pith. sign in

arxiv: 2605.22222 · v1 · pith:AROFCCG5new · submitted 2026-05-21 · 💻 cs.LG

ARC-STAR: Auditable Post-Hoc Correction for PDE Foundation Models

Chengze Li , Lingwei Wei , Li Sun , Hongbo Lv , Jie Yang , Hongrong Zhang , Kening Zheng , Wei-Chieh Huang
show 2 more authors
Enze Ma Philip S. Yu
This is my paper

Pith reviewed 2026-05-22 07:10 UTC · model grok-4.3

classification 💻 cs.LG
keywords PDE foundation modelspost-hoc correctionspatial triagefrozen solverauditable refinementrisk calibrationvelocity rollout
0
0 comments X

The pith

ARC-STAR corrects pretrained PDE foundation models after training by routing refinement only to high-risk spatial blocks while keeping the solver frozen.

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

The paper sets out to show that drifting forecasts from pretrained PDE solvers on new flows can be fixed without retraining the network or applying corrections everywhere at once. It does so by first removing broad biases with a global corrector, then cleaning residuals in selected blocks with a local refiner, and finally using a label-free risk score to decide which blocks receive limited compute at deployment time. This structure keeps the original solver untouched, lets each stage be checked on its own, and focuses effort where errors actually cluster. A reader would care because it offers a practical route to reuse large pretrained models on unfamiliar physical regimes with clear, measurable gains in rollout accuracy and without risking instability.

Core claim

ARC-STAR organizes correction into three stages: a global corrector removes broad solver bias, a blockwise local refiner cleans the post-global residual, and a label-free score routes refinement to high-risk blocks under a compute budget. The framework is frozen-host, auditable, and budget-aware. Across five flow benchmarks spanning ten regime cells, it is the only method that cuts velocity rollout error by at least 36x over raw Poseidon on every cell, with the global stage reducing raw host error by 91-99% and the local stage further reducing the remaining residual by up to 94.4%.

What carries the argument

The ARC-STAR three-stage pipeline that separates global bias removal from blockwise local refinement and uses a label-free risk score to allocate refinement under a budget.

If this is right

  • The global corrector alone removes 91-99% of the raw host error.
  • The local refiner can cut up to 94.4% of whatever residual error remains after the global stage.
  • The full reduction of at least 36x holds on every one of the ten regime cells across five benchmarks.
  • The original pretrained solver stays frozen and stable throughout correction.
  • Each stage can be trained and evaluated separately, making contributions measurable.

Where Pith is reading between the lines

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

  • The same blockwise triage could be tested on pretrained models for other time-dependent simulations where errors also localize spatially.
  • If the label-free risk score generalizes, it might reduce full-field correction costs in other sequential prediction settings that face compute limits.
  • The auditable separation of stages could support verification requirements in applications where corrections must be inspected before use.

Load-bearing premise

Prediction errors concentrate in identifiable spatial blocks that a label-free risk score can reliably detect without ground-truth labels or destabilizing the frozen pretrained solver.

What would settle it

A flow regime in which errors spread uniformly across the domain rather than clustering in blocks, or in which the label-free risk score consistently selects regions whose actual error does not exceed the rest of the field on unseen data, would undermine the triage mechanism.

Figures

Figures reproduced from arXiv: 2605.22222 by Chengze Li, Enze Ma, Hongbo Lv, Hongrong Zhang, Jie Yang, Kening Zheng, Lingwei Wei, Li Sun, Philip S. Yu, Wei-Chieh Huang.

Figure 1
Figure 1. Figure 1: Post-broad-correction residual is spatially concentrated, not uniform. Each cell is partitioned into 64 blocks of 16×16 pixels; tile color is per-block error share, with per-panel annotation reporting the share carried by the highest-error 20% of blocks (mean per-cell Gini coefficient 0.48, between the uniform null and the sparse-activation regime). Block ranking is an offline diagnostic; the deployed sele… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of ARC-STAR. I. Global-stage training: the frozen host stays fixed while a full-field residual operator learns a residual correction toward the ground-truth next state. II. Local￾stage training: after the global stage is fixed, a blockwise local refiner is trained on post-global fields with a halo-read, center-write patch interface, matching the inputs seen at deployment. III. Inference: a hand-de… view at source ↗
Figure 2
Figure 2. Figure 2: I. Step 2a densely pretrains Lθ on every block of the post-global field with an optional per-patch reweighting wb ≥0 (full form in Ap￾pendix D.2; wb ≡ 1 recovers the uniform vari￾ant); this phase teaches the refiner the block￾wise residual geometry it will see at deploy￾ment, before any rollout interaction is intro￾duced. Step 2b places the pretrained refiner back into the full hybrid rollout with k=B, re￾… view at source ↗
Figure 3
Figure 3. Figure 3: Routing frontier under shared compute. ARC-STAR (red) versus 9 routing policies; only block selection differs. Dashed and dotted lines mark global-only (k/B=0) and dense ARC￾STAR (k/B=1). Lower median 10-step UV-L 2 ratio is better. single stage rather than aggregating it into a suite-level statistic. A kinetic-energy spectrum check on NS-SL moderate (Appendix G.1) confirms full ARC-STAR matches GT to mean… view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative rollout diagnosis. Rows span high, medium, and low local recovery (NS￾SL(m), KF(m), NS-G(x); Jloc = 73.5%, 48.4%, 9.6%). The rightmost column shows local error reduction minus increase; strong reduction concentrates on structured vortex pockets, while NS-G(x) changes little, consistent with audit-bounded local headroom. Obs. ❼ The audit doubles as a forward-only routing diagnostic. Cells with t… view at source ↗
Figure 5
Figure 5. Figure 5: Data-efficiency curve on NS-PwC. ARC-STAR’s ten-step UV relative-L 2 ratio vs. training set size N under full retraining of Gϕ and Lθ. The curve saturates near N ≈ 100, consistent with the frozen Poseidon backbone supplying an N-independent predictive signal that the corrector aligns rather than re-learns. D.11 Block size and halo width sensitivity To verify that the deployed configuration (b=16, h=8) is n… view at source ↗
Figure 6
Figure 6. Figure 6: Kinetic-energy spectrum E(k) on NS-SL moderate at t0 + 10 averaged over n=8 rollouts. Lower curves at high k indicate less spurious dissipation-range energy [PITH_FULL_IMAGE:figures/full_fig_p034_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Per-block raw-residual energy concentration on Poseidon-T (top row) and DPOT-Ti (bottom row), five moderate-regime cells. Energy normalised to fraction per panel; shared color scale. Both hosts exhibit mild-to-moderate concentration: Gini ∈ [0.13, 0.22] (Poseidon-T) and [0.15, 0.20] (DPOT-Ti), mean cross-host |∆Gini| = 0.031 (max 0.067 on NS-PwC, min 0.008 on KF). Pearson spatial correlation between the tw… view at source ↗
Figure 8
Figure 8. Figure 8: Routing frontier on all ten benchmark–regime cells. Top row: NS-G, KF, NS-SL, NS-PwC, [PITH_FULL_IMAGE:figures/full_fig_p039_8.png] view at source ↗
read the original abstract

Partial differential equation (PDE) foundation models are pretrained networks that forecast how physical fields like velocity and pressure evolve from a single reusable solver. On unfamiliar flows their predictions drift step by step, errors concentrate in a few regions, yet retraining destabilizes the network and uniform post-hoc correction overlooks this spatial concentration. To address this, we propose a frozen-solver post-hoc correction framework, Adaptive Risk-Calibrated Spatial Triage for Auditable Refinement (ARC-STAR). ARC-STAR organizes correction into three stages: a global corrector removes broad solver bias, a blockwise local refiner cleans the post-global residual, and, at deployment, a label-free score routes refinement to high-risk blocks under a compute budget. The framework is designed to be (i) frozen-host, preserving the pretrained solver without fine-tuning; (ii) auditable, with global and local stages trained and evaluated separately for measurable contributions; and (iii) budget-aware, using a blockwise interface that either refines the full field or routes limited compute to high-risk regions. Across five flow benchmarks spanning ten regime cells, ARC-STAR is the only method that cuts velocity rollout error by at least 36x over raw Poseidon on every cell. The global stage reduces raw host error by 91-99%, and the local stage further reduces the remaining post-global residual by up to 94.4%. Our code implementation is available at https://anonymous.4open.science/r/arc_star.

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

2 major / 2 minor

Summary. The manuscript presents ARC-STAR, a frozen-solver post-hoc correction framework for PDE foundation models. It decomposes correction into a global corrector that removes broad solver bias, a blockwise local refiner that addresses post-global residuals, and a label-free risk score that routes limited compute to high-risk blocks at deployment. The framework is evaluated on five flow benchmarks spanning ten regime cells and claims to be the only method achieving at least a 36x reduction in velocity rollout error relative to raw Poseidon on every cell, with the global stage reducing raw host error by 91-99% and the local stage reducing the remaining residual by up to 94.4%. The design emphasizes auditability through separately trained and evaluated stages and budget awareness via the blockwise interface.

Significance. If the reported error reductions are supported by detailed per-cell baselines, statistical tests, and explicit validation of the label-free routing, the work would offer a practical, auditable alternative to retraining or uniform correction for improving out-of-distribution performance of PDE foundation models. The separate staging for measurability and the public code release are clear strengths that aid reproducibility and community verification.

major comments (2)
  1. [§3.3] §3.3 (Routing mechanism): The central claim that selective refinement under a compute budget delivers the 36x velocity error reduction on every cell rests on the label-free risk score correctly identifying spatially concentrated error blocks. The manuscript should report the correlation (e.g., Spearman or Pearson) between the risk score and ground-truth per-block rollout error on held-out regime cells, together with the fraction of high-error blocks that would be missed under the chosen budget threshold. Absent this, the contribution of the local stage to the headline gains cannot be isolated from uniform refinement.
  2. [Table 2] Table 2 or main results table (per-cell breakdown): The statement that ARC-STAR is the only method achieving at least 36x reduction 'on every cell' requires explicit per-cell velocity rollout errors for raw Poseidon, global-only, local-only, and full ARC-STAR (plus competing post-hoc baselines). Aggregate or best-case reporting leaves open whether the minimum factor holds uniformly or is driven by a subset of the ten regime cells.
minor comments (2)
  1. [Abstract] Abstract: Specify the precise rollout error metric (e.g., relative L2 norm averaged over 50 steps) and name the full set of baselines used to support the 'only method' claim.
  2. [§4] §4 (Experimental setup): Clarify the data splits, number of random seeds, and statistical test used for the 91-99% and 94.4% figures so that the numerical claims can be reproduced from the released code.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight opportunities to strengthen the empirical support for our claims. We respond to each major comment below and indicate the specific revisions we will make.

read point-by-point responses

  1. Referee: [§3.3] §3.3 (Routing mechanism): The central claim that selective refinement under a compute budget delivers the 36x velocity error reduction on every cell rests on the label-free risk score correctly identifying spatially concentrated error blocks. The manuscript should report the correlation (e.g., Spearman or Pearson) between the risk score and ground-truth per-block rollout error on held-out regime cells, together with the fraction of high-error blocks that would be missed under the chosen budget threshold. Absent this, the contribution of the local stage to the headline gains cannot be isolated from uniform refinement.

    Authors: We agree that an explicit correlation analysis would help isolate the local stage's contribution from uniform refinement. In the revised manuscript we will add to §3.3 the Spearman rank correlation between the label-free risk score and ground-truth per-block rollout error on the held-out regime cells, together with the fraction of high-error blocks missed at the operating budget threshold. These quantities will be computed using the same evaluation splits and error metric as the main results. revision: yes

  2. Referee: [Table 2] Table 2 or main results table (per-cell breakdown): The statement that ARC-STAR is the only method achieving at least 36x reduction 'on every cell' requires explicit per-cell velocity rollout errors for raw Poseidon, global-only, local-only, and full ARC-STAR (plus competing post-hoc baselines). Aggregate or best-case reporting leaves open whether the minimum factor holds uniformly or is driven by a subset of the ten regime cells.

    Authors: We acknowledge that aggregate reporting leaves the uniformity of the 36x claim open to question. We will revise Table 2 (or add a dedicated supplementary table if length constraints apply) to list the velocity rollout error for each of the ten regime cells under raw Poseidon, global-only, local-only, full ARC-STAR, and all competing post-hoc baselines. This will allow direct verification that the minimum 36x factor is attained on every cell. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper describes an empirical post-hoc correction framework with separately trained global and local stages whose contributions are measured on benchmarks. No equations, derivations, or self-referential definitions appear that reduce the reported error reductions (36x, 91-99%, 94.4%) to quantities defined by fitted parameters or prior self-citations. The label-free routing score is presented as a design choice evaluated empirically rather than derived from the target metrics by construction. The central claims rest on experimental results across ten regime cells rather than any load-bearing self-citation chain or ansatz smuggled via prior work.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities; the contribution is a methodological pipeline whose training details are not visible here.

pith-pipeline@v0.9.0 · 5829 in / 1089 out tokens · 47801 ms · 2026-05-22T07:10:32.713688+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

76 extracted references · 76 canonical work pages · 3 internal anchors

  1. [1]

    Angelopoulos, Stephen Bates, Adam Fisch, Lihua Lei, and Tal Schuster

    Anastasios N. Angelopoulos, Stephen Bates, Adam Fisch, Lihua Lei, and Tal Schuster. Confor- mal risk control, 2022. URLhttps://arxiv.org/abs/2208.02814

    work page arXiv 2022

  2. [2]

    Test-time data augmentation for estimation of heteroscedastic aleatoric uncertainty in deep neural networks

    Murat Seckin Ayhan and Philipp Berens. Test-time data augmentation for estimation of heteroscedastic aleatoric uncertainty in deep neural networks. InInternational Conference on Medical Imaging with Deep Learning, 2018. URL https://openreview.net/forum?id= rJZz-knjz

    work page 2018

  3. [3]

    Yohai Bar-Sinai, Stephan Hoyer, Jason Hickey, and Michael P. Brenner. Learning data-driven discretizations for partial differential equations.Proceedings of the National Academy of Sciences, 116(31):15344–15349, 2019. doi: 10.1073/pnas.1814058116

    work page doi:10.1073/pnas.1814058116 2019

  4. [4]

    Berger and Joseph Oliger

    Marsha J. Berger and Joseph Oliger. Adaptive mesh refinement for hyperbolic partial differential equations.Journal of Computational Physics, 53(3):484–512, 1984. doi: 10.1016/0021-9991(84)90073-1

    work page doi:10.1016/0021-9991(84)90073-1 1984

  5. [5]

    Message passing neural PDE solvers

    Johannes Brandstetter, Daniel Worrall, and Max Welling. Message passing neural PDE solvers. InInternational Conference on Learning Representations, 2022. URL https://openreview. net/forum?id=vSix3HPYKSU

    work page 2022

  6. [6]

    Choose a transformer: Fourier or Galerkin

    Shuhao Cao. Choose a transformer: Fourier or Galerkin. InAdvances in Neural Information Processing Systems, volume 34, 2021. URL https://proceedings.neurips.cc/paper/ 2021/hash/d0921d442ee91b896ad95059d13df618-Abstract.html

    work page 2021

  7. [7]

    Numerical solution of the Navier–Stokes equations.Mathematics of Computation, 22(104):745–762, 1968

    Alexandre Joel Chorin. Numerical solution of the Navier–Stokes equations.Mathematics of Computation, 22(104):745–762, 1968. doi: 10.1090/S0025-5718-1968-0242392-2

    work page doi:10.1090/s0025-5718-1968-0242392-2 1968

  8. [8]

    Donoho and Iain M

    David L. Donoho and Iain M. Johnstone. Ideal spatial adaptation by wavelet shrinkage. Biometrika, 81(3):425–455, 1994. doi: 10.1093/biomet/81.3.425

    work page doi:10.1093/biomet/81.3.425 1994

  9. [9]

    A convergent adaptive algorithm for Poisson’s equation.SIAM Journal on Numerical Analysis, 33(3):1106–1124, 1996

    Willy Dörfler. A convergent adaptive algorithm for Poisson’s equation.SIAM Journal on Numerical Analysis, 33(3):1106–1124, 1996. doi: 10.1137/0733054

    work page doi:10.1137/0733054 1996

  10. [10]

    Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.Journal of Machine Learning Research, 23 (120):1–39, 2022

    William Fedus, Barret Zoph, and Noam Shazeer. Switch transformers: Scaling to trillion parameter models with simple and efficient sparsity.Journal of Machine Learning Research, 23 (120):1–39, 2022. URLhttps://jmlr.org/papers/v23/21-0998.html

    work page 2022

  11. [11]

    Dropout as a bayesian approximation: Representing model uncertainty in deep learning

    Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Representing model uncertainty in deep learning. InInternational Conference on Machine Learning, pages 1050–1059, 2016. URLhttps://proceedings.mlr.press/v48/gal16.html

    work page 2016

  12. [12]

    Adaptive Computation Time for Recurrent Neural Networks

    Alex Graves. Adaptive computation time for recurrent neural networks, 2016. URL https: //arxiv.org/abs/1603.08983

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  13. [13]

    Multiwavelet-based operator learning for differential equations

    Gaurav Gupta, Xiongye Xiao, and Paul Bogdan. Multiwavelet-based operator learning for differential equations. InAdvances in Neural Information Processing Systems, volume 34, 2021

    work page 2021

  14. [14]

    Gupta and Johannes Brandstetter

    Jayesh K. Gupta and Johannes Brandstetter. Towards multi-spatiotemporal-scale general- ized PDE modeling.Transactions on Machine Learning Research, 2023. URL https: //openreview.net/forum?id=dPSTDbGtBY

    work page 2023

  15. [15]

    Dpot: Auto-regressive denoising operator transformer for large- scale pde pre-training.arXiv preprint arXiv:2403.03542,

    Zhongkai Hao, Chang Su, Songming Liu, Julius Berner, Chengyang Ying, Hang Su, Anima Anandkumar, Jian Song, and Jun Zhu. DPOT: Auto-regressive denoising operator transformer for large-scale PDE pre-training. InInternational Conference on Machine Learning, 2024. URL https://arxiv.org/abs/2403.03542

    work page arXiv 2024

  16. [16]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. InIEEE Conference on Computer Vision and Pattern Recognition, pages 770–778,

    work page

  17. [17]

    doi: 10.1109/CVPR.2016.90. 10

    work page doi:10.1109/cvpr.2016.90 2016

  18. [18]

    Poseidon: Efficient foundation models for pdes.arXiv preprint arXiv:2405.19101,

    Maximilian Herde, Bogdan Raoni ´c, Tobias Rohner, Roger Käppeli, Roberto Molinaro, Emmanuel de Bézenac, and Siddhartha Mishra. Poseidon: Efficient foundation models for PDEs. InAdvances in Neural Information Processing Systems, 2024. URL https: //arxiv.org/abs/2405.19101

    work page arXiv 2024

  19. [19]

    Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen

    Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. Lora: Low-rank adaptation of large language models. In International Conference on Learning Representations, 2022. URL https://openreview. net/forum?id=nZeVKeeFYf9

    work page 2022

  20. [20]

    PhysicsCorrect: A training-free approach for stable neural PDE simulations

    Xinquan Huang and Paris Perdikaris. PhysicsCorrect: A training-free approach for stable neural PDE simulations. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40,

    work page

  21. [21]

    doi: 10.1609/aaai.v40i26.39360

    work page doi:10.1609/aaai.v40i26.39360

  22. [22]

    Smith, Ayya Alieva, Qing Wang, Michael P

    Dmitrii Kochkov, Jamie A. Smith, Ayya Alieva, Qing Wang, Michael P. Brenner, and Stephan Hoyer. Machine learning–accelerated computational fluid dynamics.Proceedings of the National Academy of Sciences, 118(21):e2101784118, 2021. doi: 10.1073/pnas.2101784118

    work page doi:10.1073/pnas.2101784118 2021

  23. [23]

    Multi-grid tensorized Fourier neural operator for high-resolution PDEs.Transactions on Machine Learning Research, 2023

    Jean Kossaifi, Nikola Kovachki, Kamyar Azizzadenesheli, and Anima Anandkumar. Multi-grid tensorized Fourier neural operator for high-resolution PDEs.Transactions on Machine Learning Research, 2023. URLhttps://openreview.net/forum?id=AWiDlO63bH

    work page 2023

  24. [24]

    Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023

    Nikola Kovachki, Zongyi Li, Burigede Liu, Kamyar Azizzadenesheli, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Neural operator: Learning maps between function spaces with applications to pdes.Journal of Machine Learning Research, 24(89):1–97, 2023. URLhttps://jmlr.org/papers/v24/21-1524.html

    work page 2023

  25. [25]

    Temporal ensembling for semi-supervised learning

    Samuli Laine and Timo Aila. Temporal ensembling for semi-supervised learning. InInter- national Conference on Learning Representations, 2017. URLhttps://openreview.net/ forum?id=BJ6oOfqge

    work page 2017

  26. [26]

    Simple and scalable predictive uncertainty estimation using deep ensembles

    Balaji Lakshminarayanan, Alexander Pritzel, and Charles Blundell. Simple and scalable predictive uncertainty estimation using deep ensembles. InAdvances in Neural Information Processing Systems, volume 30, 2017. URL https://proceedings.neurips.cc/paper_ files/paper/2017/file/9ef2ed4b7fd2c810847ffa5fa85bce38-Paper.pdf

    work page 2017

  27. [27]

    Learning skillful medium-range global weather forecasting.Science, 382(6677): 1416–1421, 2023

    Remi Lam, Alvaro Sanchez-Gonzalez, Matthew Willson, Peter Wirnsberger, Meire Fortunato, Ferran Alet, Suman Ravuri, Timo Ewalds, Zach Eaton-Rosen, Weihua Hu, Alexander Merose, Stephan Hoyer, George Holland, Jacklynn Stott, Oriol Vinyals, Shakir Mohamed, and Peter Battaglia. Learning skillful medium-range global weather forecasting.Science, 382(6677): 1416–...

    work page doi:10.1126/science.adi2336 2023

  28. [28]

    A simple unified framework for detecting out-of-distribution samples and adversarial attacks

    Kimin Lee, Kibok Lee, Honglak Lee, and Jinwoo Shin. A simple unified framework for detecting out-of-distribution samples and adversarial attacks. InAdvances in Neural Information Processing Systems, volume 31, pages 7167–7177, 2018

    work page 2018

  29. [29]

    Fourier neural operator for parametric partial dif- ferential equations

    Zongyi Li, Nikola Kovachki, Kamyar Azizzadenesheli, Burigede Liu, Kaushik Bhattacharya, Andrew Stuart, and Anima Anandkumar. Fourier neural operator for parametric partial dif- ferential equations. InInternational Conference on Learning Representations, 2021. URL https://openreview.net/forum?id=c8P9NQVtmnO

    work page 2021

  30. [30]

    Fourier neural operator with learned deformations for PDEs on general geometries.Journal of Machine Learn- ing Research, 24(388), 2023

    Zongyi Li, Daniel Zhengyu Huang, Burigede Liu, and Anima Anandkumar. Fourier neural operator with learned deformations for PDEs on general geometries.Journal of Machine Learn- ing Research, 24(388), 2023. URL https://www.jmlr.org/papers/volume24/23-0064/ 23-0064.pdf

    work page 2023

  31. [31]

    Physics-informed neural operator for learning partial differential equations, 2024

    Zongyi Li, Hongkai Zheng, Nikola Kovachki, David Jin, Haoxuan Chen, Burigede Liu, Kamyar Azizzadenesheli, and Anima Anandkumar. Physics-informed neural operator for learning partial differential equations, 2024. URLhttps://arxiv.org/abs/2111.03794

    work page arXiv 2024

  32. [32]

    Veeling, Paris Perdikaris, Richard E

    Phillip Lippe, Bastiaan S. Veeling, Paris Perdikaris, Richard E. Turner, and Jo- hannes Brandstetter. PDE-Refiner: Achieving accurate long rollouts with neu- ral PDE solvers. InAdvances in Neural Information Processing Systems, 2023. 11 URL https://proceedings.neurips.cc/paper_files/paper/2023/hash/ d529b943af3dba734f8a7d49efcb6d09-Abstract-Conference.html

    work page 2023

  33. [33]

    PROSE-FD: A multimodal PDE foundation model for learning multiple operators for forecasting fluid dynamics, 2024

    Yuxuan Liu, Jingmin Sun, Xinjie He, Griffin Pinney, Zecheng Zhang, and Hayden Schaeffer. PROSE-FD: A multimodal PDE foundation model for learning multiple operators for forecasting fluid dynamics, 2024. URLhttps://arxiv.org/abs/2409.09811

    work page arXiv 2024

  34. [34]

    Kevin Denamgana¨ ı, Sondess Missaoui, and James Alfred Walker

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A ConvNet for the 2020s. InIEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 11966–11976, 2022. doi: 10.1109/CVPR52688.2022.01167

    work page doi:10.1109/cvpr52688.2022.01167 2022

  35. [35]

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators

    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. doi: 10.1038/s42256-021-00302-5

    work page doi:10.1038/s42256-021-00302-5 2021

  36. [36]

    Academic Press, 2 edition, 1999

    Stéphane Mallat.A Wavelet Tour of Signal Processing. Academic Press, 2 edition, 1999

    work page 1999

  37. [37]

    Multiple physics pretraining for physical surrogate models, 2023

    Michael McCabe, Bruno Régaldo-Saint Blancard, Liam Holden Parker, Ruben Ohana, Miles Cranmer, Alberto Bietti, Michael Eickenberg, Siavash Golkar, Geraud Krawezik, Francois Lanusse, Mariel Pettee, Tiberiu Tesileanu, Kyunghyun Cho, and Shirley Ho. Multiple physics pretraining for physical surrogate models, 2023. URL https://arxiv.org/abs/2310. 02994

    work page 2023

  38. [38]

    Michael McCabe, Payel Mukhopadhyay, Tanya Marwah, Bruno Regaldo-Saint Blancard, Fran- cois Rozet, Cristiana Diaconu, Lucas Meyer, Kaze W. K. Wong, Hadi Sotoudeh, Alberto Bietti, Irina Espejo, Rio Fear, Siavash Golkar, Tom Hehir, Keiya Hirashima, Geraud Krawezik, Fran- cois Lanusse, Rudy Morel, Ruben Ohana, Liam Parker, Mariel Pettee, Jeff Shen, Kyunghyun ...

    work page 2025

  39. [39]

    JAWS: Enhancing long-term rollout of neural PDE solvers via spatially-adaptive Jacobian regularization, 2026

    Fengxiang Nie and Yasuhiro Suzuki. JAWS: Enhancing long-term rollout of neural PDE solvers via spatially-adaptive Jacobian regularization, 2026. URL https://arxiv.org/abs/2603. 05538

    work page 2026

  40. [40]

    FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

    Jaideep Pathak, Shashank Subramanian, Peter Harrington, Sanjeev Raja, Ashesh Chattopadhyay, Morteza Mardani, Thorsten Kurth, David Hall, Zongyi Li, Kamyar Azizzadenesheli, Pedram Hassanzadeh, Karthik Kashinath, and Anima Anandkumar. FourCastNet: A global data- driven high-resolution weather model using adaptive Fourier neural operators, 2022. URL https://...

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  41. [41]

    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,

    work page

  42. [42]

    URLhttps://openreview.net/forum?id=roNqYL0_XP

    work page

  43. [43]

    Pope.Turbulent Flows

    Stephen B. Pope.Turbulent Flows. Cambridge University Press, Cambridge, UK, 2001. ISBN 9780521598866. doi: 10.1017/CBO9780511840531

    work page doi:10.1017/cbo9780511840531 2001

  44. [44]

    Ross, and Kamyar Azizzadenesheli

    Md Ashiqur Rahman, Zachary E. Ross, and Kamyar Azizzadenesheli. U-NO: U-shaped neural operators.Transactions on Machine Learning Research, 2022. URL https://openreview. net/forum?id=j3oQF9coJd

    work page 2022

  45. [45]

    Maziar Raissi, Paris Perdikaris, and George Em Karniadakis. Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations.Journal of Computational Physics, 378:686–707, 2019. doi: 10.1016/j.jcp.2018.10.045

    work page doi:10.1016/j.jcp.2018.10.045 2019

  46. [46]

    Dynam- icViT: Efficient vision transformers with dynamic token sparsification

    Yongming Rao, Wenliang Zhao, Benlin Liu, Jiwen Lu, Jie Zhou, and Cho-Jui Hsieh. Dynam- icViT: Efficient vision transformers with dynamic token sparsification. InAdvances in Neural Information Processing Systems, volume 34, 2021. URL https://proceedings.neurips. cc/paper/2021/hash/747d3443e319a22747fbb873e8b2f9f2-Abstract.html. 12

    work page 2021

  47. [47]

    Convolutional neural operators for robust and accurate learning of PDEs

    Bogdan Raoni´c, Roberto Molinaro, Tim De Ryck, Tobias Rohner, Francesca Bartolucci, Rima Alaifari, Siddhartha Mishra, and Emmanuel de Bézenac. Convolutional neural operators for robust and accurate learning of PDEs. InAdvances in Neural Information Processing Systems,

    work page

  48. [48]

    URLhttps://openreview.net/forum?id=MtekhXRP4h

    work page

  49. [49]

    Love, Alexander Scheinker, Diane Oyen, Nathan Debardeleben, Earl Lawrence, and Ayan Biswas

    Mahindra Singh Rautela, Alexander Most, Siddharth Mansingh, Bradley C. Love, Alexander Scheinker, Diane Oyen, Nathan Debardeleben, Earl Lawrence, and Ayan Biswas. MORPH: PDE foundation models with arbitrary data modality, 2025

    work page 2025

  50. [50]

    , author Fischer, P

    Olaf Ronneberger, Philipp Fischer, and Thomas Brox. U-Net: Convolutional networks for biomedical image segmentation. InMedical Image Computing and Computer-Assisted Inter- vention (MICCAI), pages 234–241, 2015. doi: 10.1007/978-3-319-24574-4_28

    work page doi:10.1007/978-3-319-24574-4_28 2015

  51. [51]

    The best of both worlds: Hybridizing neural operators and solvers for stable long-horizon inference, 2025

    Rajyasri Roy, Dibyajyoti Nayak, and Somdatta Goswami. The best of both worlds: Hybridizing neural operators and solvers for stable long-horizon inference, 2025. URL https://arxiv. org/abs/2512.19643

    work page arXiv 2025

  52. [52]

    Ryoo, AJ Piergiovanni, Anurag Arnab, Mostafa Dehghani, and Anelia Angelova

    Michael S. Ryoo, AJ Piergiovanni, Anurag Arnab, Mostafa Dehghani, and Anelia Angelova. Tokenlearner: Adaptive space-time tokenization for videos. InAdvances in Neural Information Processing Systems, volume 34, 2021. URL https://proceedings.neurips.cc/paper/ 2021/hash/6a30e32e56fce5cf381895dfe6ca7b6f-Abstract.html

    work page 2021

  53. [53]

    Battaglia

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter W. Battaglia. Learning to simulate complex physics with graph networks. InInternational Conference on Machine Learning, 2020. URL https://proceedings.mlr.press/v119/ sanchez-gonzalez20a.html

    work page 2020

  54. [54]

    Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra

    Ramprasaath R. Selvaraju, Michael Cogswell, Abhishek Das, Ramakrishna Vedantam, Devi Parikh, and Dhruv Batra. Grad-CAM: Visual explanations from deep networks via gradient- based localization. InIEEE International Conference on Computer Vision, pages 618–626,

    work page

  55. [56]

    Zebra: In-context generative pretraining for solving parametric PDEs

    Louis Serrano, Armand Kassaï Koupaï, Thomas X Wang, Pierre Erbacher, and Patrick Gallinari. Zebra: In-context generative pretraining for solving parametric PDEs. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 53940–53988. PMLR, 2025. URL https://proceedings.mlr. press/v2...

    work page 2025

  56. [57]

    Sebastian Seung, Manfred Opper, and Haim Sompolinsky

    H. Sebastian Seung, Manfred Opper, and Haim Sompolinsky. Query by committee. In Proceedings of the Fifth Annual Workshop on Computational Learning Theory, pages 287–294,

    work page

  57. [58]

    doi: 10.1145/130385.130417

    work page doi:10.1145/130385.130417

  58. [59]

    Le, Geoffrey E

    Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc V . Le, Geoffrey E. Hinton, and Jeff Dean. Outrageously large neural networks: The sparsely-gated mixture- of-experts layer. InInternational Conference on Learning Representations, 2017. URL https://openreview.net/forum?id=B1ckMDqlg

    work page 2017

  59. [60]

    A physics-informed diffusion model for high- fidelity flow field reconstruction.Journal of Computational Physics, 478:111972, 2023

    Dule Shu, Zijie Li, and Amir Barati Farimani. A physics-informed diffusion model for high- fidelity flow field reconstruction.Journal of Computational Physics, 478:111972, 2023. doi: 10.1016/j.jcp.2023.111972

    work page doi:10.1016/j.jcp.2023.111972 2023

  60. [61]

    Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps

    Karen Simonyan, Andrea Vedaldi, and Andrew Zisserman. Deep inside convolutional networks: Visualising image classification models and saliency maps, 2013. URL https://arxiv.org/ abs/1312.6034

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  61. [62]

    Fielding, Dmitrii Kochkov, Miles Cranmer, Tobias Pfaff, Jonathan Godwin, Can Cui, Shirley Ho, Peter Battaglia, and Alvaro Sanchez-Gonzalez

    Kim Stachenfeld, Drummond B. Fielding, Dmitrii Kochkov, Miles Cranmer, Tobias Pfaff, Jonathan Godwin, Can Cui, Shirley Ho, Peter Battaglia, and Alvaro Sanchez-Gonzalez. Learned simulators for turbulence. InInternational Conference on Learning Representations, 2022. URLhttps://openreview.net/forum?id=msRBojTz-Nh

    work page 2022

  62. [63]

    ReAct: Out-of-distribution detection with rectified activations

    Yiyou Sun, Chuan Guo, and Yixuan Li. ReAct: Out-of-distribution detection with rectified activations. InAdvances in Neural Information Processing Systems, volume 34, 2021. URL https://openreview.net/forum?id=IBVBtz_sRSm. 13

    work page 2021

  63. [64]

    Out-of-distribution detection with deep nearest neighbors

    Yiyou Sun, Yifei Ming, Xiaojin Zhu, and Yixuan Li. Out-of-distribution detection with deep nearest neighbors. InInternational Conference on Machine Learning, pages 20827–20840,

    work page

  64. [65]

    URLhttps://proceedings.mlr.press/v162/sun22d.html

    work page

  65. [66]

    PDEBench: An extensive benchmark for scientific machine learning

    Makoto Takamoto, Timothy Praditia, Raphael Leiteritz, Dan MacKinlay, Francesco Alesiani, Dirk Pflüger, and Mathias Niepert. PDEBench: An extensive benchmark for scientific machine learning. InAdvances in Neural Information Processing Systems, 2022

    work page 2022

  66. [67]

    Mean teachers are better role models: Weight- averaged consistency targets improve semi-supervised deep learning results

    Antti Tarvainen and Harri Valpola. Mean teachers are better role models: Weight- averaged consistency targets improve semi-supervised deep learning results. In Advances in Neural Information Processing Systems, volume 30, pages 1195– 1204, 2017. URL https://proceedings.neurips.cc/paper/2017/hash/ 68053af2923e00204c3ca7c6a3150cf7-Abstract.html

    work page 2017

  67. [68]

    American Mathe- matical Society, 2024

    Roger Temam.Navier–Stokes Equations: Theory and Numerical Analysis. American Mathe- matical Society, 2024

    work page 2024

  68. [69]

    Solver-in-the-loop: Learning from differentiable physics to interact with itera- tive PDE-solvers

    Kiwon Um, Robert Brand, Yun Raymond Fei, Philipp Holl, and Nils Thuerey. Solver-in-the-loop: Learning from differentiable physics to interact with itera- tive PDE-solvers. InAdvances in Neural Information Processing Systems, vol- ume 33, 2020. URL https://proceedings.neurips.cc/paper/2020/hash/ 43e4e6a6f341e00671e123714de019a8-Abstract.html

    work page 2020

  69. [70]

    TENT: Fully test-time adaptation by entropy minimization

    Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. TENT: Fully test-time adaptation by entropy minimization. InInternational Conference on Learning Representations (ICLR), 2021. URLhttps://openreview.net/forum?id=uXl3bZLkr3c

    work page 2021

  70. [71]

    INC: An indirect neural corrector for auto-regressive hybrid PDE solvers

    Hao Wei, Aleksandra Franz, Bjoern List, and Nils Thuerey. INC: An indirect neural corrector for auto-regressive hybrid PDE solvers. InAdvances in Neural Information Processing Systems,

    work page

  71. [72]

    URLhttps://arxiv.org/abs/2511.12764

    doi: 10.48550/arXiv.2511.12764. URLhttps://arxiv.org/abs/2511.12764

    work page doi:10.48550/arxiv.2511.12764

  72. [73]

    Physics-informed temporal alignment for auto-regressive PDE foundation models

    Congcong Zhu, Xiaoyan Xu, Jiayue Han, and Jingrun Chen. Physics-informed temporal alignment for auto-regressive PDE foundation models. InProceedings of the 42nd Interna- tional Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 80223–80258. PMLR, 2025. URL https://proceedings.mlr.press/ v267/zhu25w.html

    work page 2025

  73. [74]

    Orb: An efficient alternative to sift or surf

    Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. InIEEE International Conference on Computer Vision, pages 2242–2251, 2017. doi: 10.1109/ICCV .2017.244. 14 Table 3: Notation used throughout the paper. Symbol Meaning xt,x ∗ t+1 State at timet; ground-truth state ...

    work page doi:10.1109/iccv 2017

  74. [75]

    Compute the eight per-run ratiosρ r =L method r /Lraw r

    work page

  75. [76]

    , B boot = 10,000 bootstrap iterations, draw ρ(b) 1 ,

    For b= 1, . . . , B boot = 10,000 bootstrap iterations, draw ρ(b) 1 , . . . , ρ(b) 8 with replacement from{ρ r}8 r=1, and recordm (b) = median(ρ(b))

    work page

  76. [77]

    Ratio” is the median ten-step UV relative- L2 relative to raw Poseidon; “× Raw

    Report eRmethod together with the 2.5th and 97.5th percentiles of the bootstrap median distribution{m (b)}B b=1. On every one of the ten cells, a paired bootstrap on the per-run difference ρARC-STAR r −ρ glob r (Bboot = 10,000, seed 42) yields a 95% confidence interval that strictly excludes zero. Because the sample space is discrete with n=8 paired runs,...

    work page arXiv 2000