pith. sign in

arxiv: 2606.29723 · v1 · pith:2YOAQEPZnew · submitted 2026-06-29 · 💻 cs.LG · astro-ph.IM· cs.CV· physics.comp-ph

ScaleAware-JEPA: Latent Representation for Discovery in Multiscale Physical Fields

Guang-Xing Li This is my paper

Pith reviewed 2026-06-30 07:44 UTC · model grok-4.3

classification 💻 cs.LG astro-ph.IMcs.CVphysics.comp-ph
keywords self-supervised learningmultiscale physical fieldslatent representationsJEPAdiffusion decompositionstructural atlaseslabel-free discovery
0
0 comments X

The pith

ScaleAware-JEPA builds label-free latent coordinates for multiscale physical fields by aligning prediction masks to diffusion scale components.

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

The paper presents a self-supervised framework that learns dense representations of continuous scalar fields such as turbulence and molecular gas. It first applies Constrained Diffusion Decomposition to split each field into pixel-registered scale components, then uses those components to set the context and target regions for a JEPA prediction task. The resulting latent geometry recovers coherent morphological structures across tested datasets without any labels or hand-crafted segmentation rules. A reader would care if the approach lets scientists inspect complex patterns in raw field data before deciding which features matter.

Core claim

By tying the JEPA objective to the scale hierarchy provided by Constrained Diffusion Decomposition, ScaleAware-JEPA generates latent representations that map back to coherent morphology in fields such as MHD turbulence, interstellar molecular gas, and urban nighttime-light structure, forming dense structural atlases without labels or predefined segmentation rules.

What carries the argument

Constrained Diffusion Decomposition (CDD), which separates each field into pixel-registered scale components and supplies the scale coordinates that define the masking geometry for the JEPA predictive task.

If this is right

  • The learned geometry forms dense structural atlases without labels or predefined segmentation rules.
  • Latent prediction is performed with a context footprint tied to the diffusion scale of each component rather than an arbitrary patch size.
  • Complex physical patterns can be inspected before their relevant structures have been prescribed.
  • The same pipeline produces usable coordinates across MHD turbulence, interstellar molecular gas, and urban nighttime-light structure.

Where Pith is reading between the lines

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

  • The same scale-tied masking strategy might transfer to other continuous fields such as climate or fluid-flow data where hierarchical organization is present but unlabeled.
  • Comparing the recovered atlases against known physical catalogs in the tested domains would provide a direct check on whether the latents capture established features.
  • Extending the framework to vector or tensor fields could test whether the scale-coordinate principle generalizes beyond scalar data.

Load-bearing premise

The scale coordinates supplied by Constrained Diffusion Decomposition define a masking geometry that produces a predictive task aligned with the field's intrinsic multiscale organization.

What would settle it

Running the method on an additional multiscale field and finding that the learned latent coordinates show no correspondence to any coherent morphological features identified by independent analysis would falsify the alignment claim.

Figures

Figures reproduced from arXiv: 2606.29723 by Guang-Xing Li.

Figure 1
Figure 1. Figure 1: ScaleAware-JEPA architecture and design. (a) Architecture: the CDD frontend decomposes the raw field into scale-separated components. The context branch applies scale-aware masking and encodes the masked context; the target branch bypasses masking and uses an EMA-updated copy. A lightweight predictor maps context to target latent space. Training combines latent prediction and a weak spread regularizer. (b)… view at source ↗
Figure 2
Figure 2. Figure 2: Masking-strategy diagnostics for the MHD sweep. Left: target effective rank and hinge ratio as functions of the pyramid-mask footprint multiplier. Right: the same diagnostics for fixed-box masks. The pyramid sweep shows a gradual increase in both diagnostics through 1.2×σs, followed by a sharp rise in target effective rank at 1.6×σs and near-complete hinge saturation at 2.0×σs. The selected 1.2×σs setting … view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Back-mapping representative latent neighborhoods in the MHD density field. Left: the input density map with selected latent groups overlaid at their original spatial locations. Right: the corresponding three-dimensional UMAP projection of target-encoder embeddings, with the same groups highlighted in matching colors (rendered with the default perspective projection rather than orthographic, so the 3D point… view at source ↗
Figure 5
Figure 5. Figure 5 [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Dense latent topology learned for the NGC 3627 molecular-gas field. Target-encoder embeddings of the PHANGS–ALMA CO field are projected with PCA and UMAP and then mapped back to their original spatial locations. The PCA projection captures large-scale latent variation across the molecular disk, separating the bright central concentration and spiral-arm molecular gas from lower-surface￾brightness interarm e… view at source ↗
Figure 7
Figure 7. Figure 7: CDD versus log-normal wavelet decomposition. Each panel compares the CDD primitive (left half) with the matched log-normal wavelet coefficient (right half) across six scales on an MHD turbulence field. CDD produces sparse, spatially compact features with near-zero response in smooth regions; the wavelet representation exhibits diffuse oscillatory leakage and ringing artefacts that are absent in the CDD pri… view at source ↗
Figure 8
Figure 8. Figure 8: Wavelet-frontend control on the MHD field. A matched JEPA run using a wavelet frontend produces an organized UMAP manifold, but the mapped-back UMAP RGB field shows broad halo-like bands around high-contrast structures. This indicates a non-collapsed but less localized frontend behavior, consistent with oscillatory leakage in the wavelet decomposition and with ringing near sharp transitions [PITH_FULL_IMA… view at source ↗
Figure 9
Figure 9. Figure 9: Selected-run input fields and training losses. For each selected vanilla run, the left panel shows the scalar input field and the right panel shows the corresponding training-loss history. Rows corre￾spond to MHD turbulence, Chengdu nighttime lights, and the NGC molecular-gas field, respectively. MHD uses pyramid masking with footprint 1.2 × σs; Chengdu uses 0.8 × σs; NGC uses 1.6 × σs, where σs is the CDD… view at source ↗
Figure 10
Figure 10. Figure 10: Plain ConvNeXt control versus the scale-aware encoder. Comparison on a matched random-mask example. The dense ConvNeXt control receives only the masked scalar field and a binary mask-indicator channel, whereas ScaleAware-JEPA receives pixel-registered multiscale CDD components. In the control runs, varying the box footprint from 7 to 19 px does not produce a meaningful change in sampled embedding usage an… view at source ↗
read the original abstract

Continuous physical fields represent a large fraction of data under scientific investigation. Their multiscale structures are central to discovery, yet useful coordinates are not known in advance. Standard self-supervised methods define context and targets in fixed image coordinates, posing a predictive task misaligned with fields organized across a continuous scale hierarchy. We introduce ScaleAware-JEPA, a framework that constructs dense, label-free latent coordinates for continuous scalar fields. Constrained Diffusion Decomposition (CDD) separates each field into pixel-registered scale components and provides the scale coordinates that define the masking geometry. The resulting JEPA objective predicts hidden structure with a context footprint tied to the diffusion scale of each component rather than to an arbitrary patch size. Across MHD turbulence, interstellar molecular gas and urban nighttime-light structure, the learned geometry maps back to coherent morphology, forming dense structural atlases without labels or predefined segmentation rules. By tying latent prediction to the scale hierarchy of a field, ScaleAware-JEPA constructs latent coordinates through which complex physical patterns can be inspected before their relevant structures have been prescribed. Code is available at https://github.com/gxli/SA-JEPA.

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 paper introduces ScaleAware-JEPA, which combines Constrained Diffusion Decomposition (CDD) with a JEPA-style self-supervised objective to learn dense latent coordinates for continuous scalar fields. CDD supplies pixel-registered scale components whose diffusion scales define the context/target masking geometry; the resulting representations are claimed to map to coherent morphological structures across MHD turbulence, interstellar molecular gas, and urban nighttime-light fields, yielding label-free structural atlases.

Significance. If the central claim holds, the framework offers a route to inspect multiscale physical patterns before structures are prescribed, with potential utility in astrophysics and fluid dynamics. Public code release is a clear strength supporting reproducibility.

major comments (2)
  1. [Abstract and §3] Abstract and §3: the assertion that ScaleAware-JEPA forms atlases 'without labels or predefined segmentation rules' is load-bearing for the discovery claim, yet CDD is introduced with explicit scale-separation parameters and pixel-registration constraints; it is not shown whether these parameters recover a data-intrinsic hierarchy or impose an external one, leaving open the possibility that the JEPA objective simply learns within CDD's inductive bias rather than discovering field-intrinsic organization.
  2. [Empirical results] Empirical results (across MHD, molecular gas, and nighttime lights): the claim that 'the learned geometry maps back to coherent morphology' is central but requires quantitative support (e.g., overlap metrics with known structures, ablation on CDD scale parameters, or comparison against standard patch-based JEPA); without such evidence the results risk reflecting CDD's decomposition more than emergent discovery.
minor comments (2)
  1. [§3] Notation for CDD scale coordinates and the precise form of the JEPA objective (context footprint tied to diffusion scale) should be formalized with an equation in §3 to allow direct inspection of the masking geometry.
  2. The abstract states code is available; the repository should include the full CDD implementation and hyper-parameter settings used for each dataset.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below.

read point-by-point responses

  1. Referee: [Abstract and §3] Abstract and §3: the assertion that ScaleAware-JEPA forms atlases 'without labels or predefined segmentation rules' is load-bearing for the discovery claim, yet CDD is introduced with explicit scale-separation parameters and pixel-registration constraints; it is not shown whether these parameters recover a data-intrinsic hierarchy or impose an external one, leaving open the possibility that the JEPA objective simply learns within CDD's inductive bias rather than discovering field-intrinsic organization.

    Authors: We agree that the role of CDD parameters requires explicit clarification to support the discovery claim. The scale-separation parameters are chosen according to each field's measured diffusion properties to produce pixel-registered components; they do not encode morphological labels or segmentation rules. The JEPA objective then learns cross-scale predictions on these components. We will revise the abstract and §3 to state the data-driven selection criterion for the parameters and to distinguish the CDD decomposition step from the subsequent label-free learning performed by the JEPA objective. revision: partial

  2. Referee: [Empirical results] Empirical results (across MHD, molecular gas, and nighttime lights): the claim that 'the learned geometry maps back to coherent morphology' is central but requires quantitative support (e.g., overlap metrics with known structures, ablation on CDD scale parameters, or comparison against standard patch-based JEPA); without such evidence the results risk reflecting CDD's decomposition more than emergent discovery.

    Authors: We accept that quantitative evidence is needed to substantiate the mapping to coherent morphology. In the revised manuscript we will report overlap metrics against known structures (where available, e.g., in the MHD case), include ablations that vary the CDD scale parameters, and add a direct comparison against a standard patch-based JEPA baseline. These additions will allow readers to assess the contribution of the scale-tied objective relative to the decomposition alone. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation remains self-contained against external benchmarks

full rationale

The paper introduces ScaleAware-JEPA by using CDD-derived scale components to define JEPA masking geometry, with the objective tied to diffusion scales rather than arbitrary patches. No equation or step reduces a claimed prediction or discovery result to a fitted parameter or input definition by construction. The central assertion—that learned geometry maps to coherent morphology forming label-free atlases—does not equate to the CDD inputs; it is presented as an empirical outcome across MHD, molecular gas, and nighttime-light fields. CDD is treated as an external decomposition tool providing coordinates, and the JEPA training is a separate predictive task. No self-citation chain, uniqueness theorem, or ansatz smuggling is load-bearing in the provided text. This is the common honest finding for methods that combine existing techniques without the output being definitionally identical to the inputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 1 invented entities

Abstract-only review supplies insufficient detail to enumerate free parameters or background axioms; the only identifiable invented entity is Constrained Diffusion Decomposition.

invented entities (1)
  • Constrained Diffusion Decomposition (CDD) no independent evidence
    purpose: separates each field into pixel-registered scale components and supplies the scale coordinates that define the masking geometry
    Introduced in the abstract as the component that replaces arbitrary patch sizes with diffusion-scale footprints.

pith-pipeline@v0.9.1-grok · 5728 in / 1199 out tokens · 35907 ms · 2026-06-30T07:44:46.384255+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

85 extracted references · 29 canonical work pages · 20 internal anchors

  1. [1]

    and Hohenberg, Pierre C

    Cross, Michael C. and Hohenberg, Pierre C. , title =. Reviews of Modern Physics , volume =. 1993 , doi =

    1993

  2. [2]

    Philosophical Transactions of the Royal Society of London

    Turing, Alan Mathison , title =. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences , volume =. 1952 , doi =

    1952

  3. [3]

    and Brendel, Wieland , title =

    Geirhos, Robert and Rubisch, Patricia and Michaelis, Claudio and Bethge, Matthias and Wichmann, Felix A. and Brendel, Wieland , title =. International Conference on Learning Representations , year =

  4. [4]

    and Chen, Ting and Kornblith, Simon , title =

    Hermann, Katherine L. and Chen, Ting and Kornblith, Simon , title =. Advances in Neural Information Processing Systems , volume =. 2020 , url =

    2020

  5. [5]

    Monthly Notices of the Royal Astronomical Society , volume =

    Li, Guang-Xing and Zhao, Mengke , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2025 , doi =. 2409.02769 , archivePrefix=

    work page arXiv 2025

  6. [6]

    arXiv preprint arXiv:2509.20832 , year =

    Zhao, Mengke and Li, Guang-Xing , title =. arXiv preprint arXiv:2509.20832 , year =. doi:10.48550/arXiv.2509.20832 , url =. 2509.20832 , archivePrefix =

    work page doi:10.48550/arxiv.2509.20832

  7. [7]

    LeWorldModel: Stable End-to-End Joint-Embedding Predictive Architecture from Pixels

    Maes, Lucas and Le Lidec, Quentin and Scieur, Damien and LeCun, Yann and Balestriero, Randall , title =. arXiv preprint arXiv:2603.19312 , year =. doi:10.48550/arXiv.2603.19312 , url =. 2603.19312 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2603.19312

  8. [8]

    and Roli, Andrea , title =

    Kauffman, Stuart A. and Roli, Andrea , title =. Interface Focus , volume =. 2023 , doi =

    2023

  9. [9]

    , title =

    Kauffman, Stuart A. , title =. 1996 , institution =

    1996

  10. [10]

    The Multiscale Morphology Filter: Identifying and Extracting Spatial Patterns in the Galaxy Distribution , journal =

    Arag. The Multiscale Morphology Filter: Identifying and Extracting Spatial Patterns in the Galaxy Distribution , journal =. 2007 , doi =. 0705.2072 , archivePrefix=

    work page arXiv 2007

  11. [11]

    Gibbs Phenomenon of Framelet Expansions and Quasi-projection Approximation

    Han, Bin , title =. Journal of Fourier Analysis and Applications , volume =. 2019 , doi =. 1808.09414 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  12. [12]

    1922 , publisher =

    Richardson, Lewis Fry , title =. 1922 , publisher =

    1922

  13. [13]

    Doklady Akademii Nauk SSSR , volume =

    Kolmogorov, Andrey Nikolaevich , title =. Doklady Akademii Nauk SSSR , volume =

  14. [14]

    Masked Autoencoders Are Scalable Vision Learners

    He, Kaiming and Chen, Xinlei and Xie, Saining and Li, Yanghao and Doll. Masked Autoencoders Are Scalable Vision Learners , booktitle =. 2022 , doi =. 2111.06377 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  15. [15]

    Self-Supervised Learning from Im- ages with a Joint-Embedding Predictive Architecture, 2023

    Assran, Mahmoud and Duval, Quentin and Misra, Ishan and Bojanowski, Piotr and Vincent, Pascal and Rabbat, Michael and LeCun, Yann and Ballas, Nicolas , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =. 2023 , doi =. 2301.08243 , archivePrefix=

    work page arXiv 2023

  16. [16]

    Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , booktitle =

    Grill, Jean-Bastien and Strub, Florian and Altch. Bootstrap Your Own Latent: A New Approach to Self-Supervised Learning , booktitle =. 2020 , url =

    2020

  17. [17]

    Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

    Chen, Xinlei and Xie, Saining and He, Kaiming , title =. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =. 2021 , eprint =

    2021

  18. [18]

    LeJEPA: Provable and Scalable Self-Supervised Learning Without the Heuristics

    Balestriero, Randall and LeCun, Yann , title =. 2025 , doi =. 2511.08544 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  19. [19]

    , title =

    Jolliffe, Ian T. , title =. 2002 , publisher =

    2002

  20. [20]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    McInnes, Leland and Healy, John and Melville, James , title =. arXiv preprint arXiv:1802.03426 , year =. doi:10.48550/arXiv.1802.03426 , url =. 1802.03426 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426

  21. [21]

    Journal of Open Source Software , volume =

    McInnes, Leland and Healy, John and Saul, Nathaniel and Grossberger, Lukas , title =. Journal of Open Source Software , volume =. 2018 , doi =

    2018

  22. [22]

    Proceedings of the 37th International Conference on Machine Learning , volume =

    Chen, Ting and Kornblith, Simon and Norouzi, Mohammad and Hinton, Geoffrey , title =. Proceedings of the 37th International Conference on Machine Learning , volume =. 2020 , publisher =

    2020

  23. [23]

    Advances in Neural Information Processing Systems , volume =

    Balestriero, Randall and LeCun, Yann , title =. Advances in Neural Information Processing Systems , volume =

  24. [24]

    Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style , booktitle =

    von K. Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style , booktitle =. 2021 , eprint =

    2021

  25. [25]

    Proceedings of the 40th International Conference on Machine Learning , volume =

    Garrido, Quentin and Balestriero, Randall and Najman, Laurent and LeCun, Yann , title =. Proceedings of the 40th International Conference on Machine Learning , volume =. 2023 , publisher =

    2023

  26. [26]

    The information bottleneck method

    Tishby, Naftali and Pereira, Fernando C and Bialek, William , title =. arXiv preprint physics/0004057 , year =. physics/0004057 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  27. [27]

    Crawling the Cosmic Network: Identifying and Quantifying Filamentary Structure

    Bond, Nicholas A. and Strauss, Michael A. and Cen, Renyue , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2010 , doi =. 1003.3237 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  28. [28]

    Structural Analysis of Molecular Clouds: Dendrograms

    Rosolowsky, Erik W. and Pineda, Jaime E. and Kauffmann, Jens and Goodman, Alyssa A. , title =. The Astrophysical Journal , volume =. 2008 , doi =. 0802.2944 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2008

  29. [29]

    Platen, Erwin and van de Weygaert, Rien and Jones, Bernard J. T. , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2007 , doi =. 0706.2788 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2007

  30. [30]

    The Stellar IMF from Turbulent Fragmentation

    Padoan, Paolo and Nordlund,. The Stellar. The Astrophysical Journal , volume =. 2002 , doi =. astro-ph/0011465 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  31. [31]

    Modeling Collapse and Accretion in Turbulent Gas Clouds: Implementation and Comparison of Sink Particles in AMR and SPH

    Federrath, Christoph and Banerjee, Robi and Clark, Paul C and Klessen, Ralf S , title =. The Astrophysical Journal , volume =. 2010 , doi =. 1001.4456 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  32. [32]

    The persistent cosmic web and its filamentary structure I: Theory and implementation

    Sousbie, Thierry , title =. Monthly Notices of the Royal Astronomical Society , volume =. 2011 , doi =. 1009.4015 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  33. [33]

    A Dynamical Classification of the Cosmic Web

    Forero-Romero, JE and Hoffman, Yehuda and Gottl. A Dynamical Classification of the Cosmic Web , journal =. 2009 , doi =. 0809.4135 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  34. [34]

    , title =

    Murray, James D. , title =. 2003 , publisher =

    2003

  35. [35]

    How Filaments are Woven into the Cosmic Web

    Bond, J. Richard and Kofman, Lev and Pogosyan, Dmitry , title =. Nature , volume =. 1996 , doi =. astro-ph/9512141 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 1996

  36. [36]

    and Scalo, John , title =

    Elmegreen, Bruce G. and Scalo, John , title =. Annual Review of Astronomy and Astrophysics , volume =. 2004 , doi =

    2004

  37. [37]

    Particles and fields in fluid turbulence

    Falkovich, Gregory and Gaw. Particles and Fields in Fluid Turbulence , journal =. 2001 , doi =. cond-mat/0105199 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2001

  38. [38]

    1995 , publisher =

    Frisch, Uriel , title =. 1995 , publisher =

    1995

  39. [39]

    , title =

    Pope, Stephen B. , title =. 2000 , publisher =

    2000

  40. [40]

    Biometrika , volume =

    Epps, Thomas W and Pulley, Lawrence B , title =. Biometrika , volume =. 1983 , doi =

    1983

  41. [41]

    Proceedings of the 45th Annual

    Meng, Xiangrui and Mahoney, Michael W , title =. Proceedings of the 45th Annual. 2013 , doi =

    2013

  42. [42]

    VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning

    Bardes, Adrien and Ponce, Jean and LeCun, Yann , title =. International Conference on Learning Representations , year =. 2105.04906 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  43. [43]

    Accurate, Large Minibatch SGD: Training ImageNet in 1 Hour

    Goyal, Priya and Doll. Accurate, Large Minibatch. arXiv preprint arXiv:1706.02677 , year =. 1706.02677 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    and Noack, Bernd R

    Brunton, Steven L. and Noack, Bernd R. and Koumoutsakos, Petros , title =. Annual Review of Fluid Mechanics , volume =. 2020 , doi =

    2020

  45. [45]

    Machine Learning for Turbulent Flow Control , journal =

    Jim. Machine Learning for Turbulent Flow Control , journal =. 2018 , doi =

    2018

  46. [46]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    Bengio, Yoshua and Courville, Aaron and Vincent, Pascal , title =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =. 2013 , doi =

    2013

  47. [47]

    Testing the Manifold Hypothesis

    Fefferman, Charles and Mitter, Sanjoy and Narayanan, Hariharan , title =. Journal of the American Mathematical Society , volume =. 2016 , doi =. 1310.0425 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  48. [48]

    Dynamical Systems and Turbulence, Warwick 1980 , editor =

    Takens, Floris , title =. Dynamical Systems and Turbulence, Warwick 1980 , editor =. 1981 , publisher =

    1980

  49. [49]

    Spectral Properties of Dynamical Systems, Model Reduction and Decompositions , journal =

    Mezi. Spectral Properties of Dynamical Systems, Model Reduction and Decompositions , journal =. 2005 , doi =

    2005

  50. [50]

    2022 , month =

    LeCun, Yann , title =. 2022 , month =

    2022

  51. [51]

    The Astrophysical Journal Supplement Series , volume =

    Li, Guang-Xing , title =. The Astrophysical Journal Supplement Series , volume =. 2022 , doi =. 2201.05484 , archivePrefix=

    work page arXiv 2022

  52. [52]

    Group Invariant Scattering , journal =

    Mallat, St. Group Invariant Scattering , journal =. 2012 , doi =

    2012

  53. [53]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =

    Perona, Pietro and Malik, Jitendra , title =. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume =. 1990 , doi =

    1990

  54. [54]

    Communications in Mathematics and Statistics , volume =

    E, Weinan , title =. Communications in Mathematics and Statistics , volume =. 2017 , doi =

    2017

  55. [55]

    , title =

    Ericsson, Linus and Gouk, Henry and Hospedales, Timothy M. , title =. 33rd British Machine Vision Conference 2022,. 2022 , publisher =. 2111.11398 , archivePrefix=

    work page arXiv 2022

  56. [56]

    Inverse Problems , volume =

    Haber, Eldad and Ruthotto, Lars , title =. Inverse Problems , volume =. 2017 , doi =

    2017

  57. [57]

    , title =

    Kauffman, Stuart A. , title =. 2000 , publisher =

    2000

  58. [58]

    , title =

    Hewitt, Edwin and Hewitt, Robert E. , title =. Archive for History of Exact Sciences , volume =. 1979 , doi =

    1979

  59. [59]

    Proceedings of the 32nd International Conference on Machine Learning , volume =

    Ioffe, Sergey and Szegedy, Christian , title =. Proceedings of the 32nd International Conference on Machine Learning , volume =. 2015 , publisher =

    2015

  60. [60]

    Gaussian Error Linear Units (GELUs)

    Hendrycks, Dan and Gimpel, Kevin , title =. arXiv preprint arXiv:1606.08415 , year =. 1606.08415 , archivePrefix=

    work page internal anchor Pith review Pith/arXiv arXiv

  61. [61]

    Mixed Precision Training

    Micikevicius, Paulius and Narang, Sharan and Alben, Jonah and Diamos, Gregory and Elsen, Erich and Garcia, David and Ginsburg, Boris and Houston, Michael and Kuchaiev, Oleksii and Venkatesh, Ganesh and Wu, Hao , title =. arXiv preprint arXiv:1710.03740 , year =. doi:10.48550/arXiv.1710.03740 , eprint =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1710.03740

  62. [62]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =

    Liu, Zhuang and Mao, Hanzi and Wu, Chao-Yuan and Feichtenhofer, Christoph and Darrell, Trevor and Xie, Saining , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages =. 2022 , doi =. 2201.03545 , archivePrefix=

    work page arXiv 2022

  63. [63]

    Feature Pyramid Networks for Object Detection , booktitle =

    Lin, Tsung-Yi and Doll. Feature Pyramid Networks for Object Detection , booktitle =. 2017 , doi =

    2017

  64. [64]

    Advances in Neural Information Processing Systems , volume =

    Tarvainen, Antti and Valpola, Harri , title =. Advances in Neural Information Processing Systems , volume =

  65. [65]

    and Cadima, Jorge , title =

    Jolliffe, Ian T. and Cadima, Jorge , title =. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences , volume =. 2016 , doi =

    2016

  66. [66]

    Direct Observation of the Intermittency of Intense Vorticity Filaments in Turbulence , journal =

    Douady, St. Direct Observation of the Intermittency of Intense Vorticity Filaments in Turbulence , journal =. 1991 , doi =

    1991

  67. [67]

    2003 , publisher =

    Biskamp, Dieter , title =. 2003 , publisher =

    2003

  68. [68]

    and Nelson, Martin R

    Gandy, Demi L. and Nelson, Martin R. , title =. SIAM Review , volume =. 2022 , doi =

    2022

  69. [69]

    , title =

    Pearson, John E. , title =. Science , volume =. 1993 , doi =

    1993

  70. [70]

    and Pindza, Edson and Mar

    Owolabi, Kolade M. and Pindza, Edson and Mar. Turing Patterns Across Geometries: A Proven. Results in Applied Mathematics , volume =. 2025 , doi =

    2025

  71. [71]

    , title =

    Al Saadi, Fahad and Champneys, Alan R. , title =. Philosophical Transactions of the Royal Society A , volume =. 2021 , doi =

    2021

  72. [72]

    and Kritsuk, Alexei G

    Collins, David C. and Kritsuk, Alexei G. and Padoan, Paolo and Li, Hui and Xu, Hao and Ustyugov, Sergey D. and Norman, Michael L. , title =. The Astrophysical Journal , volume =. 2012 , doi =

    2012

  73. [73]

    and Appel, S

    Burkhart, B. and Appel, S. and Bialy, S. and Cho, J. and Christensen, A. J. and Collins, D. and Federrath, C. and Fielding, D. and Finkbeiner, D. and Hill, A. S. and Ibanez-Mejia, J. C. and Krumholz, M. R. and Lazarian, A. and Li, M. and Mocz, P. and Mac Low, M.-M. and Naiman, J. and Portillo, S. K. N. and Shane, B. and Slepian, Z. and Yuan, Y. , title =....

    work page arXiv 2020

  74. [74]

    1992 , publisher =

    Daubechies, Ingrid , title =. 1992 , publisher =

    1992

  75. [75]

    A Wavelet Tour of Signal Processing: The Sparse Way , year =

    Mallat, St. A Wavelet Tour of Signal Processing: The Sparse Way , year =

  76. [76]

    International Conference on Learning Representations , year =

    Loshchilov, Ilya and Hutter, Frank , title =. International Conference on Learning Representations , year =

  77. [77]

    Transactions on Machine Learning Research , year =

    Oquab, Maxime and Darcet, Timoth. Transactions on Machine Learning Research , year =

  78. [78]

    2024 , eprint =

    Bardes, Adrien and Garrido, Quentin and Ponce, Jean and Chen, Xinlei and Rabbat, Michael and LeCun, Yann and Assran, Mahmoud and Ballas, Nicolas , title =. 2024 , eprint =

    2024

  79. [79]

    2025 , eprint =

    Assran, Mido and Bardes, Adrien and Fan, David and Garrido, Quentin and Howes, Russell and Komeili, Mojtaba and Muckley, Matthew and Rizvi, Ammar and Roberts, Claire and Sinha, Koustuv and Zholus, Artem and Arnaud, Sergio and Gejji, Abha and Martin, Ada and Hogan, Francois Robert and Dugas, Daniel and Bojanowski, Piotr and Khalidov, Vasil and Labatut, Pat...

    2025

  80. [80]

    Representation Learning for Spatiotemporal Physical Systems , year =

    Qu, Helen and Morel, Rudy and McCabe, Michael and Bietti, Alberto and Lanusse, Fran. Representation Learning for Spatiotemporal Physical Systems , year =. 2603.13227 , archivePrefix =

    work page arXiv

Showing first 80 references.