Pith. sign in

REVIEW 4 major objections 4 minor 72 references

Resultant force on grains of a real sand dune: How to measure it?

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

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

Pith's one-line read A U-Net trained on simulated barchan dunes estimates the longitudinal force on individual grains from top-view images, and the authors apply it to real experimental dune images.

desk verdict A genuinely new sim-to-real force mapping idea with open code, but the experimental claim rests on unvalidated, post-hoc-adjusted outputs. read the letter →

arxiv 2501.05869 v1 pith:6QZQ5FO7 submitted 2025-01-10 physics.geo-ph

Renato F. Miotto , Carlos A. Alvarez , Danilo S. Borges , William R. Wolf , Erick M. Franklin This is my paper · ORCID
classification physics.geo-ph
keywords barchandunesgrain-scaleforcesconvolutionalneuralnetworkCFD-DEMsimulationimage-to-imagetranslationsubaqueousexperimentsdomainadaptationdeeplearning
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper claims that a convolutional neural network can turn a plain top-view image of a sand dune into a quantitative map of the longitudinal component of the resultant force acting on each grain. The network is trained entirely on CFD-DEM simulations of subaqueous barchan dunes, where the true forces are known, and is then applied to high-speed camera images of real laboratory dunes. The authors report that the network predicts force distributions on simulated dunes it has never seen, and that it produces physically sensible force maps when applied to experimental images. If the approach holds, images from cameras or satellites could become a remote sensor for grain-scale forces on Earth and Mars.

What carries the argument

The workhorse is a U-Net, an encoder-decoder convolutional network originally designed for image segmentation with limited training data. The authors treat force estimation as an image-to-image regression: the input is a binarized top-view image of the dune (binarization strips photometric noise and forces the network to focus on shape and contour), and the output is a color-coded map of the longitudinal component of the resultant force per grain. Training pairs come from CFD-DEM simulations that solve the fluid flow and the motion of individual grains, providing ground-truth forces. A domain-adversarial module with gradient reversal was tried to align simulation and experimental image statistics but was found insufficient; the final pipeline relies on input binarization, data augmentation (rotation, translation, zoom), and pooling of multiple simulation conditions in the training set.

What would settle it

Run the trained network on experimental images of a subaqueous barchan while simultaneously measuring actual grain-scale forces, for example by tracking individual grains with a high-speed camera to compute their accelerations from the trajectories, and compare the predicted force map to these measured forces. Any systematic discrepancy between the two would falsify the claim that the CNN estimates real forces from images.

Watch

Extended reading notes

Core claim

The central discovery is that the morphology of a barchan dune, as seen from above, carries enough information to recover the longitudinal component of the resultant force on its individual grains. By training a U-Net with paired inputs (binarized top-view dune images) and outputs (force-color maps from simulation), the network learns a mapping from shape to force that generalizes to dunes with different Reynolds numbers and particle counts. The same network, applied to experimental images captured in a water tank under matching flow conditions, yields force estimates on groups of grains, reproducing the expected pattern of strong downstream force at the dune's convex periphery and recirculation force in the lee. The authors are explicit that the experimental force maps cannot be directly verified because no force measurements exist for experimental dunes; validation is limited to the numerical domain.

Load-bearing premise

The load-bearing premise is that the mapping from dune shape to grain forces learned from the authors' own CFD-DEM simulations remains valid for real experimental dunes once the images are binarized, even though no experimental force measurements exist to check it.

Editorial extensions

If this is right

  • Applied to experimental images under the same flow conditions, the network yields force estimates on groups of grains without any modification.
  • The CNN generalizes to simulated dunes never seen during training, including cases with different Reynolds numbers and particle counts.
  • The same approach could be extended to remote-sensing images to estimate forces on rocks, boulders, rovers, and man-made structures on Earth and Mars.
  • By providing grain-scale force information from images alone, the method avoids the need to embed accelerometers or other sensors in the dune.

Reading between the lines

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

  • The experimental force maps in the paper's Fig. 2d should be read as model projections rather than verified measurements, since the paper itself states that no experimental ground-truth forces exist.
  • A natural next step is to test the pipeline on eolian or Martian dune images, but the shift in image statistics and grain-scale physics will likely be far larger than the simulation-to-water-tank shift.
  • Because the final pipeline depends on binarization, the accuracy of any field application is bounded by the quality of grain segmentation; misclassified grains would propagate into force errors.
  • If the morphology-to-force mapping is as general as claimed, the same architecture could be retrained to output other grain-scale fields, such as grain velocity or erosion rate, from the same images.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper proposes a deep-learning method to estimate the longitudinal component of the resultant force on individual grains of subaqueous barchan dunes from top-view images. The authors train a U-Net on synthetic images generated by their own CFD-DEM simulations, where grain forces are known, and then apply the trained network to experimental images of water-tank barchans. They report qualitative agreement between predicted and simulated mean forces and variances on unseen simulated dunes, and they present an experimental force map (Fig. 2d) obtained from an experimental image. The central claim is that the method 'can quantify these forces with high accuracy' and constitutes a breakthrough in measuring grain-scale forces from images.

Significance. If the central claim were established, the proposed technique would be a genuinely novel measurement concept with potential applications beyond dune grains, including planetary remote sensing. The paper is commendable for making data, code, and model weights publicly available, and for explicitly discussing the domain-shift problem between simulations and experiments. However, the claimed result — quantifying forces on grains of real dunes — is not actually validated in the manuscript. The experimental predictions have no ground truth, the numerical validation is only qualitative, and the experimental protocol was adjusted after observing the target-domain outputs. The significance is therefore prospective rather than demonstrated.

major comments (4)
  1. [Estimating the resultant force on grains of real dunes] The central quantitative claim about experimental dunes is unsupported. The authors explicitly write that 'there are no force measurements for experimental dunes (we have no access to Y in the experimental domain),' so the performance check is restricted to the numerical domain. The experiment shown in Fig. 2d is never compared against any measured force distribution, yet the abstract states that the method can 'quantify these forces with high accuracy.' This is a load-bearing gap: the title promises a way to measure forces on grains of a real dune, but the paper provides no evidence that the experimental force map is accurate.
  2. [Supplementary Materials, CNN model] The experimental evaluation is post-hoc and therefore not an independent test. The text reports that the domain-adversarial model 'was not sufficient to infer correct results in experimental images,' and that only after switching to binarized inputs did the outputs match the authors' expectation that the convex periphery should be red and the concave region predominantly blue. Because the preprocessing choice was selected after observing the experimental outputs, the subsequent qualitative agreement is not a valid confirmation of the method. The force map in Fig. 2d remains an unvalidated model projection rather than a measurement.
  3. [Generalization to dunes never seen by the CNN model, Fig. 3] The numerical-domain generalization claim is supported only by qualitative visual comparison of time-averaged forces and variances. No quantitative error metrics, confidence intervals, or per-grain instantaneous errors are reported. The text describes the results with phrases such as 'well predicted' and 'reasonably estimate' without any numerical support. Given the abstract's 'high accuracy' claim, the absence of any quantitative assessment of prediction error is a major omission even within the simulation domain where ground truth exists.
  4. [Supplementary Materials, Numerical simulations] The training labels are exclusively force maps from the authors' own CFD-DEM simulations, and the paper's validation of those simulations refers to prior work (Refs. 21, 22) without showing that the force field itself was experimentally validated. Since the entire morphology-to-force mapping is learned from these simulated labels, any systematic error in the simulated forces would propagate into all predictions, including the experimental ones. The manuscript does not address this source of uncertainty.
minor comments (4)
  1. [Abstract] The abstract overclaims by asserting 'high accuracy' and a 'breakthrough' while the body of the paper explicitly states that accuracy can only be verified in the numerical domain. These claims should be tempered to match the evidence.
  2. [Main text, 'Estimating the resultant force' section] There is a grammatical slip: 'We also want aim to develop the capability' should read 'We also aim to develop the capability.'
  3. [Supplementary Materials, CNN model] The notation in the description of the mapping function, 'Φ : R → R3, g(x) ↦ f(x) | x ∈ Rn', is confusing; the domain and codomain are not written consistently and should be clarified.
  4. [Fig. 3] The figure compares ground truth and model outputs only for time-averaged forces and variances; adding a panel with a quantitative error metric (e.g., mean absolute error or structural similarity) would make the comparison interpretable.

Circularity Check

1 steps flagged · score 5.0 of 10

Experimental force maps are validated only against the simulation-derived prior: the paper concedes no access to experimental Y and changed the input preprocessing after seeing unexpected experimental outputs.

  1. other [Supplementary Materials, section 'CNN model', paragraph beginning 'We observed, however...' (PDF p. 29); main text 'Estimating the resultant force...' discusses the binarization preprocessing.]
    "Despite not knowing what the true forces acting on the experimental really are, we expect them to be similar to those from the simulations, in the sense that the convex peripheral region of the dune is red, due to strong shearing forces, and the concave part is predominantly blue due to the fluid recirculation zone. But, that is not what we saw in the output of the neural network (now shown for brevity)."

    The only experimental check is the expectation imported from the same CFD-DEM simulations that produced the training labels ('we expect them to be similar to those from the simulations'). After the original model violated that expectation, the input representation was changed (binarization) and the model effectively retrained, so the final experimental output was selected to reproduce the simulation-based prior. Since the paper explicitly concedes 'we have no access to Y in the experimental domain,' there is no independent experimental measurement certifying Fig. 2d. The qualitative agreement is thus not an out-of-sample validation but a consequence of choosing the pipeline until it matched the training-domain expectation.

full rationale

The simulation-domain part of the paper is largely self-contained: the CNN is trained on CFD-DEM images with force labels and tested on unseen simulations (Fig. 3), which is a genuine supervised generalization check. The circularity concerns the experimental claim. The authors openly state that no force measurements exist for experimental dunes, so the only check of the transfer to real images is qualitative similarity to simulation output. That check is contaminated by the fact that, after seeing the experimental outputs did not look like simulation outputs, the authors changed the preprocessing (binarization) and retrained. This is a form of using the target-domain outputs to select the model, so the final agreement in Fig. 2d is not independent evidence that forces on the real dune are being measured. The numerical-domain accuracy reported in Fig. 3 remains meaningful, but it cannot certify sim-to-real transfer. Therefore the experimental 'measurement' is partially circular rather than fully invalid, and the abstract's 'quantify these forces with high accuracy' overstates the evidence.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The central claim rests on the fidelity of the authors' own CFD-DEM simulations and on the unvalidated sim-to-experiment transfer. The CNN weights and preprocessing choices are fitted to simulation data, and no independent experimental force measurements are used. No new physical entities are introduced.

free parameters (3)
  • CNN weights = trained on simulation force maps
    Millions of U-Net parameters are fit to CFD-DEM-derived images and force labels; they encode the entire morphology-to-force mapping and are the core fitted quantities.
  • Force color scale limits = [-2, 2] x 10^-7 N
    The authors fix the force scale range to this constant 'to minimize the effects of interval shift'; this is a hand-chosen normalization that affects the regression targets and visual comparisons.
  • Binarization threshold = not stated in main text
    Input images are thresholded to binary grain masks to reduce covariate shift; the threshold is a hand-set preprocessing parameter described in the supplementary.
assumptions (3)
  • domain assumption CFD-DEM with LES/WALE provides accurate ground-truth grain-scale forces for the experiments.
    All training labels come from these simulations (Eqs. 1-8); the paper cites prior validation in Refs. 21-22 but provides no direct experimental force measurement to confirm force magnitudes.
  • domain assumption After binarization, the morphology-to-force mapping learned on simulations transfers to experimental images.
    The paper states there is no experimental Y and relies on the expectation that experimental forces are 'similar to those from the simulations'; this sim-to-real transfer is the core unvalidated premise.
  • domain assumption The U-Net architecture can learn a generalizing function from dune morphology to per-pixel force with the provided datasets.
    The authors explicitly decline to provide generalization bounds and assume that training on varied Reynolds numbers and particle counts is sufficient for generalization to unseen dunes.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Resultant force on grains of a real sand dune: How to measure it?." pith.science (2026). https://pith.science/paper/6QZQ5FO7

@misc{pith2026250105869,
  author       = {Pith},
  title        = {Pith review of: Resultant force on grains of a real sand dune: How to measure it?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/6QZQ5FO7}},
  note         = {Machine review of arXiv:2501.05869}
}
read the original abstract

Dunes are bedforms found on sandy terrains shaped by fluid flow on Earth, Mars, and other celestial bodies. Despite their prevalence, understanding dune dynamics at the grain scale is challenging due to the vast number of grains involved. In this study, we demonstrate a novel approach to estimate the forces acting on individual dune grains using images. By combining subaqueous experiments, high-speed camera recordings, discrete numerical simulations, and a specially trained convolutional neural network, we can quantify these forces with high accuracy. This method represents a breakthrough in studying granular dynamics, offering a new way to measure forces not only on dune grains but also on smaller objects, such as rocks, boulders, rovers, and man-made structures, observed in satellite images of both Earth and Mars. This technique expands our ability to analyze and understand fluid-grain interactions in diverse environments.

Figures

Figures reproduced from arXiv: 2501.05869 by the authors.

Figure 1
Figure 1. (a) HiRISE image (35) showing a field of barchans undergoing complex interactions on the surface of Mars: 23.190◦ latitude (centered), 339.585◦ longitude (East), spacecraft al￾titude 287.3 km. Courtesy NASA/JPL-Caltech/UArizona; (b) Field of barchans on the Nazca desert: -15.278◦ latitude, -74.878◦ longitude, June 2012. Courtesy Google Earth Pro; (c) Ex￾perimental image of barchan obtained in water channel. Estimati… view at source ↗
Figure 2
Figure 2. (a) Top view of subaqueous dune obtained from numerical simulation showing the [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 4
Figure 4. Sketch of the experimental setup. The experiment begins with a head tank that serves [PITH_FULL_IMAGE:figures/full_fig_p024_4.png] view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: Photograph of the experimental setup. On the bottom, the photograph shows the test [PITH_FULL_IMAGE:figures/full_fig_p025_5.png]
Figure 6
Figure 6. Figure 6: Snapshots showing top views of bedforms for the typical initial condition and devel [PITH_FULL_IMAGE:figures/full_fig_p026_6.png]
Figure 7
Figure 7. Figure 7: Diagram of the U-Net architecture coupled with a domain adversarial discriminative [PITH_FULL_IMAGE:figures/full_fig_p027_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

72 extracted references · 68 canonical work pages

  1. [1]

    R. A. Bagnold, The Physics of Blown Sand and Desert Dunes\/ (Chapman and Hall, London, 1941)

    work page 1941

  2. [2]

    H. J. Herrmann, G. Sauermann, Physica A (Amsterdam)\/ 283 , 24 (2000)

    work page 2000

  3. [3]

    Hersen, Eur

    P. Hersen, Eur. Phys. J. B\/ 37 , 507 (2004)

    work page 2004

  4. [4]

    Elbelrhiti, P

    H. Elbelrhiti, P. Claudin, B. Andreotti, Nature\/ 437 (2005)

    work page 2005

  5. [5]

    Claudin, B

    P. Claudin, B. Andreotti, Earth Plan. Sci. Lett.\/ 252 , 20 (2006)

    work page 2006

  6. [6]

    E. J. R. Parteli, H. J. Herrmann, Phys. Rev. E\/ 76 , 041307 (2007)

    work page 2007

  7. [7]

    Courrech du Pont , C

    S. Courrech du Pont , C. R. Phys.\/ 16 , 118 (2015)

    work page 2015

  8. [8]

    E. M. Franklin, Appl. Math. Model.\/ 39 , 7390 (2015)

    work page 2015

Show all 72 references
  1. [9]

    Hersen, S

    P. Hersen, S. Douady, B. Andreotti, Phys. Rev. Lett.\/ 89 , 264301 (2002)

    2002

  2. [10]

    Rubanenko, S

    L. Rubanenko, S. Pérez-López, J. Schull, M. G. A. Lapôtre, IEEE J. Sel. Top. Appl.\/ 14 , 9364 (2021)

    2021

  3. [11]

    Rubanenko, M

    L. Rubanenko, M. G. A. Lapôtre, R. C. Ewing, L. K. Fenton, A. Gunn, Nat. Commun.\/ 13 (2022)

    2022

  4. [12]

    A. C. W. Baas, L. A. Delobel, Nat. Clim. Chang.\/ 12 , 999 (2022)

    2022

  5. [13]

    N. Endo, H. Kubo, T. Sunamura, Earth Surf. Process. Landforms\/ 29 , 31 (2004)

    2004

  6. [14]

    N. Hori, A. Yamada, Y. Oshiro, O. Sano, J. Phys. Soc. Jpn.\/ 76 , 024401 (2007)

    2007

  7. [15]

    C. A. Alvarez, E. M. Franklin, Phys. Rev. Lett.\/ 121 , 164503 (2018)

    2018

  8. [16]

    C. A. Alvarez, E. M. Franklin, Phys. Rev. E\/ 100 , 042904 (2019)

    2019

  9. [17]

    Bin Yang , et al.\/ , Eur. J. Mech. B Fluids\/ 103 , 246 (2024)

    2024

  10. [18]

    W. R. Assis, E. M. Franklin, Geophys. Res. Lett.\/ 47 , e2020GL089464 (2020)

    2020

  11. [19]

    W. R. Assis, E. M. Franklin, J. Geophys. Res.: Earth Surf.\/ 126 , e2021JF006237 (2021)

    2021

  12. [20]

    N. He, Y. Lin, Y. Zhang, B. Yang, X. Gao, Phys. Fluids\/ 35 , 106609 (2023)

    2023

  13. [21]

    C. A. Alvarez, E. M. Franklin, Phys. Rev. E\/ 101 , 012905 (2020)

    2020

  14. [22]

    C. A. Alvarez, E. M. Franklin, Phys. Fluids\/ 33 , 013313 (2021)

    2021

  15. [23]

    N. C. Lima, W. R. Assis, C. A. Alvarez, E. M. Franklin, Phys. Fluids\/ 34 , 123320 (2022)

    2022

  16. [24]

    Silvestro, D

    S. Silvestro, D. A. Vaz, L. K. Fenton, P. E. Geissler, Geophys. Res. Lett.\/ 38 (2011)

    2011

  17. [25]

    L. K. Fenton, Icarus\/ 352 , 114018 (2020)

    2020

  18. [26]

    Tsoar, R

    H. Tsoar, R. Greeley, A. R. Peterfreund, J. Geophys. Res.\/ 84 , 8167 (1979)

    1979

  19. [27]

    Tsoar, E

    H. Tsoar, E. J. R. Parteli, Environ. Earth Sci.\/ 75 (2016)

    2016

  20. [28]

    Zhang, Z

    Z. Zhang, Z. Dong, G. Hu, E. J. R. Parteli, Geosciences\/ 8 (2018)

    2018

  21. [29]

    M. A. Azzaoui, M. Adnani, H. El Belrhiti, I. E. Chaouki, C. Masmoudi, The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences\/ XLI-B7 , 153 (2016)

    2016

  22. [30]

    Carrera, L

    D. Carrera, L. Bandeira, R. Santana, J. A. Lozano, Knowledge-Based Systems\/ 163 , 858 (2019)

    2019

  23. [31]

    K. He, G. Gkioxari, P. Dollár, R. Girshick, 2017 IEEE International Conference on Computer Vision (ICCV)\/ (2017), pp. 2980--2988

    2017

  24. [32]

    E. A. C\'u\ nez, E. M. Franklin, Sci. Rep.\/ 14 , 18381 (2024)

    2024

  25. [33]

    Li, Photogramm

    Z. Li, Photogramm. Rec.\/ 14 , 113 (1992)

    1992

  26. [34]

    C. J. Okolie, J. L. Smit, ISPRS J. Photogramm. Remote Sens\/ 188 , 1 (2022)

    2022

  27. [35]

    High resolution imaging science experiment, https://www.uahirise.org/, HiRISE Operations Center - University of Arizona

  28. [36]

    C. A. Alvarez, E. M. Franklin, Mendeley Data, http://dx.doi.org/10.17632/g666hxgrty.1\/ (2020)

    2020 doi

  29. [37]

    N. C. Lima, W. R. Assis, C. A. Alvarez, E. M. Franklin, Mendeley Data, https://data.mendeley.com/datasets/zhnngnvvz8\/ (2022)

    2022

  30. [38]

    Ronneberger, P

    O. Ronneberger, P. Fischer, T. Brox, Medical Image Computing and Computer-Assisted Intervention -- MICCAI 2015\/ (Springer International Publishing, Cham, 2015), pp. 234--241

    2015

  31. [39]

    I. J. Goodfellow, Y. Bengio, A. Courville, Deep Learning\/ (MIT Press, Cambridge, MA, USA, 2016)

    2016

  32. [40]

    Bengio, A

    Y. Bengio, A. Courville, P. Vincent, IEEE Trans. Pattern Anal. Mach. Intell.\/ 35 , 1798–1828 (2013)

    2013

  33. [41]

    Ganin, et al.\/ , Journal of Machine Learning Research\/ 17 , 1 (2016)

    Y. Ganin, et al.\/ , Journal of Machine Learning Research\/ 17 , 1 (2016)

    2016

  34. [42]

    Farahani, S

    A. Farahani, S. Voghoei, K. Rasheed, H. R. Arabnia, Advances in Data Science and Information Engineering\/ , R. Stahlbock, et al.\/ , eds. (Springer International Publishing, Cham, 2021), pp. 877--894

    2021

  35. [43]

    Achille, M

    A. Achille, M. Rovere, S. Soatto, International Conference on Learning Representations\/ (2019)

    2019

  36. [44]

    S. J. Pan, Q. Yang, IEEE Transactions on Knowledge and Data Engineering\/ 22 , 1345 (2010)

    2010

  37. [45]

    H. Zhao, R. T. D. Combes, K. Zhang, G. Gordon, Proceedings of the 36th International Conference on Machine Learning\/ , K. Chaudhuri, R. Salakhutdinov, eds. (PMLR, 2019), vol. 97 of Proceedings of Machine Learning Research\/ , pp. 7523--7532

    2019

  38. [46]

    WOODTV8, Silver L ake D unes swallow up house, https://www.youtube.com/watch?v=ifxzs MA4IWY (2017)

    2017

  39. [47]

    R. F. Miotto, C. A. Alvarez, D. S. Borges, W. R. Wolf, E. M. Franklin, Dataset on Zenodo, https://doi.org/10.5281/zenodo.14262069\/ (2025)

    2025 doi

  40. [48]

    N. C. Lima, W. R. Assis, C. A. Alvarez, E. M. Franklin, Mendeley Data, http://dx.doi.org/10.17632/zhnngnvvz8.1\/ (2022)

    2022 doi

  41. [49]

    N. C. Lima, W. R. Assis, E. M. Franklin, Mendeley Data, http://dx.doi.org/10.17632/ypkgwjfr4r.1\/ (2024)

    2024 doi

  42. [50]

    Goniva, C

    C. Goniva, C. Kloss, N. G. Deen, J. A. M. Kuipers, S. Pirker, Particuology\/ 10 , 582 (2012)

    2012

  43. [51]

    Kloss, C

    C. Kloss, C. Goniva, Proc. 5th Int. Conf. on Discrete Element Methods\/ (London, UK, 2010)

    2010

  44. [52]

    Berger, C

    R. Berger, C. Kloss, A. Kohlmeyer, S. Pirker, Powder Technology\/ 278 , 234 (2015)

    2015

  45. [53]

    Z. Y. Zhou, S. B. Kuang, K. W. Chu, A. B. Yu, J. Fluid Mech.\/ 661 , 482–510 (2010)

    2010

  46. [54]

    Schlichting, K

    H. Schlichting, K. Gersten, Berlin, Heidelberg\/ (2000)

    2000

  47. [55]

    Glorot, A

    X. Glorot, A. Bordes, Y. Bengio, Proceedings of the 28th International Conference on International Conference on Machine Learning\/ , ICML'11 (Omnipress, Madison, WI, USA, 2011), p. 513–520

    2011

  48. [56]

    Oquab, L

    M. Oquab, L. Bottou, I. Laptev, J. Sivic, 2014 IEEE Conference on Computer Vision and Pattern Recognition\/ (2014), pp. 1717--1724

    2014

  49. [57]

    M. Long, Z. Cao, J. Wang, M. I. Jordan, Proceedings of the 32nd International Conference on Neural Information Processing Systems\/ , NIPS'18 (Curran Associates Inc., Red Hook, NY, USA, 2018), p. 1647–1657

    2018

  50. [58]

    Tzeng, J

    E. Tzeng, J. Hoffman, N. Zhang, K. Saenko, T. Darrell, CoRR\/ abs/1412.3474 (2014)

    2014 arXiv

  51. [59]

    M. Long, Y. Cao, J. Wang, M. Jordan, Proceedings of the 32nd International Conference on Machine Learning\/ , F. Bach, D. Blei, eds. (PMLR, Lille, France, 2015), vol. 37 of Proceedings of Machine Learning Research\/ , pp. 97--105

    2015

  52. [60]

    Zhao, et al.\/ , IEEE Transactions on Neural Networks and Learning Systems\/ 33 , 473 (2022)

    S. Zhao, et al.\/ , IEEE Transactions on Neural Networks and Learning Systems\/ 33 , 473 (2022)

    2022

  53. [61]

    Stojanov, et al.\/ , Advances in Neural Information Processing Systems\/ , A

    P. Stojanov, et al.\/ , Advances in Neural Information Processing Systems\/ , A. Beygelzimer, Y. Dauphin, P. Liang, J. W. Vaughan, eds. (2021)

    2021

  54. [62]

    F. D. Johansson, D. A. Sontag, R. Ranganath, International Conference on Artificial Intelligence and Statistics\/ (2019)

    2019

  55. [63]

    Li, et al.\/ , Computer Vision -- ECCV 2018\/ , V

    Y. Li, et al.\/ , Computer Vision -- ECCV 2018\/ , V. Ferrari, M. Hebert, C. Sminchisescu, Y. Weiss, eds. (Springer International Publishing, Cham, 2018), pp. 647--663

    2018

  56. [64]

    A. T. Nguyen, T. Tran, Y. Gal, A. G. Baydin, Advances in Neural Information Processing Systems\/ , M. Ranzato, A. Beygelzimer, Y. Dauphin, P. Liang, J. W. Vaughan, eds. (Curran Associates, Inc., 2021), vol. 34, pp. 5264--5275

    2021

  57. [65]

    J. Ren, M. Zhang, C. Yu, Z. Liu, Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)\/ (2022), pp. 7926--7935

    2022

  58. [66]

    H. Wang, H. He, D. Katabi, Proceedings of the 37th International Conference on Machine Learning\/ , H. D. III, A. Singh, eds. (PMLR, 2020), vol. 119 of Proceedings of Machine Learning Research\/ , pp. 9898--9907

    2020

  59. [67]

    Chidlovskii, A

    B. Chidlovskii, A. Sadek, C. Wolf, Universal domain adaptation in ordinal regression (2021)

    2021

  60. [68]

    Y. Yang, K. Zha, Y.-C. Chen, H. Wang, D. Katabi, ArXiv\/ abs/2102.09554 (2021)

    2021 arXiv

  61. [69]

    X. Chen, S. Wang, J. Wang, M. Long, Proceedings of the 38th International Conference on Machine Learning\/ , M. Meila, T. Zhang, eds. (PMLR, 2021), vol. 139 of Proceedings of Machine Learning Research\/ , pp. 1749--1759

    2021

  62. [70]

    Ganin, V

    Y. Ganin, V. Lempitsky, Proceedings of the 32nd International Conference on International Conference on Machine Learning - Volume 37\/ , ICML'15 (JMLR.org, 2015), p. 1180–1189

    2015

  63. [71]

    , " * write output.state after.block = add.period write newline

    ENTRY address author booktitle chapter edition editor howpublished institution journal key month note number organization pages publisher school series title type volume year label INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.con...

  64. [72]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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