(e^t_{ij})$ and $x_{ij}$ the bond vector, and a Stress Head defined via outer products of normalized lattice vectors combined with final atom features. The Force Head is rotation-equivariant because bond features are invariant and bond vectors rotate with the system; the heads remove the computational overhead of second-order derivatives in the backward pass. The second key mechanism is breaking the sequential dependency inside the interaction blocks—atom convolution, bond convolution, and angle update can all be computed from the same input features—so the forward pass is parallelized. On the system side, a load-balancing sampler assigns the largest and smallest graphs to the same GPU, reducing workload variance from 0.186 to 0.064.

What would settle it

Run the trained FastCHGNet with decoupled heads in an NVE molecular dynamics simulation on a lithium-containing system and measure total energy drift over hundreds of picoseconds; systematic drift or instability would show the model is not a physics-safe drop-in for CHGNet. A sharper check is to compute the gradient of the predicted energy with respect to atomic positions and compare it to the negative of the predicted forces: if they differ, the model is not a conservative force field.

Watch

Extended reading notes

Core claim

The paper's central discovery is that forces and stresses do not need to be computed as derivatives of a learned energy in a universal interatomic potential. FastCHGNet instead fits them with separate readout heads: a Force Head that sums learned per-bond magnitudes times bond vectors (provably rotation-equivariant) and a Stress Head built from lattice-vector outer products and final atom features. This eliminates the need to store and compute second-order derivatives during training, cutting memory by up to 3.59x and adding roughly a 2x speedup. Combined with system-level optimizations such as parallel graph-basis construction, kernel fusion, redundancy removal, a load-balancing sampler, and communication overlap, the model scales to 32 GPUs at 66% efficiency in strong scaling and completes a full training run in 1.53 hours.

Load-bearing premise

The load-bearing premise is that fitting forces and stresses directly—instead of deriving them from a learned energy—still yields a model that behaves correctly in real molecular dynamics, not just in aggregate test-set error.

Editorial extensions

If this is right

  • Training a state-of-the-art GNN-UIP from scratch drops from over a week to under two hours on a 32-GPU cluster, enabling rapid iteration on architectures and hyperparameters.
  • The decoupled Force/Stress readout strategy can be transferred to other energy-based GNN interatomic potentials to reduce their memory footprint and training time.
  • The load-balancing sampler and parallel basis computation are general techniques for scaling graph-neural-network training on material datasets with long-tail size distributions.
  • The up-to-3.59x memory reduction makes larger minibatches feasible on a single GPU, benefiting groups without multi-GPU resources.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If direct force fitting proves stable in long molecular dynamics runs, energy labels might become unnecessary for future training, letting models learn purely from the much more abundant force and stress data in trajectory datasets.
  • The rotation-equivariance proof given for the Force Head does not cover reflections; the Stress Head's construction from outer products of normalized lattice vectors may break equivariance under improper rotations, which could matter for chiral or layered systems.
  • The reported 130x speedup was measured on A100 GPUs with PyTorch 2.3.1; the algorithmic gains should transfer to other hardware, while the kernel-fusion and communication-overlap components may shrink on newer GPU generations.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper presents FastCHGNet, an optimized implementation of CHGNet, a graph neural network universal interatomic potential. The authors introduce force/stress readout heads that predict forces and stresses directly rather than via energy derivatives, along with system-level optimizations including kernel fusion, redundancy elimination, parallel basis computation, a load-balanced multi-GPU sampler, and a batch-size-scaled learning rate. They report that FastCHGNet reduces training time from 8.3 days on one A100 GPU to 1.53 hours on 32 A100 GPUs, claiming this is achieved without sacrificing model accuracy. The paper also reports memory-footprint reduction, scaling studies on up to 32 GPUs, MAE comparisons on the MPtrj test set, and MD inference speedups on three lithium-based systems.

Significance. If the central claim were supported, this would be a substantial practical contribution, as it would remove the training bottleneck for a widely used universal interatomic potential and enable rapid iteration on model architectures. The engineering optimizations are described in detail and the reported iteration-time and memory gains are plausible. The rotation-equivariance proof for the force head (Eq. 8) is a correct and useful formal check. However, the headline accuracy claim is internally contradicted by the paper's own Table I and Conclusion, and the physical usability of the decoupled force/stress heads for molecular dynamics is not validated beyond aggregate MAE values. The corrected contribution -- a 3.79-hour training run without accuracy sacrifice and a 1.53-hour run with a quantifiable accuracy tradeoff -- is still interesting but is not what the abstract and introduction claim.

major comments (4)
  1. [Abstract; Section I; Table I; Section VII] The central claim that the 1.53-hour training run is achieved 'without sacrificing model accuracy' is contradicted by the paper's own data. Table I reports that the F/S-head variant used for the 1.53-hour run has Force MAE 73 meV/Å and Stress MAE 0.479 GPa, versus 68 meV/Å and 0.314 GPa for CHGNet, i.e., 7% worse force and 53% worse stress MAE. The Conclusion explicitly states: 'Without sacrificing accuracy, the training time of FastCHGNet(without force/stress decoupling) can reduce to 3.79 hours.' Thus the 1.53-hour result is not a no-sacrifice result, and the abstract's 'without sacrificing model accuracy' and Section I's 'With no sacrifice of accuracy ... can be reduced to 1.53 hours' are unsupported. This is a load-bearing inconsistency that must be corrected by reframing the contribution or by providing a 1.53-hour version that actually preserves CHGNet's accuracy.
  2. [Section V-A, Table I] The parity claim for the w/o-head variant (which is the basis for the 3.79-hour no-sacrifice claim) rests on single-run MAE values without any statistical uncertainty. The observed differences (e.g., Force 62 vs 68 meV/Å, Stress 0.270 vs 0.314 GPa) are in the expected direction, but with no error bars, repeated runs, or significance tests, it is not established that these differences are meaningful. This is load-bearing because the no-sacrifice claim depends on the w/o-head variant being at least as accurate as CHGNet, not merely comparable in one run.
  3. [Section V-D, Table II] The load-bearing assumption that the decoupled Force and Stress heads produce forces usable for molecular dynamics is not tested. Table II reports only one-step inference time on three Li-based systems; there is no evidence of trajectory stability, energy conservation, or long-timescale behavior. A model that predicts forces directly without deriving them from a single energy surface may not conserve energy or satisfy Newton's third law, so the claim that the 1.53-hour F/S-head model is a drop-in replacement for CHGNet in MD is unsupported. The authors should validate the F/S-head model in at least short NVE and NVT simulations, reporting energy drift and trajectory quality.
  4. [Section III-B (Dependency Elimination)] The claim that breaking the dependency between bond convolution and angle update 'does not affect accuracy' (Eq. 11 vs Eq. 10) is not supported by any ablation. The comparison of FastCHGNet w/o head against CHGNet in Table I confounds dependency elimination with other changes, such as larger batch size, modified learning-rate schedule, and kernel fusion. A controlled experiment isolating the dependency-elimination change is needed to justify this design choice, which is presented as a key model innovation.
minor comments (5)
  1. [Section IV and Section V-A, Fig. 6] Section IV states 'The initial learning rate is 0.0003', but the text describing Fig. 6 says 'the default learning rate (0.003)'. This apparent factor-of-ten discrepancy should be resolved.
  2. [Table I] The Magmom column header reads 'Magmom( mµB)' which mixes an SI prefix with a Greek mu; it should be 'μB' for consistency.
  3. [Section V-C, Fig. 10(b)] The weak-scaling sentence reports 'scaling efficiencies for 4, 8, 16, and 32 GPUs are 91.5%, 84.6%, and 74.6%, respectively' but lists only three values for four GPU counts; the 4-GPU baseline efficiency (presumably 100%) is missing.
  4. [Section III-D (Learning Rate Schedule)] Equation (14) introduces a free hyperparameter k, and the paper only reports results for k=128 and one global batch size (2048). The sensitivity of convergence to k and the comparison with standard scaling rules (e.g., square-root scaling) should be discussed.
  5. [Algorithm 2] The construction of a block-diagonal matrix for neighbor images (line 11) may incur significant memory overhead; the paper should comment on the memory cost of this 'Parallel Computation of Basis' design, especially for very large batches.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: training-time results are measured outcomes; the equivariance proof is a direct algebraic consequence; no fitted quantity is presented as a prediction.

full rationale

No circular step was found. FastCHGNet's central claims are measured engineering outcomes (training time, memory footprint, kernel count, scaling efficiency, inference speed), obtained from concrete system optimizations rather than derived from the claims themselves. The rotation-equivariance proof for the Force Head is a direct algebraic consequence of invariant bond features and the linear action of rotation matrices; it does not assume the conclusion. The Force and Stress heads are fit to MPtrj labels and then evaluated on a held-out test set, so Table I is a standard supervised evaluation, not a fitted parameter renamed as a prediction. The learning-rate heuristic is presented as a tuning choice with convergence curves, not as a model output. The citations to the authors' own prior work appear only in the Related Work discussion of optimizers and are not load-bearing for the 1.53-hour or 3.79-hour training-time claims. One internal tension does exist: the abstract's statement that the 1.53-hour run is achieved 'without sacrificing model accuracy' is not supported by Table I, where the F/S-head variant used for that run has worse Force MAE (73 vs 68 meV/A) and Stress MAE (0.479 vs 0.314 GPa) than CHGNet, and the conclusion itself attributes the no-sacrifice result to the 3.79-hour w/o-head variant. That is an accuracy-claim consistency problem, not a circularity problem, because the reported numbers are independent measurements rather than consequences of the paper's assumptions. Accordingly, the circularity score is 0.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The speedup claim depends on a small set of tuned hyperparameters (learning rate schedule, loss weights, batch size). The physically consequential new ingredient is the force/stress decoupling, whose MD usability is assumed rather than demonstrated.

free parameters (4)
  • learning rate scaling constant k = 128
    Eq 14 sets initLR = batchsize/k * 0.0003 with k=128. This constant is chosen by hand; no sensitivity analysis is shown.
  • multi-task loss prefactors = energy 2, force 1.5, stress 0.1, magmom 0.1
    Section IV specifies these weights for the Huber loss. They directly control the accuracy tradeoff among energy, force, stress, and magmom, and are not derived.
  • initial learning rate = 0.0003
    Section IV states the initial learning rate is 0.0003 with cosine annealing; after Eq 14 with batchsize 2048 the effective rate becomes 0.0048, but Figure 6 labels the default as 0.003, an inconsistency.
  • global batch size = 2048 for multi-GPU runs
    Section V-C uses global batch 2048 to enable large-batch training and the load-balancing sampler; Section IV says batch size 128. The mismatch is not explained.
assumptions (4)
  • domain assumption MPtrj labels (energies, forces, stresses, magnetic moments) are treated as exact DFT ground truth.
    The entire accuracy evaluation compares predictions against these labels without questioning DFT accuracy or dataset noise.
  • domain assumption The 0.9/0.05/0.05 train/validation/test split and the single-run MAE comparison against pretrained CHGNet v0.3.0 constitute a valid accuracy benchmark.
    No repeated splits or statistical tests are reported, yet the paper draws conclusions about accuracy parity from this comparison.
  • standard math Bond features e^t_ij are invariant under rotation, which underpins the force-head equivariance proof in Eq 8.
    The proof that F_i transforms as R F_i relies on this invariance; it holds by construction of the CHGNet features, but is stated without proof in the paper.
  • ad hoc to paper Direct force/stress prediction without energy consistency yields forces usable for molecular dynamics.
    The paper validates only inference speed on three Li-containing systems (Table II), not trajectory accuracy, energy conservation, or long-timescale stability. This is an unvalidated premise behind the claim that accuracy is not sacrificed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of FastCHGNet: Training one Universal Interatomic Potential to 1.5 Hours with 32 GPUs." pith.science (2026). https://pith.science/paper/MF7HM4HS

@misc{pith2026241220796,
  author       = {Pith},
  title        = {Pith review of: FastCHGNet: Training one Universal Interatomic Potential to 1.5 Hours with 32 GPUs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MF7HM4HS}},
  note         = {Machine review of arXiv:2412.20796}
}
read the original abstract

Graph neural network universal interatomic potentials (GNN-UIPs) have demonstrated remarkable generalization and transfer capabilities in material discovery and property prediction. These models can accelerate molecular dynamics (MD) simulation by several orders of magnitude while maintaining \textit{ab initio} accuracy, making them a promising new paradigm in material simulations. One notable example is Crystal Hamiltonian Graph Neural Network (CHGNet), pretrained on the energies, forces, stresses, and magnetic moments from the MPtrj dataset, representing a state-of-the-art GNN-UIP model for charge-informed MD simulations. However, training the CHGNet model is time-consuming(8.3 days on one A100 GPU) for three reasons: (i) requiring multi-layer propagation to reach more distant atom information, (ii) requiring second-order derivatives calculation to finish weights updating and (iii) the implementation of reference CHGNet does not fully leverage the computational capabilities. This paper introduces FastCHGNet, an optimized CHGNet, with three contributions: Firstly, we design innovative Force/Stress Readout modules to decompose Force/Stress prediction. Secondly, we adopt massive optimizations such as kernel fusion, redundancy bypass, etc, to exploit GPU computation power sufficiently. Finally, we extend CHGNet to support multiple GPUs and propose a load-balancing technique to enhance GPU utilization. Numerical results show that FastCHGNet reduces memory footprint by a factor of 3.59. The final training time of FastCHGNet can be decreased to \textbf{1.53 hours} on 32 GPUs without sacrificing model accuracy.

Figures

Figures reproduced from arXiv: 2412.20796 by the authors.

Figure 1
Figure 1. The CHGNet workflow. (a) The input and output of CHGNet; (b)The graph representation of crystal structure; (c) The [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. The architecture of FastCHGNet. (a) the high-level workflow of FastCHGNet; (b) the detailed implementation of [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Packing strategy. (a) Sharing the same input can [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Load balance sampler. The strategy for distributing [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: The convergence of fine-tuned learning rate in terms of [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: CHGNet and FastCHGNet performance contrasted with [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 10
Figure 10. Figure 10: The strong scaling and weak scaling of FastCHGNet [PITH_FULL_IMAGE:figures/full_fig_p009_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Facet: highly efficient E(3)-equivariant networks for interatomic potentials

    cond-mat.mtrl-sci 2025-09 conditional novelty 6.0 of 10

    Facet trains an E(3)-equivariant interatomic potential on MPTrj with accuracy close to SevenNet and MACE while using under 10% of the training compute, via spline radial filters and an S2-MLP-Mixer node update.

Reference graph

Works this paper leans on

41 extracted references · 33 canonical work pages · cited by 1 Pith paper

  1. [1]

    Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations,

    Y . Park, J. Kim, S. Hwang, and S. Han, “Scalable parallel algorithm for graph neural network interatomic potentials in molecular dynamics simulations,” Journal of Chemical Theory and Computation, 2024

    work page 2024

  2. [2]

    A foundation model for atomistic materials chemistry,

    I. Batatia, P. Benner, Y . Chiang, A. M. Elena, D. P. Kov´acs, J. Riebesell, X. R. Advincula, M. Asta, W. J. Baldwin, N. Bernstein et al., “A foundation model for atomistic materials chemistry,” arXiv preprint arXiv:2401.00096, 2023

    arXiv 2023

  3. [3]

    Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling,

    B. Deng, P. Zhong, K. Jun, J. Riebesell, K. Han, C. J. Bartel, and G. Ceder, “Chgnet as a pretrained universal neural network potential for charge-informed atomistic modelling,” Nature Machine Intelligence, vol. 5, no. 9, pp. 1031–1041, 2023

    work page 2023

  4. [4]

    Deepmd-kit: A deep learning package for many-body potential energy representation and molecular dynamics,

    H. Wang, L. Zhang, J. Han, and E. Weinan, “Deepmd-kit: A deep learning package for many-body potential energy representation and molecular dynamics,” Computer Physics Communications, vol. 228, pp. 178–184, 2018

    work page 2018

  5. [5]

    Quantum-chemical insights from deep tensor neural networks,

    K. T. Sch ¨utt, F. Arbabzadah, S. Chmiela, K. R. M ¨uller, and A. Tkatchenko, “Quantum-chemical insights from deep tensor neural networks,” Nature communications, vol. 8, no. 1, p. 13890, 2017

    work page 2017

  6. [6]

    Schnet: A continuous-filter convo- lutional neural network for modeling quantum interactions,

    K. Sch ¨utt, P.-J. Kindermans, H. E. Sauceda Felix, S. Chmiela, A. Tkatchenko, and K.-R. M ¨uller, “Schnet: A continuous-filter convo- lutional neural network for modeling quantum interactions,” Advances in neural information processing systems, vol. 30, 2017

    work page 2017

  7. [7]

    Hierarchical modeling of molecular energies using a deep neural network,

    N. Lubbers, J. S. Smith, and K. Barros, “Hierarchical modeling of molecular energies using a deep neural network,” The Journal of chemical physics, vol. 148, no. 24, 2018

    work page 2018

  8. [8]

    Physnet: A neural network for predicting energies, forces, dipole moments, and partial charges,

    O. T. Unke and M. Meuwly, “Physnet: A neural network for predicting energies, forces, dipole moments, and partial charges,” Journal of chemical theory and computation, vol. 15, no. 6, pp. 3678–3693, 2019

    work page 2019

Show all 41 references
  1. [9]

    Directional message pass- ing for molecular graphs,

    J. Gasteiger, J. Groß, and S. G ¨unnemann, “Directional message pass- ing for molecular graphs,” in International Conference on Learning Representations (ICLR), 2020

    2020

  2. [10]

    Mace: Higher order equivariant message passing neural networks for fast and accurate force fields,

    I. Batatia, D. P. Kovacs, G. Simm, C. Ortner, and G. Cs ´anyi, “Mace: Higher order equivariant message passing neural networks for fast and accurate force fields,” Advances in Neural Information Processing Systems, vol. 35, pp. 11 423–11 436, 2022

    2022

  3. [11]

    Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties,

    T. Xie and J. C. Grossman, “Crystal graph convolutional neural networks for an accurate and interpretable prediction of material properties,” Physical review letters, vol. 120, no. 14, p. 145301, 2018

    2018

  4. [12]

    Gemnet: Universal di- rectional graph neural networks for molecules,

    J. Gasteiger, F. Becker, and S. G ¨unnemann, “Gemnet: Universal di- rectional graph neural networks for molecules,” Advances in Neural Information Processing Systems, vol. 34, pp. 6790–6802, 2021

    2021

  5. [13]

    Ani-1: an extensible neural network potential with dft accuracy at force field computational cost,

    J. S. Smith, O. Isayev, and A. E. Roitberg, “Ani-1: an extensible neural network potential with dft accuracy at force field computational cost,” Chemical science, vol. 8, no. 4, pp. 3192–3203, 2017

    2017

  6. [14]

    Extending the applicability of the ani deep learning molecular potential to sulfur and halogens,

    C. Devereux, J. S. Smith, K. K. Huddleston, K. Barros, R. Zubatyuk, O. Isayev, and A. E. Roitberg, “Extending the applicability of the ani deep learning molecular potential to sulfur and halogens,” Journal of Chemical Theory and Computation, vol. 16, no. 7, pp. 4192–4202, 2020

    2020

  7. [15]

    Generalized neural-network representation of high-dimensional potential-energy surfaces,

    J. Behler and M. Parrinello, “Generalized neural-network representation of high-dimensional potential-energy surfaces,” Physical review letters, vol. 98, no. 14, p. 146401, 2007

    2007

  8. [16]

    Embedded atom neural network poten- tials: Efficient and accurate machine learning with a physically inspired representation,

    Y . Zhang, C. Hu, and B. Jiang, “Embedded atom neural network poten- tials: Efficient and accurate machine learning with a physically inspired representation,” The journal of physical chemistry letters, vol. 10, no. 17, pp. 4962–4967, 2019

    2019

  9. [17]

    Equivariant message passing for the prediction of tensorial properties and molecular spectra,

    K. Sch ¨utt, O. Unke, and M. Gastegger, “Equivariant message passing for the prediction of tensorial properties and molecular spectra,” in International Conference on Machine Learning. PMLR, 2021, pp. 9377–9388

    2021

  10. [18]

    E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials,

    S. Batzner, A. Musaelian, L. Sun, M. Geiger, J. P. Mailoa, M. Kornbluth, N. Molinari, T. E. Smidt, and B. Kozinsky, “E (3)-equivariant graph neural networks for data-efficient and accurate interatomic potentials,” Nature communications, vol. 13, no. 1, p. 2453, 2022

    2022

  11. [19]

    Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces,

    M. Haghighatlari, J. Li, X. Guan, O. Zhang, A. Das, C. J. Stein, F. Heidar-Zadeh, M. Liu, M. Head-Gordon, L. Bertelset al., “Newtonnet: a newtonian message passing network for deep learning of interatomic potentials and forces,” Digital Discovery, vol. 1, no. 3, pp. 333–343, 2022

    2022

  12. [20]

    Spookynet: Learning force fields with electronic degrees of freedom and nonlocal effects,

    O. T. Unke, S. Chmiela, M. Gastegger, K. T. Sch ¨utt, H. E. Sauceda, and K.-R. M ¨uller, “Spookynet: Learning force fields with electronic degrees of freedom and nonlocal effects,” Nature communications, vol. 12, no. 1, p. 7273, 2021

    2021

  13. [21]

    Fast and uncertainty-aware directional message passing for non-equilibrium molecules,

    J. Gasteiger, S. Giri, J. T. Margraf, and S. G ¨unnemann, “Fast and uncertainty-aware directional message passing for non-equilibrium molecules,” arXiv preprint arXiv:2011.14115, 2020

    2011 arXiv

  14. [22]

    A universal graph deep learning interatomic potential for the periodic table,

    C. Chen and S. P. Ong, “A universal graph deep learning interatomic potential for the periodic table,” Nature Computational Science, vol. 2, no. 11, pp. 718–728, 2022

    2022

  15. [23]

    Atomistic line graph neural network for improved materials property predictions. npj computational materials, 7 (1): 185,

    K. Choudhary and B. DeCost, “Atomistic line graph neural network for improved materials property predictions. npj computational materials, 7 (1): 185,” 2021

    2021

  16. [24]

    Graph networks as a universal machine learning framework for molecules and crystals,

    C. Chen, W. Ye, Y . Zuo, C. Zheng, and S. P. Ong, “Graph networks as a universal machine learning framework for molecules and crystals,” Chemistry of Materials, vol. 31, no. 9, pp. 3564–3572, 2019

    2019

  17. [25]

    Gptff: A high-accuracy out-of-the- box universal ai force field for arbitrary inorganic materials,

    F. Xie, T. Lu, S. Meng, and M. Liu, “Gptff: A high-accuracy out-of-the- box universal ai force field for arbitrary inorganic materials,” Science Bulletin, 2024

    2024

  18. [26]

    A survey on compiler autotuning using machine learning,

    A. H. Ashouri, W. Killian, J. Cavazos, G. Palermo, and C. Silvano, “A survey on compiler autotuning using machine learning,” ACM Computing Surveys (CSUR), vol. 51, no. 5, pp. 1–42, 2018

    2018

  19. [27]

    and y. chen. 2018. tvm: An automated end-to-end optimizing compiler for deep learning,

    T. Chen, T. Moreau, Z. Jiang, L. Zheng, E. Yan, and H. Shen, “and y. chen. 2018. tvm: An automated end-to-end optimizing compiler for deep learning,” in 13th USENIX Symposium on Operating Systems Design and Implementation (OSDI 18), 2018, pp. 578–594

    2018

  20. [28]

    Cdnet: A real-time and robust crosswalk detection network on jetson nano based on yolov5,

    Z.-D. Zhang, M.-L. Tan, Z.-C. Lan, H.-C. Liu, L. Pei, and W.-X. Yu, “Cdnet: A real-time and robust crosswalk detection network on jetson nano based on yolov5,” Neural Computing and Applications, vol. 34, no. 13, pp. 10 719–10 730, 2022

    2022

  21. [29]

    Ansor: Generating {High-Performance} tensor programs for deep learning,

    L. Zheng, C. Jia, M. Sun, Z. Wu, C. H. Yu, A. Haj-Ali, Y . Wang, J. Yang, D. Zhuo, K. Sen et al., “Ansor: Generating {High-Performance} tensor programs for deep learning,” in 14th USENIX symposium on operating systems design and implementation (OSDI 20), 2020, pp. 863–879

    2020

  22. [30]

    Operator fusion in xla: Analysis and evalua- tion,

    D. Snider and R. Liang, “Operator fusion in xla: Analysis and evalua- tion,” arXiv preprint arXiv:2301.13062, 2023

    2023 arXiv

  23. [31]

    Pytorch: An imperative style, high-performance deep learning library,

    A. Paszke, S. Gross, F. Massa, A. Lerer, J. Bradbury, G. Chanan, T. Killeen, Z. Lin, N. Gimelshein, L. Antiga et al., “Pytorch: An imperative style, high-performance deep learning library,” Advances in neural information processing systems, vol. 32, 2019

    2019

  24. [32]

    Degree-quant: Quantization-aware training for graph neural networks,

    S. A. Tailor, J. Fernandez-Marques, and N. D. Lane, “Degree-quant: Quantization-aware training for graph neural networks,” arXiv preprint arXiv:2008.05000, 2020

    2008 arXiv

  25. [33]

    Qgtc: accelerating quantized graph neural networks via gpu tensor core,

    Y . Wang, B. Feng, and Y . Ding, “Qgtc: accelerating quantized graph neural networks via gpu tensor core,” in Proceedings of the 27th ACM SIGPLAN symposium on principles and practice of parallel programming, 2022, pp. 107–119

    2022

  26. [34]

    Vq-gnn: A universal framework to scale up graph neural networks using vector quantization,

    M. Ding, K. Kong, J. Li, C. Zhu, J. Dickerson, F. Huang, and T. Gold- stein, “Vq-gnn: A universal framework to scale up graph neural networks using vector quantization,” Advances in Neural Information Processing Systems, vol. 34, pp. 6733–6746, 2021

    2021

  27. [35]

    Sgquant: Squeezing the last bit on graph neural networks with specialized quantization,

    B. Feng, Y . Wang, X. Li, S. Yang, X. Peng, and Y . Ding, “Sgquant: Squeezing the last bit on graph neural networks with specialized quantization,” in 2020 IEEE 32nd international conference on tools with artificial intelligence (ICTAI). IEEE, 2020, pp. 1044–1052

    2020

  28. [36]

    Rlekf: an optimizer for deep potential with ab initio accuracy,

    S. Hu, W. Zhang, Q. Sha, F. Pan, L.-W. Wang, W. Jia, G. Tan, and T. Zhao, “Rlekf: an optimizer for deep potential with ab initio accuracy,” in Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence and Thirty-Fifth Conference on Innovative Applications of...

    2023 doi

  29. [37]

    Neural network force field training based on reor- ganized layer-wised extended kalman filter,

    H. Si-Yu, Z. Yuan-Chang, Z. Tong, W. Lin-Wang, J. Wei-Le, and T. Guang-Ming, “Neural network force field training based on reor- ganized layer-wised extended kalman filter,” Journal of Software, pp. 1–17

  30. [38]

    The mlip package: moment tensor potentials with mpi and active learning,

    I. S. Novikov, K. Gubaev, E. V . Podryabinkin, and A. V . Shapeev, “The mlip package: moment tensor potentials with mpi and active learning,” Machine Learning: Science and Technology, vol. 2, no. 2, p. 025002, dec

  31. [39]

    Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials,

    A. Thompson, L. Swiler, C. Trott, S. Foiles, and G. Tucker, “Spectral neighbor analysis method for automated generation of quantum-accurate interatomic potentials,” Journal of Computational Physics, vol. 285, pp. 316–330, 2015. [Online]. Available: https: //www.sciencedirect.c...

    2015

  32. [40]

    Training one deepmd model in minutes: a step towards online learning,

    S. Hu, T. Zhao, Q. Sha, E. Li, X. Meng, L. Liu, L.-W. Wang, G. Tan, and W. Jia, “Training one deepmd model in minutes: a step towards online learning,” in Proceedings of the 29th ACM SIGPLAN Annual Symposium on Principles and Practice of Parallel Programming, ser. PPoPP ’24. N...

    2024

  33. [2020]

    Available: https://dx.doi.org/10.1088/2632-2153/abc9fe

    [Online]. Available: https://dx.doi.org/10.1088/2632-2153/abc9fe

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.