pith. sign in

arxiv: 2605.25115 · v1 · pith:FCUZUZKSnew · submitted 2026-05-24 · 💻 cs.LG · cs.AI· cs.CE· physics.app-ph

Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support and Interpretable Field Decomposition

Anuj Kumar , Josiah Bjorgaard , Nikolaos Bouklas , Matteo Salvador , Alexander Lavin This is my paper

Pith reviewed 2026-06-30 12:01 UTC · model grok-4.3

classification 💻 cs.LG cs.AIcs.CEphysics.app-ph
keywords neural surrogatePerceiverscientific machine learninginterpretable modelslocal supportadaptive refinementfield decomposition
0
0 comments X

The pith

Courant neural surrogate produces latents with local support and multiscale specialization in physical domains.

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

The paper introduces Courant, a Perceiver-based model for simulating physical fields that learns latent representations with built-in interpretability. These latents adapt to the state, show local support in space, and specialize at different scales, similar to how traditional numerical methods refine grids. Trained only with standard loss on simulation data, it achieves competitive accuracy while allowing a decomposition of the field into geometry-anchored components. This matters because it bridges neural surrogates with classical solver concepts like hp-refinement and basis function expansions without extra constraints.

Core claim

Courant is a Perceiver-based encoder-processor-decoder surrogate that combines shared random Fourier feature coordinate embedding, state-adapted latent queries, and a light-weight decoder. When trained end-to-end on steady or transient simulation data using only L2 prediction loss in physical space, its latent features develop adaptive specialization and local support, enabling functionality akin to an adaptive hp-refinement scheme. The latents exhibit multiscale geometric specialization and track coherent structures over time, acting analogously to time-evolving spatial basis functions that permit a compact, geometry-anchored, partition-of-unity-like decomposition of the simulated field.

What carries the argument

State-adapted latent queries in a Perceiver architecture that, through end-to-end training with L2 loss, induce local support and multiscale specialization in the physical domain.

If this is right

  • Courant achieves competitive accuracy on benchmarks for steady and transient simulations.
  • The model functions like an adaptive hp-refinement scheme desirable in traditional solvers.
  • Latents allow decoding a partition-of-unity-like decomposition of the field.
  • Features track coherent structures in time-dependent cases.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the interpretability holds, it could reduce the need for post-hoc analysis in scientific ML models.
  • This approach might extend to other architectures by incorporating similar inductive biases for local support.
  • Visualizing the latents could serve as a diagnostic for the quality of the surrogate in capturing physical structures.

Load-bearing premise

That training the Perceiver architecture end-to-end with only a standard L2 prediction loss in physical space is sufficient to induce local support, multiscale specialization, and partition-of-unity-like decomposition without additional regularization or losses.

What would settle it

Visualizing the learned latent features on a held-out simulation and checking whether they exhibit local support and multiscale geometric specialization; if they do not, the claim fails.

read the original abstract

We introduce "Courant", a Perceiver-based encoder-processor-decoder surrogate model that has latent features exhibiting adaptive specialization and local support in the physical space, enabling functionality akin to an adaptive hp-refinement scheme, an attribute that is highly desirable in traditional numerical solvers and scientific machine learning broadly. The proposed architecture combines a shared random Fourier feature coordinate embedding, state-adapted latent queries, and a light-weight decoder. Courant is trained end-to-end with steady or transient simulation data and only a standard L_2 prediction loss in the physical space, achieving competitive accuracy on benchmarks. We demonstrate that Courant's inductive biases yield latents that are interpretable by design: they develop multiscale geometric specialization in the simulation domain and track coherent structures in the time-dependent case, acting analogously to time-evolving spatial basis functions and allowing for decoding a compact, geometry-anchored, partition-of-unity-like decomposition of the simulated field.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper introduces Courant, a Perceiver-based encoder-processor-decoder neural surrogate for steady and transient physical simulations. It combines a shared random Fourier feature (RFF) coordinate embedding, state-adapted latent queries, and a lightweight decoder, trained end-to-end solely with an L2 prediction loss in physical space. The central claims are that this architecture achieves competitive accuracy on benchmarks while its inductive biases produce latent features with local support and multiscale geometric specialization; these latents track coherent structures in time-dependent cases, act analogously to time-evolving spatial basis functions, and enable decoding of a compact, geometry-anchored, partition-of-unity-like decomposition of the simulated field, akin to adaptive hp-refinement.

Significance. If the emergence of local support, multiscale specialization, and partition-of-unity decomposition from standard L2 training is rigorously demonstrated, the work would be significant for scientific machine learning by offering interpretable neural surrogates that mimic desirable properties of traditional adaptive numerical methods without auxiliary losses or post-processing. The architecture's use of shared RFF embeddings and state-adaptation is a plausible inductive bias worth exploring, but the current description supplies no quantitative metrics, baselines, error bars, dataset details, or ablation results to support the accuracy or interpretability claims.

major comments (1)
  1. [Abstract] Abstract: The claim that 'Courant's inductive biases yield latents that are interpretable by design' with local support and a 'partition-of-unity-like decomposition' rests on the assertion that shared RFF + state-adapted queries + L2 loss alone suffice; however, the architecture description provides no mathematical enforcement (e.g., no locality bias, orthogonality constraint, or auxiliary loss) and global cross-attention does not inherently produce these properties, making the emergence claim load-bearing and requiring explicit ablation evidence that removing state-adaptation or RFF destroys the specialization while preserving accuracy.
minor comments (2)
  1. [Abstract] Abstract: No quantitative metrics, baseline comparisons, error bars, or dataset details are supplied to support the 'competitive accuracy' assertion, which is required to evaluate the practical utility of the surrogate.
  2. [Abstract] Abstract: The phrase 'acting analogously to time-evolving spatial basis functions' is used without defining the precise sense of analogy or providing a quantitative measure (e.g., overlap with traditional basis functions) that would allow readers to assess the claim.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the detailed and constructive review. We address the single major comment below and will revise the manuscript to strengthen the supporting evidence for the emergence claim.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The claim that 'Courant's inductive biases yield latents that are interpretable by design' with local support and a 'partition-of-unity-like decomposition' rests on the assertion that shared RFF + state-adapted queries + L2 loss alone suffice; however, the architecture description provides no mathematical enforcement (e.g., no locality bias, orthogonality constraint, or auxiliary loss) and global cross-attention does not inherently produce these properties, making the emergence claim load-bearing and requiring explicit ablation evidence that removing state-adaptation or RFF destroys the specialization while preserving accuracy.

    Authors: We agree that the interpretability properties are presented as emerging from the inductive biases (shared RFF coordinate embedding, state-adapted queries, and end-to-end L2 loss) without explicit mathematical constraints or auxiliary terms, and that global cross-attention alone does not guarantee locality or multiscale specialization. The manuscript demonstrates these properties via qualitative analysis and visualizations of the learned latents across benchmarks. To make the emergence claim more rigorous as requested, we will add explicit ablation studies in the revised version: we will train and evaluate variants without state-adaptation and without the shared RFF embedding, reporting both predictive accuracy (with error bars) and quantitative measures of latent support and scale specialization. The abstract and discussion will be updated to clarify that the properties are observed to emerge under the proposed biases rather than being strictly enforced. revision: yes

Circularity Check

0 steps flagged

No circularity; claims rest on observed emergence from architecture and L2 loss

full rationale

The paper introduces an architecture (shared RFF embedding + state-adapted queries + lightweight decoder) and states that training with only physical-space L2 loss produces latents with local support, multiscale specialization, and partition-of-unity-like decomposition 'by design' via inductive biases. No equations, derivations, or self-citations are present in the provided text that reduce these properties to fitted parameters, self-definitions, or load-bearing prior work by the authors. The central claim is an empirical observation about post-training behavior rather than a mathematical reduction that loops back to its inputs. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on standard machine-learning assumptions about end-to-end L2 training producing desired inductive biases; no free parameters, new physical entities, or ad-hoc axioms are introduced in the abstract.

axioms (1)
  • domain assumption End-to-end training of a Perceiver-style encoder-processor-decoder with L2 loss on physical-space data produces latent features with local support and interpretable decomposition.
    This is the core premise invoked when the abstract states that the architecture achieves the listed properties after training with only a standard L2 loss.

pith-pipeline@v0.9.1-grok · 5713 in / 1358 out tokens · 58857 ms · 2026-06-30T12:01:24.748201+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 17 canonical work pages · 3 internal anchors

  1. [1]

    Learning mesh-based simulation with graph networks

    Tobias Pfaff, Meire Fortunato, Alvaro Sanchez-Gonzalez, and Peter Battaglia. Learning mesh-based simulation with graph networks. InInternational Conference on Learning Representations (ICLR), 2021

    2021

  2. [2]

    Message passing neural pde solvers.arXiv preprint arXiv:2202.03376, 2022

    Johannes Brandstetter, Daniel Worrall, and Max Welling. Message passing neural pde solvers.arXiv preprint arXiv:2202.03376, 2022. Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support 12

    work page arXiv 2022

  3. [3]

    Fourier neural operator for parametric partial differential equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial differential equations. International Conference on Learning Representations (ICLR), 2021

    2021

  4. [4]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.Nature machine intelligence, 3(3):218–229, 2021

    Lu Lu, Pengzhan Jin, Guofei Pang, Zhongqiang Zhang, and George Em Karniadakis. Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators.Nature machine intelligence, 3(3):218–229, 2021

    2021

  5. [5]

    Universal physics transformers: A framework for efficiently scaling neural operators.Advances in Neural Information Processing Systems, 37:25152–25194, 2024

    Benedikt Alkin, Andreas Fürst, Simon Schmid, Lukas Gruber, Markus Holzleitner, and Johannes Brand- stetter. Universal physics transformers: A framework for efficiently scaling neural operators.Advances in Neural Information Processing Systems, 37:25152–25194, 2024

    2024

  6. [6]

    CALM-PDE: Continuous and adaptive convolutions for latent space modeling of time-dependent PDEs.arXiv preprint arXiv:2505.12944, 2025

    Jan Hagnberger, Daniel Musekamp, and Mathias Niepert. CALM-PDE: Continuous and adaptive convolutions for latent space modeling of time-dependent PDEs.arXiv preprint arXiv:2505.12944, 2025

    work page arXiv 2025

  7. [7]

    Transolver: A Fast Transformer Solver for PDEs on General Geometries

    Haixu Wu, Huakun Luo, Haowen Wang, Jianmin Wang, and Mingsheng Long. Transolver: A fast transformer solver for pdes on general geometries.arXiv preprint arXiv:2402.02366, 2024

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  8. [8]

    Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations.Nat

    Nick McGreivy and Ammar Hakim. Weak baselines and reporting biases lead to overoptimism in machine learning for fluid-related partial differential equations.Nat. Mac. Intell., 2024

    2024

  9. [9]

    Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    George Em Karniadakis, Ioannis G Kevrekidis, Lu Lu, Paris Perdikaris, Sifan Wang, and Liu Yang. Physics-informed machine learning.Nature Reviews Physics, 3(6):422–440, 2021

    2021

  10. [10]

    From PINNs to PIKANs: Recent advances in physics-informed machine learning.Machine Learning for Computational Science and Engineering, 1(1):15, 2025

    Juan Diego Toscano, Vivek Oommen, Alan John Varghese, Zongren Zou, Nazanin Ahmadi Daryakenari, Chenxi Wu, and George Em Karniadakis. From PINNs to PIKANs: Recent advances in physics-informed machine learning.Machine Learning for Computational Science and Engineering, 1(1):15, 2025

    2025

  11. [11]

    Luis Mandl, Somdatta Goswami, Lena Lambers, and Tim Ricken. Separable physics-informed deeponet: Breaking the curse of dimensionality in physics-informed machine learning.Computer Methods in Applied Mechanics and Engineering, 434:117586, 2025

    2025

  12. [12]

    Technology readiness levels for machine learning systems.Nature Communications, 13:6039, 2022

    Alexander Lavin, Ciarán M Gilligan-Lee, Alessya Visser, Julie Gori, Alexander Golbraikh, Roselyne B Tchoua, Chris Rackauckas, et al. Technology readiness levels for machine learning systems.Nature Communications, 13:6039, 2022

    2022

  13. [13]

    Simulation intelligence: Towards a new generation of scientific methods.arXiv preprint arXiv:2112.03235, 2021

    Alexander Lavin, David Krakauer, Hector Zenil, Jacob Gottschall, Tim Gros-Louis, Anita Karami, Peter Mattson, Albert Kolb, Bart Selman, et al. Simulation intelligence: Towards a new generation of scientific methods.arXiv preprint arXiv:2112.03235, 2021

    work page arXiv 2021

  14. [14]

    Eighty years of the finite element method: birth, evolution, and future.Archives of Computational Methods in Engineering, 29(6):4431–4453, 2022

    Wing Kam Liu, Shaofan Li, and Harold S Park. Eighty years of the finite element method: birth, evolution, and future.Archives of Computational Methods in Engineering, 29(6):4431–4453, 2022

    2022

  15. [15]

    Variational methods for the solution of problems of equilibrium and vibrations.Bulletin of the American Mathematical Society, 49(1):1–23, 1943

    Richard Courant. Variational methods for the solution of problems of equilibrium and vibrations.Bulletin of the American Mathematical Society, 49(1):1–23, 1943. doi: 10.1090/S0002-9904-1943-07818-4

    work page doi:10.1090/s0002-9904-1943-07818-4 1943

  16. [16]

    The hp version of the finite element method: Part 1: The basic approximation results.Computational Mechanics, 1(1):21–41, 1986

    Benqi Guo and Ivo Babuška. The hp version of the finite element method: Part 1: The basic approximation results.Computational Mechanics, 1(1):21–41, 1986

    1986

  17. [17]

    Ameya D Jagtap and George Em Karniadakis. Extended physics-informed neural networks (XPINNs): A generalized space-time domain decomposition based deep learning framework for nonlinear partial differential equations.Communications in Computational Physics, 28(5), 2020

    2020

  18. [18]

    Partition of unity networks: deep hp-approximation.arXiv preprint arXiv:2101.11256, 2021

    Kookjin Lee, Nathaniel A Trask, Ravi G Patel, Mamikon A Gulian, and Eric C Cyr. Partition of unity networks: deep hp-approximation.arXiv preprint arXiv:2101.11256, 2021

    work page arXiv 2021

  19. [19]

    Wavelet neural operator: a neural operator for parametric partial differential equations.arXiv preprint arXiv:2205.02191, 2022

    Tapas Tripura and Souvik Chakraborty. Wavelet neural operator: a neural operator for parametric partial differential equations.arXiv preprint arXiv:2205.02191, 2022

    work page arXiv 2022

  20. [20]

    The proper orthogonal decomposition in the analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

    Gal Berkooz, Philip Holmes, and John L Lumley. The proper orthogonal decomposition in the analysis of turbulent flows.Annual review of fluid mechanics, 25(1):539–575, 1993

    1993

  21. [21]

    Peter J. Schmid. Dynamic mode decomposition of numerical and experimental data.Journal of Fluid Mechanics, 656:5–28, 2010. doi: 10.1017/S0022112010001217

    work page doi:10.1017/s0022112010001217 2010

  22. [22]

    Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support 13 Perceiver: General perception with iterative attention

    Andrew Jaegle, Felix Gimeno, Andy Brock, Oriol Vinyals, Andrew Zisserman, and Joao Carreira. Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support 13 Perceiver: General perception with iterative attention. InInternational conference on machine learning, pages 4651–4664. PMLR, 2021

    2021

  23. [23]

    Ricky T. Q. Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. Neural ordinary differential equations. InAdvances in Neural Information Processing Systems (NeurIPS), 2018

    2018

  24. [24]

    Transolver++: An accurate neural solver for pdes on million-scale geometries.arXiv preprint arXiv:2502.02414, 2025

    Huakun Luo, Haixu Wu, Hang Zhou, Lanxiang Xing, Yichen Di, Jianmin Wang, and Mingsheng Long. Transolver++: An accurate neural solver for pdes on million-scale geometries.arXiv preprint arXiv:2502.02414, 2025

    work page arXiv 2025

  25. [25]

    Transolver-3: Scaling Up Transformer Solvers to Industrial-Scale Geometries

    Hang Zhou, Haixu Wu, Haonan Shangguan, Yuezhou Ma, Huikun Weng, Jianmin Wang, and Mingsheng Long. Transolver-3: Scaling up transformer solvers to industrial-scale geometries.arXiv preprint arXiv:2602.04940, 2026

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  26. [26]

    Aroma: Preserving spatial structure for latent pde modeling with local neural fields.Advances in Neural Information Processing Systems, 37:13489–13521, 2024

    Louis Serrano, Thomas X Wang, Etienne Le Naour, Jean-Noël Vittaut, and Patrick Gallinari. Aroma: Preserving spatial structure for latent pde modeling with local neural fields.Advances in Neural Information Processing Systems, 37:13489–13521, 2024

    2024

  27. [27]

    Geometry aware operator transformer as an efficient and accurate neural surrogate for PDEs on arbitrary domains.arXiv preprint arXiv:2505.18781, 2025

    Shizheng Wen, Arsh Kumbhat, Levi Lingsch, Sepehr Mousavi, Yizhou Zhao, Praveen Chandrashekar, and Siddhartha Mishra. Geometry aware operator transformer as an efficient and accurate neural surrogate for PDEs on arbitrary domains.arXiv preprint arXiv:2505.18781, 2025

    work page arXiv 2025

  28. [28]

    Wessels, David M

    David R. Wessels, David M. Knigge, Riccardo Valperga, Samuele Papa, Sharvaree Sonnino, Efstratios Gavves, and Erik J. Bekkers. Grounding continuous representations in geometry: Equivariant neural fields. InInternational Conference on Learning Representations (ICLR), 2025

    2025

  29. [29]

    Learning the intrinsic dynamics of spatio-temporal processes through latent dynamics networks.Nature Communica- tions, 15(1):1834, 2024

    Francesco Regazzoni, Stefano Pagani, Matteo Salvador, Luca Dede’, and Alfio Quarteroni. Learning the intrinsic dynamics of spatio-temporal processes through latent dynamics networks.Nature Communica- tions, 15(1):1834, 2024

    2024

  30. [30]

    Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    Steven L Brunton, Joshua L Proctor, and J Nathan Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems.Proceedings of the national academy of sciences, 113(15):3932–3937, 2016

    2016

  31. [31]

    Interpretable a-posteriori error indication for graph neural network surrogate models.Computer Methods in Applied Mechanics and Engineering, 433:117509, 2025

    Shivam Barwey, Hojin Kim, and Romit Maulik. Interpretable a-posteriori error indication for graph neural network surrogate models.Computer Methods in Applied Mechanics and Engineering, 433:117509, 2025

    2025

  32. [32]

    FIGNN: Feature-specific interpretability for graph neural network surrogate models.arXiv preprint arXiv:2506.11398, 2025

    Riddhiman Raut, Romit Maulik, and Shivam Barwey. FIGNN: Feature-specific interpretability for graph neural network surrogate models.arXiv preprint arXiv:2506.11398, 2025

    work page arXiv 2025

  33. [33]

    Towards interpretable deep learning and analysis of dynamical systems via the discrete empirical interpolation method.arXiv preprint arXiv:2510.21852, 2025

    Hojin Kim and Romit Maulik. Towards interpretable deep learning and analysis of dynamical systems via the discrete empirical interpolation method.arXiv preprint arXiv:2510.21852, 2025

    work page arXiv 2025

  34. [34]

    Interpreting CFD surrogates through sparse autoencoders.arXiv preprint arXiv:2507.16069, 2025

    Yeping Hu and Shusen Liu. Interpreting CFD surrogates through sparse autoencoders.arXiv preprint arXiv:2507.16069, 2025

    work page arXiv 2025

  35. [35]

    A comprehensive and fair comparison of two neural operators (with practical extensions) based on FAIR data.Computer Methods in Applied Mechanics and Engineering, 393:114778, 2022

    Lu Lu, Xuhui Meng, Shengze Cai, Zhiping Mao, Somdatta Goswami, Zhongqiang Zhang, and George Em Karniadakis. A comprehensive and fair comparison of two neural operators (with practical extensions) based on FAIR data.Computer Methods in Applied Mechanics and Engineering, 393:114778, 2022

    2022

  36. [36]

    Solving high-dimensional PDEs with latent spectral models

    Haixu Wu, Tengge Hu, Huakun Luo, Jianmin Wang, and Mingsheng Long. Solving high-dimensional PDEs with latent spectral models. InInternational Conference on Machine Learning, 2023

    2023

  37. [37]

    GeoTransolver: Learning Physics on Irregular Domains Using Multi-scale Geometry Aware Physics Attention Transformer

    Corey Adams, Rishikesh Ranade, Ram Cherukuri, and Sanjay Choudhry. GeoTransolver: Learning physics on irregular domains using multi-scale geometry aware physics attention transformer.arXiv preprint arXiv:2512.20399, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  38. [38]

    Finite basis kolmogorov-arnold networks: domain decomposition for data-driven and physics-informed problems

    Amanda A Howard, Bruno Jacob, Sarah Helfert, Alexander Heinlein, and Panos Stinis. Finite basis kolmogorov-arnold networks: domain decomposition for data-driven and physics-informed problems. arXiv preprint arXiv:2406.19662, 2024

    work page arXiv 2024

  39. [39]

    Latent neural operator for solving forward and inverse PDE problems

    Tian Wang and Chuang Wang. Latent neural operator for solving forward and inverse PDE problems. Advances in Neural Information Processing Systems, 37:33085–33107, 2024

    2024

  40. [40]

    Hybrid Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support 14 latent representations for PDE emulation

    Ali Can Bekar, Siddhant Agarwal, Christian Hüttig, Nicola Tosi, and David S Greenberg. Hybrid Courant: a State-Adaptive Perceiver-Based Neural Surrogate with Local Support 14 latent representations for PDE emulation. InThe Thirty-ninth Annual Conference on Neural Information Processing Systems, 2025

    2025

  41. [41]

    Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T

    Matthew Tancik, Pratul P. Srinivasan, Ben Mildenhall, Sara Fridovich-Keil, Nithin Raghavan, Utkarsh Singhal, Ravi Ramamoorthi, Jonathan T. Barron, and Ren Ng. Fourier features let networks learn high frequency functions in low dimensional domains. InAdvances in Neural Information Processing Systems (NeurIPS), 2020

    2020

  42. [42]

    Srinivasan, Matthew Tancik, Jonathan T

    Ben Mildenhall, Pratul P. Srinivasan, Matthew Tancik, Jonathan T. Barron, Ravi Ramamoorthi, and Ren Ng. NeRF: Representing scenes as neural radiance fields for view synthesis. InEuropean Conference on Computer Vision (ECCV), 2020

    2020

  43. [43]

    Continuous PDE dynamics forecasting with implicit neural representations

    Yuan Yin, Matthieu Kirchmeyer, Jean-Yves Franceschi, Alain Rakotomamonjy, and Patrick Gallinari. Continuous PDE dynamics forecasting with implicit neural representations. InInternational Conference on Learning Representations (ICLR), 2023

    2023

  44. [44]

    Operator learning with neural fields: Tackling PDEs on general geometries

    Yuan Yin, Matthieu Kirchmeyer, Jean-Yves Franceschi, Alain Rakotomamonjy, and Patrick Gallinari. Operator learning with neural fields: Tackling PDEs on general geometries. InAdvances in Neural Information Processing Systems (NeurIPS), 2023

    2023

  45. [45]

    Peter Yichen Chen, Jinxu Xiang, Dong Heon Cho, Yue Chang, G. A. Pershing, Henrique Teles Maia, Maurizio M. Chiaramonte, Kevin Carlberg, and Eitan Grinspun. CROM: Continuous reduced-order modeling of PDEs using implicit neural representations. InInternational Conference on Learning Representations (ICLR), 2023

    2023

  46. [46]

    Elsevier, 2024

    Pijush K Kundu, Ira M Cohen, David R Dowling, and Jesse Capecelatro.Fluid mechanics. Elsevier, 2024

    2024

  47. [47]

    Scalable transformer for PDE surrogate modeling.Advances in Neural Information Processing Systems, 36:28010–28039, 2023

    Zijie Li, Dule Shu, and Amir Barati Farimani. Scalable transformer for PDE surrogate modeling.Advances in Neural Information Processing Systems, 36:28010–28039, 2023

    2023

  48. [48]

    OpenFOAM: Open source CFD in research and industry, 2026

    OpenCFD Ltd. OpenFOAM: Open source CFD in research and industry, 2026. URL https://www. openfoam.org

    2026

  49. [49]

    ANSYS Fluent (version 2024 r1), 2024

    ANSYS, Inc. ANSYS Fluent (version 2024 r1), 2024. URL https://ansys.com. ANSYS Academic Research

    2024

  50. [50]

    Cambridge university press, 2000

    George Keith Batchelor.An introduction to fluid dynamics. Cambridge university press, 2000

    2000

  51. [51]

    PyTorch 2: Faster Machine Learning Through Dynamic Python Bytecode Transformation and Graph Compilation

    Jason Ansel, Edward Yang, Horace He, Natalia Gimelshein, Animesh Jain, Michael Voznesensky, Bin Bao, Peter Bell, David Berard, Evgeni Burovski, Geeta Chauhan, Anjali Chourdia, Will Constable, Alban Desmaison, Zachary DeVito, Elias Ellison, Will Feng, Jiong Gong, Michael Gschwind, Brian Hirsh, Sherlock Huang, Kshiteej Kalambarkar, Laurent Kirsch, Michael L...

    work page doi:10.1145/3620665.3640366 2024