Pith. sign in

REVIEW 3 major objections 4 minor 29 references

LFA applied to CNNs: Efficient Singular Value Decomposition of Convolutional Mappings by Local Fourier Analysis

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read A CNN convolution's entire singular-value spectrum can be computed by SVD-ing one small channel block per Fourier frequency, skipping the FFT log factor.

desk verdict Correct and useful O(N) algorithm for periodic boundary conditions; the zero-padding transfer is empirical only, so the 'exact SVD' claim overreaches. read the letter →

arxiv 2506.05617 v1 pith:Z5TJSYA4 submitted 2025-06-05 cs.LG cs.AI

Antonia van Betteray , Matthias Rottmann , Karsten Kahl This is my paper
classification cs.LGcs.AI MSC 65F1565T5068T07
keywords singularvaluedecompositionconvolutionalneuralnetworkslocalFourieranalysisconvolutiontheoremspectralnormsymbolboundaryconditionscomputationalcomplexity
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

The paper shows that the complete singular value decomposition of a convolutional layer can be obtained by evaluating the layer's Fourier symbol at every discrete frequency and taking the SVD of the resulting $c_{out} \times c_{in}$ block. This reduces the cost from the FFT-based $O(n^2 c^2(c+\log n))$ to $O(n^2 c^3)$ for square inputs, with no log factor, and the same idea extends to rectangular inputs and unequal channel counts. The authors argue that this makes full spectral information accessible for convolutional layers whose explicit matrix form is far too large to store. They support the claim with a runtime analysis and numerical experiments, while acknowledging that the exactness relies on periodic boundary conditions rather than the zero padding used in most CNNs.

What carries the argument

The load-bearing object is the Fourier symbol $A_k=\sum_{y\in\mathcal{N}} M_y e^{2\pi i\langle k,y\rangle}$, a $c_{out}\times c_{in}$ matrix built from the $c_{out}\times c_{in}$ kernel weights $M_y$ and the frequency $k$. Because the spatial Fourier modes are orthogonal and are eigenvectors of any shift-invariant convolution, the full operator splits into one such block per frequency; the whole spectrum is the union of the per-block spectra, and the global singular vectors are the per-block singular vectors transformed by the Fourier basis. This block diagonalization is what carries the complexity reduction: each block SVD costs $O(c^3)$ and there are $nm$ blocks, independent of any FFT log factor.

What would settle it

For a small periodic convolution, form the explicit doubly block-circulant matrix and compare its full singular-value list to Algorithm 1's output for many random kernels; any mismatch beyond numerical precision would disprove the exactness claim. Separately, for a zero-padded convolution at $n=8$ or $n=16$, compare the LFA spectrum to the explicit sparse-matrix SVD: if the maximum relative deviation does not shrink as $n$ grows, the practical approximation claim for Dirichlet boundary conditions fails.

Watch

Extended reading notes

Core claim

The central claim is that a convolutional mapping $A: \mathbb{R}^{m\times n\times c_{in}}\to\mathbb{R}^{m\times n\times c_{out}}$ is diagonalized by the Fourier basis in the spatial dimensions, so its singular values are exactly the union of the singular values of the symbols $A_k=\sum_{y\in\mathcal{N}} M_y e^{2\pi i\langle k,y\rangle}$ taken over all frequencies $k$ on the dual torus. Each symbol is only $c_{out}\times c_{in}$, and its left and right singular vectors lift back to global singular vectors through the Fourier basis matrices $F^{c_{out}}_k$ and $F^{c_{in}}_k$. Algorithm 1 performs one SVD per frequency without ever forming the global matrix, and the paper proves the per-frequency computations have cost $O(1)$ in the spatial size, giving $O(nmc^3)$ overall for equal channels. This removes the $\log n$ factor that the FFT-based approach incurs.

Load-bearing premise

The load-bearing premise is that periodic boundary conditions give the same singular-value spectrum as the zero padding used in real CNN layers; the paper only demonstrates this agreement empirically and notes that boundary conditions clearly affect accuracy for small inputs.

Editorial extensions

If this is right

  • For a fixed channel count, computing all singular values of a convolutional layer becomes linear in the number of pixels, so spectra of layers with very large inputs become tractable.
  • The full spectrum, not just the largest singular value, can be used for spectral regularization, robustness certification, low-rank compression, and exact pseudo-inverse layers.
  • The speed-up over the FFT-based method grows with input size; the paper reports a measured factor of 1.09 at $n=256$ and 1.44 at $n=16{,}384$.
  • Because each frequency's SVD is independent, the computation is embarrassingly parallel and can be distributed across cores or devices.
  • The method also produces a memory layout that makes the subsequent SVD routine faster, further widening the runtime gap.

Reading between the lines

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

  • The exactness claim is cleanest for circular convolutions; applying the algorithm to zero-padded CNN layers is an approximation whose error the paper only observes empirically. A natural next step is deriving a quantitative bound on the spectrum perturbation in terms of kernel support and input size.
  • The lattice and crystal formulation suggests the same symbol-based SVD works for non-rectangular sampling patterns, so the method could extend to spherical or octagonal convolutions with only a change of the frequency grid.
  • One testable extension is to use the per-frequency singular vectors to build exact low-rank approximations of convolutional layers in the Fourier domain, which may compress models more effectively than spatial low-rank factorizations.
  • If periodic padding were adopted in network architectures, the SVD would be exact and differentiable, enabling spectral losses computed from the full spectrum rather than estimates of the largest singular value.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 4 minor

Summary. The paper proposes a local Fourier analysis (LFA) based method for computing the singular value decomposition of convolutional mappings. Exploiting translation invariance, the method evaluates the symbol A_k = sum_y M_y exp(2π i <k,y>) for each frequency k and performs an SVD of each c_out × c_in block, yielding the complete SVD under periodic boundary conditions. The authors derive an O(N c^3) complexity for the singular value computation, improving on the FFT-based approach of Sedghi et al. by a logarithmic factor, and present runtime experiments on a CPU for input sizes up to n = 16,384. The paper also empirically compares singular value spectra under periodic and zero-padded (Dirichlet) boundary conditions, finding qualitative agreement for large input sizes.

Significance. If the claims are appropriately qualified, the periodic-boundary result is a clean and useful contribution: the per-frequency block-diagonalization of a convolutional operator is derived self-containedly, and the O(N) scaling (for fixed kernel and channel sizes) over the FFT baseline is theoretically and experimentally supported. The derivation is free of fitted parameters, and the runtime study covers a wide range of input sizes. The main caveat is that standard CNN convolutions use zero padding, not periodic boundary conditions, and the paper's evidence for transferring the exact periodic-boundary result to the zero-padded case is only qualitative. Since the named applications (low-rank compression, pseudo-inverses) require singular vectors, not just the sorted spectrum, this gap is load-bearing for the paper's headline claim about real CNNs.

major comments (3)
  1. [Section IV-A and Section III(e)] The transfer from periodic to zero-padded (Dirichlet) boundary conditions is not quantitatively established. The paper only plots sorted singular value spectra for n = 4, 8, 32 and states that small inputs are 'clearly affected'; there is no error bound, no asymptotic rate, and no input-size threshold. Moreover, the comparison is only between sorted spectra, while the applications advertised in the introduction (low-rank compression, pseudo-inverse computation) require the actual singular vectors. The abstract and conclusion should either state clearly that the exact SVD claim holds only for periodic boundary conditions, or provide a quantitative analysis of the approximation error for zero-padded convolutions, including a check of the singular vectors.
  2. [Table I, row 'cin ≥ cout'] The stated complexity for unequal channel dimensions is incorrect. For a cout × cin matrix, the SVD cost is O(cout^2 cin) when cin ≥ cout and O(cout cin^2) when cin ≤ cout (with the usual convention for rectangular SVD). The table gives O(nm cin^2 cout) for cin ≥ cout, which swaps the roles of cin and cout. This should be corrected to O(nm cout^2 cin).
  3. [Section III(d), Algorithm 1, line 4] The claimed O(N) complexity omits the dependence on the kernel support size. Each symbol B_{i,j} is a sum over y ∈ N of the cout × cin matrices M_y times an exponential factor, so the per-frequency cost is O(|N| cin cout) before the SVD. The O(1) designation in line 4 is valid only when the kernel extent is treated as a fixed constant, as in typical small CNN kernels. This assumption should be stated explicitly in the complexity theorem and in the abstract, since without it the comparison to the FFT-based method is not complete.
minor comments (4)
  1. [Section V, Conclusion] The sentence 'We verified that the time required to compute the SVD decreases as n increases' is contradicted by Table II, where total runtime increases monotonically with n (from 2.30 s at n=256 to 7521.93 s at n=16384). The authors likely mean that the runtime gap or the per-singular-value cost decreases; please rephrase.
  2. [Section III(c)] There is a typographical artifact at the end of the sentence 'without forming the global singular vectors ˆUk and ˆVk,.'; the stray comma should be removed.
  3. [Section IV-b, paragraph 2] The example 'the computation of the singular values of a 1,024 × 1,024 matrix takes 0.30 seconds' would be clearer if it explicitly says that this is the explicit unrolled matrix size, not the input feature-map size, since the latter is used elsewhere in the same paragraph.
  4. [Section IV-b, Table II] The speed-up factor column would benefit from an explicit statement that sFFT and sLFA include the transformation step, to align with the text in Section IV-b that timing starts before the transformation.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the SVD derivation is a direct application of the convolution theorem, and no fitted parameter is renamed as a prediction.

full rationale

The central derivation in Section III(c) is self-contained: for periodic boundary conditions, the Fourier modes are invariant subspaces of the convolutional operator, and the symbol A_k = sum_{y in N} M_y e^{2πi⟨k,y⟩} is explicitly computed and then factored by a c_out × c_in SVD. The paper states this reduction directly ('Applying a convolution A to these wave functions immediately yields...') and Algorithm 1 implements it. No input quantity is fitted to the target singular values; the singular values are obtained from the symbol via standard SVD, and the claimed complexity O(n²c³) follows from evaluating each symbol in O(1) for a fixed local kernel and performing n² small SVDs. The cited self-work [17] supplies lattice/crystal notation and background, but the load-bearing eigenvalue argument is presented and proven in the manuscript itself, so the self-citation is not load-bearing. The acknowledged limitation regarding periodic versus zero-padded boundary conditions is an approximation caveat, not a circular step: the paper explicitly compares both spectra and does not claim the periodic result is derived from the zero-padded spectrum. Overall, the derivation does not reduce to its inputs, and there is no fitted-input-as-prediction or self-citation-forced conclusion.

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

The paper introduces no new free parameters or invented entities. It relies on standard convolution theorem results and a domain assumption about periodic boundary conditions, which is the main caveat of the method.

assumptions (4)
  • standard math Convolution theorem: Fourier modes diagonalize circulant (periodic) convolutions
    Invoked in Section III-c to transform the convolution into pointwise multiplication per frequency.
  • standard math Orthonormality of the Fourier basis on the crystal torus T_n,m
    Required so that singular vectors in frequency domain map back to spatial domain via orthonormal matrices F_k; Section III-c.
  • domain assumption The convolution under study uses periodic boundary conditions
    The LFA framework implies periodic boundary conditions (Section III-e); real CNNs use zero padding, and the gap is only studied empirically.
  • standard math The singular values of the full operator are the union of the per-frequency block singular values
    Follows from the block-diagonal structure of the transformed operator; Section III-c.

how reviews work

0 comments
Cite this review

Pith. "Pith review of LFA applied to CNNs: Efficient Singular Value Decomposition of Convolutional Mappings by Local Fourier Analysis." pith.science (2026). https://pith.science/paper/Z5TJSYA4

@misc{pith2026250605617,
  author       = {Pith},
  title        = {Pith review of: LFA applied to CNNs: Efficient Singular Value Decomposition of Convolutional Mappings by Local Fourier Analysis},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z5TJSYA4}},
  note         = {Machine review of arXiv:2506.05617}
}
read the original abstract

The singular values of convolutional mappings encode interesting spectral properties, which can be used, e.g., to improve generalization and robustness of convolutional neural networks as well as to facilitate model compression. However, the computation of singular values is typically very resource-intensive. The naive approach involves unrolling the convolutional mapping along the input and channel dimensions into a large and sparse two-dimensional matrix, making the exact calculation of all singular values infeasible due to hardware limitations. In particular, this is true for matrices that represent convolutional mappings with large inputs and a high number of channels. Existing efficient methods leverage the Fast Fourier transformation (FFT) to transform convolutional mappings into the frequency domain, enabling the computation of singular values for matrices representing convolutions with larger input and channel dimensions. For a constant number of channels in a given convolution, an FFT can compute N singular values in O(N log N) complexity. In this work, we propose an approach of complexity O(N) based on local Fourier analysis, which additionally exploits the shift invariance of convolutional operators. We provide a theoretical analysis of our algorithm's runtime and validate its efficiency through numerical experiments. Our results demonstrate that our proposed method is scalable and offers a practical solution to calculate the entire set of singular values - along with the corresponding singular vectors if needed - for high-dimensional convolutional mappings.

Figures

Figures reproduced from arXiv: 2506.05617 by the authors.

Figure 2
Figure 2. Figure 2a Lattice L(A) with basis A = [a1 a2].fig. 2b A crystal L u (A) contains copies of A, each shifted by u1,...,uν. (a) Sublattice with primitive vectors c1 and c2. (b) Crystal L u (A) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Sublattice L(C) ⊆ L(A) with C = A ·  3 2 1 3 and the periodicity of the corresponding crystal torus TA,C . The SVD can be used for low rank approximations to enable model compression. Extensive studies on low rank approx￾imation in neural networks have been conducted by [10]– [12], [26]. These methods have been further developed to compress word embedding layers [13]–[15]. The application of the SVD in CNN interpr… view at source ↗
Figure 4
Figure 4. 3×3 kernel operator N , centered at x (green dot). The extension to the crystal case is straight-forward and a canonical basis of wave functions can be fixed. Applying a convolution A to these wave functions immediately yields (A∗ fk)(x) = ∑ y∈N My · e 2πihk,yi  e 2πihk,xi . That is, the wave functions a.k.a. Fourier modes yield invariant subspaces w.r.t. the convolution A. Note, that due to the fact that the convo… view at source ↗
Figures from the paper (2 more)
Figure 5
Figure 5. Figure 5: Dirichlet boundary conditions (left), the white dot [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Effect of boundary conditions for increasing input s [PITH_FULL_IMAGE:figures/full_fig_p006_6.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

29 extracted references · 28 canonical work pages

  1. [1]

    ImageNet Large Scale Visual Recognition Chall enge,

    O. Russakovsky, J. Deng, H. Su, J. Krause, S. Satheesh, S. Ma, Z. Huang, A. Karpathy, A. Khosla, M. Bernstein, A. C. Berg, an d L. Fei-Fei, “ImageNet Large Scale Visual Recognition Chall enge,” International Journal of Computer Vision , vol. 115, pp. 211–252, Dec. 2015

    work page 2015

  2. [2]

    ImageNet Classification with Deep Convolutional Neural Networks,

    A. Krizhevsky, I. Sutskever, and G. E. Hinton, “ImageNet Classification with Deep Convolutional Neural Networks,” in Advances in Neural Information Processing Systems , vol. 25, Curran Associates, Inc., 2012

    work page 2012

  3. [3]

    Deep Residual Learnin g for Image Recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep Residual Learnin g for Image Recognition,” in 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR) , pp. 770–778, June 2016. ISSN: 1063-6919

    work page 2016

  4. [4]

    Gradient-based learning applied to document recognition,

    Y . Lecun, L. Bottou, Y . Bengio, and P . Haffner, “Gradient-based learning applied to document recognition,” Proceedings of the IEEE , vol. 86, pp. 2278–2324, Nov. 1998

    work page 1998

  5. [5]

    Spectral norm regularization for improving the generalizability of deep learning,

    Y . Y oshida and T. Miyato, “Spectral norm regularization for improving the generalizability of deep learning,” 05 2017

    work page 2017

  6. [6]

    The singular values o f convolu- tional layers,

    H. Sedghi, V . Gupta, and P . M. Long, “The singular values o f convolu- tional layers,” in International Conference on Learning Representations , 2019

    work page 2019

  7. [7]

    Regular isation of neural networks by enforcing lipschitz continuity,

    H. Gouk, E. Frank, B. Pfahringer, and M. J. Cree, “Regular isation of neural networks by enforcing lipschitz continuity,” Mach. Learn. , vol. 110, p. 393–416, Feb. 2021

    work page 2021

  8. [8]

    Parseval networks: improving robustness to adversarial e xamples,

    M. Cisse, P . Bojanowski, E. Grave, Y . Dauphin, and N. Usun ier, “Parseval networks: improving robustness to adversarial e xamples,” in Proceedings of the 34th International Conference on Machin e Learning - V olume 70, ICML’17, p. 854–863, JMLR.org, 2017

    work page 2017

Show all 29 references
  1. [9]

    Invisible back door attack through singular value decomposition,

    W. Chen, X. Xu, X. Wang, Z. Li, and Y . Chen, “Invisible back door attack through singular value decomposition,” in Pattern Recognition and Computer Vision: 7th Chinese Conference, PRCV 2024, Uru mqi, China, October 18–20, 2024, Proceedings, Part II , (Berlin, Heidelberg), p. ...

    2024

  2. [10]

    Speeding u p convolutional neural networks with low rank expansions,

    M. Jaderberg, A. V edaldi, and A. Zisserman, “Speeding u p convolutional neural networks with low rank expansions,” BMVC 2014 - Proceedings of the British Machine Vision Conference 2014 , 05 2014

    2014

  3. [11]

    Accelerating very de ep convolu- tional networks for classification and detection,

    X. Zhang, J. Zou, K. He, and J. Sun, “Accelerating very de ep convolu- tional networks for classification and detection,” IEEE Transactions on Pattern Analysis and Machine Intelligence , vol. 38, no. 10, pp. 1943– 1955, 2016

    1943

  4. [12]

    Compressing pre-trained language models by matrix decomposition,

    M. Ben Noach and Y . Goldberg, “Compressing pre-trained language models by matrix decomposition,” in Proceedings of the 1st Confer- ence of the Asia-Pacific Chapter of the Association for Compu tational Linguistics and the 10th International Joint Conference on Natural Lan- gua...

    2020

  5. [13]

    GroupR educe: Block- wise low-rank approximation for neural language model shri nking,

    P . Chen, S. Si, Y . Li, C. Chelba, and C.-J. Hsieh, “GroupR educe: Block- wise low-rank approximation for neural language model shri nking,” in Advances in Neural Information Processing Systems (S. Bengio, H. Wal- lach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Gar n...

    2018

  6. [14]

    Onl ine embedding compression for text classification using low rank matrix fa ctorization,

    A. Acharya, R. Goel, A. Metallinou, and I. Dhillon, “Onl ine embedding compression for text classification using low rank matrix fa ctorization,” vol. 33, no. 1, pp. 6196–6203, 2019

    2019

  7. [15]

    L anguage model compression with weighted low-rank factorization,

    Y .-C. Hsu, T. Hua, S. Chang, Q. Lou, Y . Shen, and H. Jin, “L anguage model compression with weighted low-rank factorization,” 2022

    2022

  8. [16]

    The svd of convolutional weights: A cnn interpreta bility framework,

    B. Praggastis, D. Brown, C. O. Marrero, E. Purvine, M. Sh apiro, and B. Wang, “The svd of convolutional weights: A cnn interpreta bility framework,” ArXiv, vol. abs/2208.06894, 2022

    2022 arXiv

  9. [17]

    Automated local fourier anal ysis (aLFA),

    K. Kahl and N. Kintscher, “Automated local fourier anal ysis (aLFA),” vol. 60, no. 3, pp. 651–686, 2020

    2020

  10. [18]

    MgNet: A unified framework of multigrid a nd convolutional neural network,

    J. He and J. Xu, “MgNet: A unified framework of multigrid a nd convolutional neural network,” Science China Mathematics , vol. 62, pp. 1331–1354, July 2019

    2019

  11. [19]

    M gic: Multigrid- in-channels neural network architectures,

    M. Eliasof, J. Ephrath, L. Ruthotto, and E. Treister, “M gic: Multigrid- in-channels neural network architectures,” SIAM Journal on Scientific Computing, vol. 0, no. 0, pp. S307–S328, 2020

    2020

  12. [20]

    Mgiad: Multi grid in all dimensions. efficiency and robustness by weight sharing and coars- ening in resolution and channel dimensions,

    A. van Betteray, M. Rottmann, and K. Kahl, “Mgiad: Multi grid in all dimensions. efficiency and robustness by weight sharing and coars- ening in resolution and channel dimensions,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICC V) W ork- shops, ...

    2023

  13. [21]

    Poly-mgnet: Polynomial building blocks in multigrid-inspired resnets,

    A. van Betteray, M. Rottmann, and K. Kahl, “Poly-mgnet: Polynomial building blocks in multigrid-inspired resnets,” in Proceedings of the 14th International Conference on Pattern Recognition Appl ications and Methods , p. 181–191, SCITEPRESS - Science and Technology Publications, 2025

    2025

  14. [22]

    Lipschitz-margi n training: Scal- able certification of perturbation invariance for deep neur al networks,

    Y . Tsuzuku, I. Sato, and M. Sugiyama, “Lipschitz-margi n training: Scal- able certification of perturbation invariance for deep neur al networks,” in Neural Information Processing Systems , 2018

    2018

  15. [23]

    A PAC-ba yesian ap- proach to spectrally-normalized margin bounds for neural n etworks,

    B. Neyshabur, S. Bhojanapalli, and N. Srebro, “A PAC-ba yesian ap- proach to spectrally-normalized margin bounds for neural n etworks,” in International Conference on Learning Representations , 2018

    2018

  16. [24]

    Fantastic four: Differentiabl e and efficient bounds on singular values of convolution layers,

    S. Singla and S. Feizi, “Fantastic four: Differentiabl e and efficient bounds on singular values of convolution layers,” in International Conference on Learning Representations , 2021

    2021

  17. [25]

    Plug -and- play methods provably converge with properly trained denoi sers,

    E. Ryu, J. Liu, S. Wang, X. Chen, Z. Wang, and W. Yin, “Plug -and- play methods provably converge with properly trained denoi sers,” in Proceedings of the 36th International Conference on Machin e Learning (K. Chaudhuri and R. Salakhutdinov, eds.), vol. 97 of Proceedings of Ma...

    2019

  18. [26]

    Exploiting linear structure within convolutional networks for efficie nt evaluation,

    E. Denton, W. Zaremba, J. Bruna, Y . LeCun, and R. Fergus, “Exploiting linear structure within convolutional networks for efficie nt evaluation,” in Proceedings of the 27th International Conference on Neural Information Processing Systems - V olume 1 , NIPS’14, (Cambridge, MA, ...

    2014

  19. [27]

    Pseudoinvertible Neur al Networks ,

    E. D. Bolluyt and C. Comaniciu, “ Pseudoinvertible Neur al Networks ,” IEEE Transactions on Artificial Intelligence , vol. 5, pp. 602–612, Feb. 2024

    2024

  20. [28]

    L. N. Trefethen and D. Bau, Numerical linear algebra. SIAM textbooks, siam, Society for Industrial and Applied Mathematics, twen ty-fifth anniversary edition ed., 2022

    2022

  21. [29]

    Array programming with NumPy,

    C. R. Harris, K. J. Millman, S. J. van der Walt, R. Gommers , P . Virtanen, D. Cournapeau, E. Wieser, J. Taylor, S. Berg, N. J. Smith, R. K ern, M. Picus, S. Hoyer, M. H. van Kerkwijk, M. Brett, A. Haldane, J. Fernandez del Rio, M. Wiebe, P . Peterson, P . G´ erard-Marc hant, ...

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.