Pith. sign in

REVIEW 3 major objections 5 minor 21 references

Transfer learning in Scalable Graph Neural Network for Improved Physical Simulation

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read Pre-training on simulated CAD shapes lets graph-network physics simulators fine-tune from a fraction of the usual training data, with an 11% RMSE gain on a deformable-plate benchmark.

desk verdict A useful new architecture and dataset, but the transfer gains are confounded by an asymmetric regularization term that needs a control before the central claim holds. read the letter →

arxiv 2502.06848 v1 pith:U7EWVY2Q submitted 2025-02-07 cs.LG cs.AI

Siqi Shen , Yu Liu , Daniel Biggs , Omar Hafez , Jiandong Yu , Wentao Zhang , Bin Cui , Jiulong Shan This is my paper · ORCID
classification cs.LGcs.AI
keywords graphneuralnetworksphysicalsimulationtransferlearningmesh-basedU-netpre-trainingfiniteelementanalysisdeformablebodies
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

This paper tries to make transfer learning work for graph-network physics simulators, which until now were trained from scratch for every new simulation task. The paper proposes a scalable graph U-net (SGUNET) whose depth-first-search pooling lets one architecture be reconfigured for different mesh sizes, together with mapping functions that copy or average pre-trained weights into a differently sized target network and a regularization term that keeps fine-tuned weights close to the pre-trained ones. To have something to pre-train on, the paper built a dataset of 20,000 finite-element simulations of random 3D CAD shapes pressed into contact. The central result is that fine-tuning this pre-trained model on a fraction of a benchmark's training data can beat training from scratch on the same fraction: on the 2D Deformable Plate benchmark, fine-tuning on 1/16 of the data improves position RMSE by 11.05%, and on the 3D Deforming Plate benchmark the same accuracy is reached in about 40% of the training time. If this holds, GNN simulators could stop requiring large, expensive FEA-generated datasets for every new task.

What carries the argument

The load-bearing mechanism is the combination of a configurable architecture and two parameter-alignment functions. DFS pooling performs a depth-first random walk over the element-node graph, clustering nearby nodes of the same material into pooled super-nodes with averaging, and the pooling ratios per stage can be changed so the same SGUNET can be built for different mesh resolutions. The Uniform mapping divides or averages the pre-trained Graph-Net blocks and GUnet stages to fit a target model of different size; the First-N mapping copies only the first shared blocks and randomly initializes the rest. During fine-tuning, the Frobenius norm between pre-trained and fine-tuned weights is added to the task loss as a regularizer. These pieces together let a checkpoint trained on ABCD be loaded into models with different architecture configurations, which is what makes transfer learning possible.

What would settle it

Run the same fine-tuning protocol with the ABCD pre-training outputs replaced by randomized or non-physical displacement fields; if the RMSE gains over from-scratch training persist, the improvement is not caused by transferred physics knowledge but by the fine-tuning procedure itself.

Watch

Extended reading notes

Core claim

The paper's central claim is that pre-training can be decoupled from task-specific architecture choices in graph-network simulators. SGUNET is an Encoder-Processor-Decoder network with a staged U-Net whose depth-first-search pooling clusters nearby element nodes into pooled nodes, making the receptive field configurable through pooling ratios and message-passing counts. Because downstream simulators may need different numbers of stages and blocks, the paper defines two weight-mapping functions, Uniform and First-N, that respectively spread pre-trained weights across the target network or copy only the earliest shared blocks, and it regularizes fine-tuning with a Frobenius-norm penalty against the pre-trained weights. Pre-trained on ABCD, a dataset of 20,000 simulations of randomly paired 3D CAD shapes deformed by contact, the model is then fine-tuned on two public benchmarks. The reported results are that on the 2D Deformable Plate benchmark, SGUNET fine-tuned on 1/16 of the training data improves position RMSE by 11.05% over training from scratch on the same fraction; on the 3D Deforming Plate benchmark, the pre-trained model reaches the same performance as the from-scratch model with 1/8 of the data in roughly 40% of the training time; and the same transfer protocol also improves the MGN baseline.

Load-bearing premise

The central bet is that pre-training on ABCD—random 3D CAD shapes pressed into contact—teaches a representation that transfers to the specific 2D and 3D plate benchmarks, and the paper does not analyze how similar those domains are.

Editorial extensions

If this is right

  • Pre-trained and fine-tuned SGUNET models reach lower rollout validation RMSE than their from-scratch counterparts across every data scale tested, and the gap grows as the training set shrinks.
  • Fine-tuning on 1/16 of the Deformable Plate training set yields an 11.05% improvement in position RMSE over from-scratch training, putting the model close to the full-data fine-tuned performance.
  • On the 3D Deforming Plate benchmark, fine-tuning with 1/8 of the data reaches the from-scratch accuracy in roughly 40% of the training steps.
  • Uniform mapping beats First-N mapping in the reported experiments, indicating that distributing pre-trained weights over the whole target network is more effective than keeping only the first blocks.
  • The same transfer protocol improves MGN, the baseline optimized for the target datasets, so the benefit is not tied exclusively to SGUNET.

Reading between the lines

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

  • Beyond the paper: the reported gains are consistent with the pre-trained checkpoint acting mainly as a strong regularizer; the paper does not measure domain similarity between ABCD and the plate benchmarks, so the learned-physics explanation is not the only one consistent with the data.
  • Beyond the paper: if the transfer is genuinely about contact-deformation priors, the same ABCD checkpoint should also speed up other quasi-static solid-mechanics tasks with different loads or materials, which would be a direct test of the paradigm.
  • Beyond the paper: the mapping-function recipe is architecture-agnostic and could be applied to other encoder-processor-decoder mesh simulators, including fluid or cloth models, provided their graph construction and pooling are compatible.
  • Beyond the paper: a cheaper way to test the mechanism is to pre-train on ABCD with randomized or non-physical targets; if fine-tuning gains persist, the improvement comes from the optimization procedure rather than from transferred physics.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper introduces SGUNET, a scalable graph U-Net with depth-first-search (DFS) pooling for mesh-based physical simulation, together with a transfer-learning scheme for GNN simulators. It also contributes ABCD, a pre-training dataset of 20,000 FEA simulations derived from the ABC CAD dataset. The proposed method maps parameters between pre-trained and fine-tuned models of different sizes using Uniform and First-N mapping functions, and adds a Frobenius-norm regularization term to the loss. Experiments on the 2D Deformable Plate and 3D Deforming Plate benchmarks claim that fine-tuning with small fractions of the training data improves position RMSE over from-scratch training, including an 11.05% improvement on 1/16 of the Deformable Plate data, and reduces training time on the 3D benchmark.

Significance. If the claims are established, this would be a useful practical contribution to a relatively unexplored area: pre-training and fine-tuning for graph-network-based physics simulators. The SGUNET architecture with variable pooling ratios and the ABCD dataset are potentially reusable resources. The work also gives a concrete recipe for parameter mapping between differently sized GNNs. However, the current experimental design does not isolate the effect of pre-training from the effect of the added regularization term, and some of the headline numerical claims rest on comparisons with overlapping error bars. These issues need to be addressed before the central transfer-learning claim is convincing.

major comments (3)
  1. [§3.4.2, Appendix C.2] The fine-tuned models minimize L_reg = L_task + λ||W_pt - W_ft||^2_F, whereas the from-scratch baselines minimize only L_task (Appendix C.2). This asymmetry confounds the transfer-learning comparison: the benefit attributed to pre-training in Tables 3 and 4 could be produced by the regularization term alone (e.g., as a weight-decay-like stabilizer), independent of the actual values of W_pt. Please add control experiments that (i) train from-scratch models with the same Frobenius penalty relative to a randomly initialized reference, and (ii) fine-tune with the pre-trained initialization but without the penalty. Without these ablations, the claim that ABCD pre-training transfers is not isolated from the regularizer's effect.
  2. [§4.4, Table 3 (1/16 row)] The 11.05% improvement reported in the abstract compares the means of SGUNET-FT Uniform (0.057560±0.0034) and SGUNET from scratch (0.064714±0.0096) on 1/16 Deformable Plate. The standard deviations overlap substantially, and with only 5 seeds the difference is not demonstrably significant. The abstract's stronger claim that fine-tuning on 1/16 beats from-scratch on the full dataset (0.059615±0.0005) also relies on overlapping intervals. Please report significance tests or bootstrap confidence intervals for these pairwise comparisons, and adjust the abstract if the effect is not statistically supported.
  3. [§4.1, §4.4] The method assumes that pre-training on ABCD (3D CAD parts in contact) transfers to the 2D Deformable Plate and 3D Deforming Plate benchmarks, but no evidence of domain similarity or covariate shift is provided. If the regularization confound is resolved and gains persist, it would still be valuable to show that the pre-trained representations are actually reused; please add a quantitative analysis of feature reuse or a similarity measure between the pre-training and downstream data distributions.
minor comments (5)
  1. [§3.4.2, Table 1] Please report the value of λ used in the experiments and any sensitivity analysis; it is missing from the hyperparameter tables and the experiment details.
  2. [Appendix C.3, Tables 3 and 4] Tables 3 and 4 are difficult to read in the current typesetting; please ensure proper column alignment and clear row/column labels.
  3. [§5 Discussion] The statement that this is the first time transfer learning has been applied to GNNs predicting physics simulations is too strong given the emerging literature on transfer and foundation models for PDE solvers; please soften the claim or provide a more comprehensive related-work discussion.
  4. [Figure 5] Figure 5's caption says all models are trained on 1/8 of the original training size, while the surrounding text also discusses 1/16 results; please clarify which figure corresponds to which data fraction.
  5. [Appendix B.1, Algorithm 1] In Algorithm 1, the variables cid, cnt, and left are modified inside the DFS function without being declared as parameters; please clarify the scoping to avoid ambiguity.

Circularity Check

0 steps flagged · score 0.0 of 10

No circular derivation: the transfer-learning claims are empirical benchmark comparisons, not quantities forced by construction.

full rationale

The paper's central claims are empirical comparisons of fine-tuned versus from-scratch models on the Deformable Plate and Deforming Plate test sets (Tables 3 and 4). The 11.05% improvement quoted in the abstract is direct arithmetic on those independently evaluated test-set means, not a fitted quantity renamed as a prediction. The mapping functions of Section 3.4.1 and the Frobenius regularization term of Section 3.4.2 define the fine-tuning procedure, but the reported RMSE numbers are not equal to those definitions by construction; they come from running the training procedures and evaluating held-out rollouts. The from-scratch baselines use only L_task (Appendix C.2) while fine-tuning uses L_reg = L_task + lambda*||W_pt - W_ft||^2_F, which is a legitimate potential confound for the transfer claim but is not a circularity: the result is not forced by the equations. The only self-citation, Hafez and Rashid (2023) for the Shabaka meshing tool used in constructing ABCD, is instrumental rather than load-bearing, and no uniqueness or external-validation claim rests on it. No step in the paper reduces to its own input, so the circularity score is 0.

Assumptions & free parameters 5 free parameters · 4 assumptions · 0 invented entities

The paper's central claims rest on standard ML training assumptions plus two domain assumptions: that graph networks can represent the deformation physics, and that pre-training on the authors' ABCD compression dataset transfers to plate-bending benchmarks. Several hyperparameters (lambda, pooling ratios, block counts, noise, edge thresholds) are chosen per dataset and some are not reported, which limits independent reproduction.

free parameters (5)
  • Regularization weight lambda = not reported
    Controls the Frobenius norm penalty between pre-trained and fine-tuned weights in L_reg = L_task + lambda * ||W_pt - W_ft||_F^2. The value is never given, so the effect on the reported improvements cannot be assessed or reproduced. Appears in Section 3.4.2.
  • Pooling ratios per dataset = ABCD: [4,2,2]; Deforming Plate: [4,2]; Deformable Plate: [2]
    The number of pooling stages and ratios in SGUNET are chosen per dataset in Table 1 to adapt to mesh sizes and are part of the model configuration selection.
  • Number of message passing blocks = Table 1, e.g., Encoder/Processor/GUnet counts per dataset
    The depth of the encoder, processor, and GUnet stages is set per dataset and selected based on validation performance, affecting receptive field and capacity.
  • Noise standard deviation = 0.0003 (ABCD), 0.003 (Deforming Plate), 0.05 (Deformable Plate)
    Added to input node features during next-step prediction; values follow prior papers for the benchmarks and are chosen smaller for ABCD, as stated in Appendix C.1.
  • Edge construction distance threshold = 0.0003 (ABCD), 0.003 (Deforming Plate), 0.05 (Deformable Plate)
    Used to build edges between elements or nodes; values are taken from prior work for benchmarks and chosen for ABCD, as described in Appendix B.4.
assumptions (4)
  • domain assumption Graph neural networks with message passing and pooling can learn to predict quasi-static hyper-elastic deformations from mesh data.
    The entire method assumes that stacked message passing blocks and a U-net with pooling can represent the physics; this is the standard assumption in Pfaff et al. (2021) and related work that the paper builds on. It appears in Sections 3.1 to 3.3.
  • domain assumption Pre-training on the ABCD dataset (compression of random 3D CAD shapes) transfers to the 2D Deformable Plate and 3D Deforming Plate benchmarks.
    The transfer experiments in Section 4.4 rely on this domain-similarity assumption; the paper does not quantify the domain gap.
  • domain assumption Averaging or selective copying of pre-trained Graph-Net block weights (Uniform and First-N mappings) yields a meaningful initialization for the fine-tuned model.
    This is an assumption about the geometry of parameter space for GNNs; there is no theoretical justification beyond empirical validation. See Section 3.4.1 and Appendix B.2.
  • standard math Standard backpropagation training with MSE loss and the given regularization will converge to a good solution.
    Standard ML training assumptions, used throughout the experimental section.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Transfer learning in Scalable Graph Neural Network for Improved Physical Simulation." pith.science (2026). https://pith.science/paper/U7EWVY2Q

@misc{pith2026250206848,
  author       = {Pith},
  title        = {Pith review of: Transfer learning in Scalable Graph Neural Network for Improved Physical Simulation},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U7EWVY2Q}},
  note         = {Machine review of arXiv:2502.06848}
}
read the original abstract

In recent years, Graph Neural Network (GNN) based models have shown promising results in simulating physics of complex systems. However, training dedicated graph network based physics simulators can be costly, as most models are confined to fully supervised training, which requires extensive data generated from traditional physics simulators. To date, how transfer learning could improve the model performance and training efficiency has remained unexplored. In this work, we introduce a pre-training and transfer learning paradigm for graph network simulators. We propose the scalable graph U-net (SGUNET). Incorporating an innovative depth-first search (DFS) pooling, the SGUNET is adaptable to different mesh sizes and resolutions for various simulation tasks. To enable the transfer learning between differently configured SGUNETs, we propose a set of mapping functions to align the parameters between the pre-trained model and the target model. An extra normalization term is also added into the loss to constrain the difference between the pre-trained weights and target model weights for better generalization performance. To pre-train our physics simulator we created a dataset which includes 20,000 physical simulations of randomly selected 3D shapes from the open source A Big CAD (ABC) dataset. We show that our proposed transfer learning methods allow the model to perform even better when fine-tuned with small amounts of training data than when it is trained from scratch with full extensive dataset. On the 2D Deformable Plate benchmark dataset, our pre-trained model fine-tuned on 1/16 of the training data achieved an 11.05\% improvement in position RMSE compared to the model trained from scratch.

Figures

Figures reproduced from arXiv: 2502.06848 by the authors.

Figure 1
Figure 1. (a) An illustration of the composition of mesh data [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. (a) A detailed depiction of our proposed model, S [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Randomized FEA simulation dataset using geometry from ABC dataset. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: The FEA simulation results using ABC CAD dataset highlight various deformation modes, [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 1
Figure 1. Figure 1: fig 1.00 50 acem [PITH_FULL_IMAGE:figures/full_fig_p008_1.png]
Figure 6
Figure 6. Figure 6: The best validation loss of the two models on the Deformable Plate dataset when trained [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Comparison of the best validation loss tendencies on the Deformable Plate. (a) Two [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 9
Figure 9. Figure 9: The best validation loss of the two models on the Deforming Plate dataset when trained [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Comparison of the best validation loss tendencies on the Deforming Plate. (a) Two [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]
Figure 11
Figure 11. Figure 11: Simulated meshes at various stages (t=30 at the top row, t=50 at the bottom row) for dif￾ferent models. All models are trained on the full training dataset. The colors indicate displacement magnitude. (a) MGN (b) MGN-FT (Uniform) (d) SGUNET (e) SGUNET-FT (Uniform) (g)…
Figure 12
Figure 12. Figure 12: Simulated meshes at various stages (t=200 at the top row, t=300 at the bottom row) for different models. All models are trained on the full training dataset. The colors indicate displace￾ment magnitude. 17 [PITH_FULL_IMAGE:figures/full_fig_p017_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

21 extracted references · 10 canonical work pages

  1. [1]

    Allen, Tatiana Lopez-Guevara, Yulia Rubanova, Kimberly L

    Kelsey R. Allen, Tatiana Lopez-Guevara, Yulia Rubanova, Kimberly L. Stachenfeld, Alvaro Sanchez-Gonzalez, Peter W. Battaglia, and Tobias Pfaff. Graph network simulators can learn discontinuous, rigid contact dynamics. In Karen Liu, Dana Kulic, and Jeffrey Ichnowski (eds.), Conference on Robot Learning, CoRL 2022, 14-18 December 2022, Auckland, New Zealand...

    arXiv 2022

  2. [2]

    15 C E XPERIMENT DETAILS C.1 M ODEL CONFIGURATION MGN: For the Deforming Plate and Deformable Plate datasets, we adhere to the settings outlined in the original paper Pfaff et al

    set the material properties λ and µ to zero. 15 C E XPERIMENT DETAILS C.1 M ODEL CONFIGURATION MGN: For the Deforming Plate and Deformable Plate datasets, we adhere to the settings outlined in the original paper Pfaff et al. (2021); Linkerh ¨agner et al. (2023). For the ABCD dataset, which involves larger-scale mesh sizes, more message passing steps are r...

    work page 2021

  3. [6]

    Predicting physics in mesh- reduced space with temporal attention

    Xu Han, Han Gao, Tobias Pfaff, Jian-Xun Wang, and Liping Liu. Predicting physics in mesh- reduced space with temporal attention. In The Tenth International Conference on Learning Representations, ICLR 2022, Virtual Event, April 25-29, 2022 . OpenReview.net,

    work page 2022

  4. [8]

    Scaling face interaction graph networks to real world scenes

    Tatiana Lopez-Guevara, Yulia Rubanova, William F Whitney, Tobias Pfaff, Kimberly Stachenfeld, and Kelsey R Allen. Scaling face interaction graph networks to real world scenes. arXiv preprint arXiv:2401.11985,

    arXiv

  5. [9]

    Language models are few-shot learners

    Ben Mann, N Ryder, M Subbiah, J Kaplan, P Dhariwal, A Neelakantan, P Shyam, G Sas- try, A Askell, S Agarwal, et al. Language models are few-shot learners. arXiv preprint arXiv:2005.14165, 1,

    arXiv 2005

  6. [11]

    Transfer learning with resnet-50 for malaria cell- image classification

    A Sai Bharadwaj Reddy and D Sujitha Juliet. Transfer learning with resnet-50 for malaria cell- image classification. In 2019 International conference on communication and signal processing (ICCSP), pp. 0945–0949. IEEE,

    work page 2019

  7. [12]

    Malicious software classification using transfer learning of resnet-50 deep neural network

    Edmar Rezende, Guilherme Ruppert, Tiago Carvalho, Fabio Ramos, and Paulo De Geus. Malicious software classification using transfer learning of resnet-50 deep neural network. In 2017 16th IEEE international conference on machine learning and applications (ICMLA) , pp. 1011–1014. IEEE,

    work page 2017

  8. [15]

    Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth ´ee Lacroix, Baptiste Rozi `ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al

    URL http://proceedings.mlr.press/v119/sanchez-gonzalez20a.html. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timoth ´ee Lacroix, Baptiste Rozi `ere, Naman Goyal, Eric Hambro, Faisal Azhar, et al. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971,

    arXiv

Show all 21 references
  1. [17]

    Learning flexible body collision dynamics with hierarchical contact mesh transformer

    Youn-Yeol Yu, Jeongwhan Choi, Woojin Cho, Kookjin Lee, Nayong Kim, Kiseok Chang, ChangSe- ung Woo, Ilho Kim, SeokWoo Lee, Joon-Young Yang, Sooyoung Yoon, and Noseong Park. Learning flexible body collision dynamics with hierarchical contact mesh transformer. CoRR, abs/2312.12467,

    arXiv

  2. [18]

    URL https://doi.org/10

    doi: 10.48550/ARXIV .2312.12467. URL https://doi.org/10. 48550/arXiv.2312.12467. Fuzhen Zhuang, Zhiyuan Qi, Keyu Duan, Dongbo Xi, Yongchun Zhu, Hengshu Zhu, Hui Xiong, and Qing He. A comprehensive survey on transfer learning. Proceedings of the IEEE , 109(1): 43–76,

    doi arXiv

  3. [19]

    The operator denotes concatenation, while | · |refers to the L2 norm

    representing whether the mesh node vi is a normal or boundary node, λi and µi are mechanical properties of the material, xt i denotes the world coordinates of node i at time t, and xt ij refers to the relative world position between nodes i and j at time t. The operator denote...

    2021

  4. [21]

    We evaluate the model performance across several configurations and select the one that performs best

    maintaining a model size that is comparable to or smaller than that of MGN. We evaluate the model performance across several configurations and select the one that performs best. Noise Std.: As we adopt the next-step prediction approach, adding noise to the input data is essen...

    2021

  5. [1998]

    Learning 3d particle-based simulators from rgb-d videos.arXiv preprint arXiv:2312.05359,

    William F Whitney, Tatiana Lopez-Guevara, Tobias Pfaff, Yulia Rubanova, Thomas Kipf, Kimberly Stachenfeld, and Kelsey R Allen. Learning 3d particle-based simulators from rgb-d videos.arXiv preprint arXiv:2312.05359,

    arXiv

  6. [2012]

    Battaglia

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter W. Battaglia. Learning mesh- based simulation with graph networks. In 9th International Conference on Learning Repre- sentations, ICLR 2021, Virtual Event, Austria, May 3-7, 2021 . OpenReview.net,

    2021

  7. [2017]

    Battaglia

    Yulia Rubanova, Alvaro Sanchez-Gonzalez, Tobias Pfaff, and Peter W. Battaglia. Constraint-based graph network simulator. In Kamalika Chaudhuri, Stefanie Jegelka, Le Song, Csaba Szepesv ´ari, Gang Niu, and Sivan Sabato (eds.), International Conference on Machine Learning, ICML ...

    2022

  8. [2018]

    Battaglia

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter W. Battaglia. Learning to simulate complex physics with graph networks. In Proceedings of the 37th International Conference on Machine Learning, ICML 2020, 13-18 July 2020, Virtual Event...

    2020

  9. [2019]

    Gnn-based physics solver for time-independent pdes

    Rini Jasmine Gladstone, Helia Rahmani, Vishvas Suryakumar, Hadi Meidani, Marta D’Elia, and Ahmad Zareei. Gnn-based physics solver for time-independent pdes. arXiv preprint arXiv:2303.15681,

    arXiv

  10. [2020]

    Grounding graph network simulators using physical sensor observations

    11 Jonas Linkerh ¨agner, Niklas Freymuth, Paul Maria Scheikl, Franziska Mathis-Ullrich, and Ger- hard Neumann. Grounding graph network simulators using physical sensor observations. arXiv preprint arXiv:2302.11864,

    arXiv

  11. [2022]

    URL https://doi.org/10.48550/arXiv.2210.00612

    doi: 10.48550/ARXIV .2210.00612. URL https://doi.org/10.48550/arXiv.2210.00612. Hongyang Gao and Shuiwang Ji. Graph u-nets. In international conference on machine learning , pp. 2083–2092. PMLR,

    doi arXiv

  12. [2023]

    Distance-based regularisation of deep networks for fine-tuning

    Henry Gouk, Timothy M Hospedales, and Massimiliano Pontil. Distance-based regularisation of deep networks for fine-tuning. arXiv preprint arXiv:2002.08253,

    2002 arXiv

  13. [2024]

    doi: https://doi.org/10.1016/j.engappai.2024.108055

    ISSN 0952-1976. doi: https://doi.org/10.1016/j.engappai.2024.108055. URL https: //www.sciencedirect.com/science/article/pii/S0952197624002136. Meire Fortunato, Tobias Pfaff, Peter Wirnsberger, Alexander Pritzel, and Peter W. Battaglia. Mul- tiscale meshgraphnets. CoRR, abs/2210.00612,

    1976

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.