Pith. sign in

REVIEW 2 major objections 4 minor 33 references

Consistency of augmentation graph and network approximability in contrastive learning

T0 review · 2 major / 4 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This paper proves that as augmented datasets grow, the augmentation graph Laplacian converges pointwise and spectrally to a weighted Laplace-Beltrami operator on the natural data manifold, and that this limit yields a bounded ReLU network…

desk verdict Solid new consistency results for augmentation graph Laplacians, but the headline network approximability theorem has a missing projection step and needs major repair. read the letter →

arxiv 2502.04312 v2 pith:47TCKIEP submitted 2025-02-06 cs.LG math.APmath.SP

Chenghui Li , A. Martina Neuman This is my paper · ORCID
classification cs.LGmath.APmath.SP MSC 58J5068T0762H30
keywords contrastivelearningaugmentationgraphLaplacianconsistencyspectralconvergenceLaplace-Beltramioperatormanifoldneuralnetworkapproximabilityrealizabilityassumption
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 sets out to prove that the augmentation graph Laplacian, the object that spectral contrastive loss is built on, becomes a consistent picture of the true data manifold as the augmented dataset grows: pointwise and spectrally, it converges to a weighted Laplace-Beltrami operator on the manifold. The authors then use that limit to address a known gap in contrastive learning theory, the realizability assumption, which says that some neural network can achieve the global minimum of the spectral contrastive loss. Because the eigenfunctions of the limit operator are smooth and the graph eigenvectors provably track them, they conclude that a ReLU network of bounded depth and width reproduces the optimal spectral embedding with a controlled error. If this is right, an assumption earlier work had to take on faith becomes a theorem with explicit capacity bounds: representation learning on augmented data is guaranteed to be realizable at the network sizes they specify.

What carries the argument

The argument is carried by a chain of three operators: the augmentation graph Laplacian $L_{\mathrm{aug}}$ is first compared to a projected graph Laplacian $L_{\mathrm{proj}}$ defined on the orthogonal projections of augmented points onto the manifold; $L_{\mathrm{proj}}$ is then compared to a nonlocal integral operator $L_M$ with kernel $\varsigma(x,y)$; and $L_M$ is Taylor-expanded to order $\varepsilon^2$, where the weighted Laplace-Beltrami operator $\Delta_{\mathrm{aug}}f = -\tfrac12\mathrm{div}(q^2\nabla f)$ emerges, with $q \approx \rho$. The parameter hierarchy $\varepsilon_w = \varepsilon^\tau$, $\varepsilon_p = \eta^{1/d}\varepsilon^{\tau+1}$, $\varepsilon_n = \varepsilon^{\tau+1}$ for $\tau \geq 3$ is what makes each comparison exact: it keeps augmented points inside an $\varepsilon_n$-tubular neighborhood of $M$ while making the edge weights behave like a smooth kernel. Spectral transfer is mediated by $\infty$-optimal-transport-based discretization and interpolation maps that act as near-isometries between graph Dirichlet energy and continuum Dirichlet energy, following a variational min-max argument. The final step combines elliptic regularity of eigenfunctions with a ReLU approximation theorem for functions on low-dimensional manifolds to obtain the network size bounds.

What would settle it

On a flat torus or a dumbbell, where the Laplace-Beltrami spectrum is known exactly, build the augmentation graph with the prescribed parameter scaling, compute the $l$-th graph eigenvector, and compare it with the $l$-th eigenfunction evaluated at the raw augmented points, not at their projections onto the manifold. If the error does not decay at the rate stated in Theorem 2 as $\varepsilon \to 0$, the operator chain $L_{\mathrm{aug}} \to L_M \to \Delta_{\mathrm{aug}}$ fails. Separately, train a ReLU network of exactly the stated depth and width on the raw augmented points and test whether its output matches the graph eigenvector on those points; failure here would falsify the unproved projection step of Theorem 3.

Watch

Extended reading notes

Core claim

The central claim is a chain of three theorems. First, for a smooth closed $m$-dimensional manifold $M$ with sampling density $\rho$, the augmentation graph Laplacian $L_{\mathrm{aug}}$, whose edge weights measure the expected probability that two augmented points came from the same natural point, differs pointwise from the weighted Laplace-Beltrami operator, the manifold analogue of the ordinary Laplacian, $\Delta_{\mathrm{aug}}f = -\tfrac12\mathrm{div}(q^2\nabla f)$, by $O(\varepsilon)$ with high probability, where $q$ is the density of the projected augmented points and $\varepsilon$ is the graph connectivity scale. Second, given an eigengap, the $l$-th normalized eigenvector of $L_{\mathrm{aug}}$, evaluated at a point through its projection $Q$ onto $M$, is within $C_{\eta,l}\varepsilon$ (in $L^2$ over the augmented sample) of the $l$-th eigenfunction of $\Delta_{\mathrm{aug}}$, with probability at least $1 - n\exp(-c n \iota_1^2 \iota_2^m) - 10n\exp(-c_\eta n \varepsilon^{m+4})$. Third, the smoothness of those eigenfunctions, via an approximation theorem for ReLU networks, yields a network of depth $O(\log(1/\tilde\delta)+\log d)$ and width $O(k(\tilde\delta^{-m}+d))$ that reproduces the spectral contrastive embedding to accuracy $O(\varepsilon+\tilde\delta)$. The convergence rate depends on the manifold dimension $m$, not the ambient dimension $d$, which the paper highlights as a mitigation of the curse of dimensionality.

Load-bearing premise

The proof of the network approximation result bounds the network's error on the projected points, while the network must run on the raw augmented points; bridging that gap would require the network to also reproduce the orthogonal projection or a constant-along-normal extension of the eigenfunction, and the paper does not show that step.

Editorial extensions

If this is right

  • With $\varepsilon = C(\log n/n)^{1/(m+4)}$, the graph spectrum converges to the manifold spectrum at rate $O((\log n/n)^{1/(m+4)})$ up to log factors, so manifold geometry is statistically accessible from augmented data alone.
  • The realizability assumption is discharged: a ReLU network of depth $O(\log(n/\log n))$ and width $O((n/\log n)^{m/(m+4)})$ approximates the spectral contrastive embedding, giving explicit capacity bounds for the first time.
  • Because rates depend on the intrinsic dimension $m$ and not the ambient dimension $d$, the guarantees remain non-vacuous for high-dimensional inputs such as images.
  • The parameter prescription ($\varepsilon_p \geq \varepsilon_w$ with $\tau \geq 3$) is corroborated empirically: with $\varepsilon_p = \varepsilon_w$, the augmentation graph's leading eigenvectors miss the bottleneck of a dumbbell manifold, and with the new scaling they detect it, supporting fine-grained recovery on a shared manifold.

Reading between the lines

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

  • If the missing projection step in Theorem 3 were supplied, the same two-scale weight structure would likely extend the consistency chain to other secondary-similarity graphs, for example shared-nearest-neighbor graphs, since the proof uses only the kernel shape and tubular-neighborhood concentration.
  • A testable design rule follows implicitly: the augmentation noise $\varepsilon_p$ must shrink faster than the edge-weight scale $\varepsilon_w$ for the graph to reveal the manifold rather than the augmentation noise, which could guide noise scheduling in practical contrastive pipelines.
  • The existence guarantee does not say that gradient-based training reaches the network realizing the embedding; connecting spectral convergence to the optimization landscape of the contrastive loss is the natural next step for an end-to-end guarantee.
  • Since the continuum limit's density $q$ is shaped by the augmentation process itself, the learned embedding encodes augmentation geometry as much as data geometry; varying the augmentation distribution and observing the resulting eigenbasis would probe this distinction.
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

2 major / 4 minor

Summary. The paper studies the augmentation graph Laplacian of HaoChen et al. when augmented data are Gaussian perturbations of samples from a compact low-dimensional manifold. It proves a pointwise consistency result (Theorem 1) and a spectral consistency result (Theorem 2) showing that, with high probability, the augmentation graph Laplacian converges to a weighted Laplace-Beltrami operator on the natural data manifold. It then uses these results to assert a neural approximability theorem for spectral contrastive embeddings (Theorem 3), with explicit depth, width, and parameter-count bounds, which is presented as resolving the realizability assumption of the spectral contrastive learning framework. The proofs systematically adapt the Calder-Trillos / Garcia-Trillos et al. framework to the augmented-data setting and contain detailed probability estimates.

Significance. If Theorem 3 were established as stated, the paper would make a valuable contribution: it would turn the realizability assumption in contrastive learning into a concrete approximation guarantee with rates depending on the manifold dimension m rather than the ambient dimension d. The paper is unusually explicit about parameter regimes, assumptions, and probability bounds, and it credits the external spectral-convergence framework rather than obscuring its dependence. Theorems 1 and 2 appear carefully executed and are the strongest part of the manuscript. However, the advertised headline claim rests on the final approximation step in Section 8, which has a load-bearing gap: the proof approximates graph eigenvectors by networks evaluated at projected points, while the theorem requires evaluation at the raw augmented inputs. For this reason the realizability resolution is not established as written.

major comments (2)
  1. [Section 8, Eqs. (8.5)-(8.7)] The proof of Theorem 3 compares the graph eigenvector f_l^aug with the network value (f_l)_theta(Q dot), that is, at the orthogonal projection Q of an augmented point onto M, while the rows of Y_theta in (3.25) are required to be f_theta(bar x_i) at the raw augmented points. The only quantitative link between bar x and Q bar x in the manuscript is dist(bar x, M) <= epsilon^{tau+1}, and no Lipschitz or off-manifold control is proved for the ReLU approximant obtained from [3, Thm 3.1]. In particular, the paper neither constructs a ReLU network approximating f_l composed with Q on the tubular neighborhood N_{epsilon_n}(M), nor shows that the approximating network is nearly constant along normal directions. Therefore the chain (8.5)-(8.6) does not yield (8.7) when Y_theta is evaluated at bar x_i. Since this is exactly the step advertised as resolving the realizability assumption, the theorem is not established as stated; a repair needs an explicit off-manifold construction or a uniform Lipschitz estimate for the approximating network.
  2. [Section 8, expression for Y^*] The displayed formula for a minimizer of (3.18) is not correct for the problem as stated. With eigenpairs (lambda_l^aug, v_l) of L_aug and v_l normalized in the Euclidean norm, the rank-k truncated SVD of a I_n - L_aug has columns with factors sqrt(a - lambda_l^aug)/sqrt(n), not sqrt(lambda_l^aug)/sqrt(n). Replacing a - lambda_l^aug by lambda_l^aug changes Y^* (Y^*)^T and therefore need not minimize the objective in (3.18). The proof goes on to construct a network approximating sqrt(lambda_l^aug) f_l, so even after fixing the raw-input issue, the constructed Y_theta approximates the wrong target. The minimizer normalization must be rederived, and the dependence of the network complexity bounds on a should be tracked. The normalization inconsistency between the operator definition (2.4) and the matrices A_aug, D_aug in Section 3.3 contributes to this confusion and should also be fixed.
minor comments (4)
  1. [Section 1.1] The simulation uses tau = 2 and eta = 1, but Assumption 3 and the main theorems require tau >= 3; the text should either justify why the simulation is informative outside the proven regime or adjust the parameters to satisfy the stated assumptions.
  2. [Section 8] The proof contains typos that make the construction hard to follow: f_theta is described as mapping R^d to R^l instead of R^k, and 'the jth normalized eigenvector' should be 'the l-th normalized eigenvector'.
  3. [Section 3.3 and Section 8] There is an unresolved convention mismatch for Y_theta: Eq. (3.25) defines rows as f_theta(bar x_i), while the proof inserts an explicit 1/sqrt(n) factor outside the stacked network outputs. The theorem should adopt one convention and state explicitly whether the 1/sqrt(n) scaling is absorbed into the network function.
  4. [Theorem 2] The probability statement of Theorem 2 does not include the failure probability of Assumption 2 from Lemma 1; if the theorem is conditional on Assumption 2, that should be stated explicitly, and if not, the event should be added to the probability bound.

Circularity Check

0 steps flagged · score 1.0 of 10

No load-bearing circularity: the spectrum proof rests on external graph-Laplacian/transport results and the network step on external ReLU approximation theory.

full rationale

Walking the derivation chain, the main consistency argument is not circular. Theorem 1 compares Laug to a projected graph Laplacian (Propositions 2 and 3) and then to the nonlocal operator LM, whose expansion yields the weighted Laplace-Beltrami operator through direct Taylor and Gaussian-integral calculations (Lemmas 4-6); the input distributions determine both the graph weights and the limit operator, but the convergence statement is a substantive probabilistic limit, not a restatement of an assumption. Theorem 2 adapts the external variational and optimal-transport machinery of Calder and Garcia Trillos [11], with the density and variable-edge-length modifications proved in Propositions 5-7, and the eigenvector closeness (4.6) is not assumed as an input. Theorem 3 uses the external ReLU approximation theorem [3, Theorem 3.1] and elliptic regularity of Laplace-Beltrami eigenfunctions; the approximating network is constructed, not fitted to the target eigenvectors, so no fitted parameter is renamed as a prediction. The only self-citations, [14] and [17], appear as background or related-work framing and are not used as load-bearing evidence in any proof. The duplicated reference [10]=[28] is an external prior work, not a self-citation chain. The proof of Theorem 3 does contain a technical gap: (8.5)-(8.6) control (f_l)_theta evaluated at Q, the projection onto M, while Ytheta is evaluated at the raw augmented points x_i, and the paper does not show that the constructed network computes or is constant along the normal direction. This is a missing argument affecting correctness, not a circular reduction; no equation in the paper makes the claimed conclusion equal to its own input by construction.

Assumptions & free parameters 4 free parameters · 8 assumptions · 0 invented entities

The central claim rests on a specific Gaussian augmentation model, a bounded-dependency assumption, and the Calder-Trillos spectral framework. No constants are fitted; the analysis contains hand-chosen scales tau, eta, iota1, iota2. The augmentation graph itself was introduced by HaoChen et al., not invented here.

free parameters (4)
  • epsilon (graph connectivity scale) = C(log n / n)^{1/(m+4)} in optimal rate statements
    Bandwidth of the augmentation graph; not fitted to data but chosen to balance bias and variance.
  • tau = tau >= 3
    Controls separation between augmentation noise epsilon_p=epsilon^{tau+1}, edge weight scale epsilon_w=epsilon^tau, and graph bandwidth epsilon; required for kernel estimates and Assumption 3.
  • eta = eta > 0
    Controls the constant in epsilon_p and appears in probability estimates and constants.
  • iota1, iota2 = iota1=C epsilon^2, iota2=c epsilon in the application
    Optimal transport and interpolation parameters in Assumption 4; chosen to satisfy spectral convergence conditions.
assumptions (8)
  • domain assumption M is a smooth closed orientable m-dimensional submanifold of R^d with bounded sectional curvature, reach and injectivity radius.
    Section 2.2; gives tangent and normal coordinates, exponential maps, and volume estimates.
  • domain assumption Natural data are i.i.d. from density rho in C^2(M) with 0 < rho_min <= rho <= rho_max < infinity.
    Equation (2.14); controls sampling and density regularity.
  • ad hoc to paper Augmented data follow Gaussian noise P(bar x | x) = pi_{epsilon_p}(bar x | x) with epsilon_p = eta^{1/d} epsilon^{tau+1}.
    Equation (2.1); specific augmentation model assumed throughout the analysis.
  • ad hoc to paper Assumption 1: each augmented point has at most C0 dependent points, equivalently at most C0 augmented points per natural point.
    Section 2.1; needed to apply concentration lemmas for dependent samples.
  • ad hoc to paper Assumption 2: all augmented points lie within the epsilon_n-tubular neighborhood of M.
    Section 3.1; holds with high probability by Lemma 1 and is then imposed to define projections.
  • ad hoc to paper Parameter smallness: epsilon <= min{1, K^{-1/2}, i0, R/2}, tau >= 3, and inequalities (3.2), (3.7) or (3.17) hold.
    Assumptions 3 and 4; ensure kernel decay and derivative bounds.
  • standard math Regularity of Laplace-Beltrami eigenfunctions and the ReLU approximation theorem of Chen, Jiang, Liao and Zhao.
    Used in Theorem 3 as external results.
  • standard math Spectral convergence framework of Calder and Garcia Trillos, including their optimal transport and interpolation propositions.
    External framework adapted to augmentation graph Laplacians.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Consistency of augmentation graph and network approximability in contrastive learning." pith.science (2026). https://pith.science/paper/47TCKIEP

@misc{pith2026250204312,
  author       = {Pith},
  title        = {Pith review of: Consistency of augmentation graph and network approximability in contrastive learning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/47TCKIEP}},
  note         = {Machine review of arXiv:2502.04312}
}
read the original abstract

Contrastive learning leverages data augmentation to develop feature representation without relying on large labeled datasets. However, despite its empirical success, the theoretical foundations of contrastive learning remain incomplete, with many essential guarantees left unaddressed, particularly the realizability assumption concerning neural approximability of an optimal spectral contrastive loss solution. In this work, we overcome these limitations by analyzing pointwise and spectral consistency of the augmentation graph Laplacian. We establish that, under specific conditions for data generation and graph connectivity, as the augmented dataset size increases, the augmentation graph Laplacian converges to a weighted Laplace-Beltrami operator on the natural data manifold. These consistency results ensure that the graph Laplacian spectrum effectively captures the manifold geometry. Consequently, they give way to a robust framework for establishing neural approximability, directly resolving the realizability assumption in a current paradigm.

Figures

Figures reproduced from arXiv: 2502.04312 by the authors.

Figure 1
Figure 1. (a) When choosing εp, εw, εn according to [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. A two-dimensional sphere, one of its tangent planes, and an isomorphic copy of the latter as the [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 4
Figure 4. Qx¯ and Qx¯ ′ are the closest points to ¯x and ¯x ′ on M, respectively. T˜Qx¯M and T˜Qx¯′M are the tangent planes of M on Qx¯ and Qx¯ ′ , respectively. where ⊥ denotes the usual orthogonal notion in R d ; see [PITH_FULL_IMAGE:figures/full_fig_p038_4.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 26 canonical work pages

  1. [3]

    Nonparametric regression on low- dimensional manifolds using deep relu networks: Function approximation and statistical recovery

    Minshuo Chen, Haoming Jiang, Wenjing Liao, and Tuo Zhao. Nonparametric regression on low- dimensional manifolds using deep relu networks: Function approximation and statistical recovery. In- formation and Inference: A Journal of the IMA , 11(4):1203–1253, 2022

    work page 2022

  2. [7]

    Provable guarantees for self-supervised deep learning with spectral contrastive loss

    Jeff Z HaoChen, Colin Wei, Adrien Gaidon, and Tengyu Ma. Provable guarantees for self-supervised deep learning with spectral contrastive loss. Advances in Neural Information Processing Systems , 34, 2021

    work page 2021

  3. [10]

    Error Estimates for Spectral Convergence of the Graph Laplacian on Random Geometric Graphs Toward the Laplace–Beltrami Operator

    Nicol´ as Garc ´ ıa Trillos, Moritz Gerlach, Matthias Hein, and Dejan Slepˇ cev. Error Estimates for Spectral Convergence of the Graph Laplacian on Random Geometric Graphs Toward the Laplace–Beltrami Operator. Foundations of Computational Mathematics , 20(4):827–887, August 2020

    work page 2020

  4. [28]

    Error estimates for spectral convergence of the graph laplacian on random geometric graphs toward the laplace–beltrami operator

    Nicol´ as Garc ´ ıa Trillos, Moritz Gerlach, Matthias Hein, and Dejan Slepˇ cev. Error estimates for spectral convergence of the graph laplacian on random geometric graphs toward the laplace–beltrami operator. Foundations of Computational Mathematics , 20(4):827–887, 2020

    work page 2020

  5. [1]

    Data augmentation: A comprehensive survey of modern ap- proaches

    Alhassan Mumuni and Fuseini Mumuni. Data augmentation: A comprehensive survey of modern ap- proaches. Array, page 100258, 2022

    work page 2022

  6. [2]

    Text data augmentation for deep learning

    Connor Shorten, Taghi M Khoshgoftaar, and Borko Furht. Text data augmentation for deep learning. Journal of big Data , 8:1–34, 2021

    work page 2021

  7. [4]

    A theoretical analysis of contrastive unsupervised representation learning

    Sanjeev Arora, Hrishikesh Khandeparkar, Mikhail Khodak, Orestis Plevrakis, and Nikunj Saun- shi. A theoretical analysis of contrastive unsupervised representation learning. arXiv preprint arXiv:1902.09229, 2019

    arXiv 1902

  8. [5]

    Contrastive estimation reveals topic poste- rior information to linear models

    Christopher Tosh, Akshay Krishnamurthy, and Daniel Hsu. Contrastive estimation reveals topic poste- rior information to linear models. Journal of Machine Learning Research , 22(281):1–31, 2021

    work page 2021

Show all 33 references
  1. [6]

    Contrastive learning, multi-view redun- dancy, and linear models

    Christopher Tosh, Akshay Krishnamurthy, and Daniel Hsu. Contrastive learning, multi-view redun- dancy, and linear models. In Algorithmic Learning Theory, pages 1179–1206. PMLR, 2021

    2021

  2. [8]

    Elliptic partial differential equations of second order, volume 224

    David Gilbarg, Neil S Trudinger, David Gilbarg, and NS Trudinger. Elliptic partial differential equations of second order, volume 224. Springer, 1977

    1977

  3. [9]

    A graph discretization of the Laplace-Beltrami operator

    Dmitri Burago, Sergei Ivanov, and Yaroslav Kurylev. A graph discretization of the Laplace-Beltrami operator. Journal of Spectral Theory , 4(4):675–714, 2014

    2014

  4. [11]

    Improved spectral convergence rates for graph laplacians on ε-graphs and k-nn graphs

    Jeff Calder and Nicol´ as Garc ´ ıa Trillos. Improved spectral convergence rates for graph laplacians on ε-graphs and k-nn graphs. Applied and Computational Harmonic Analysis , 60:123–175, 2022

    2022

  5. [12]

    Spectral convergence of graph laplacian and heat kernel reconstruction in l∞ from random samples

    David B Dunson, Hau Tieng Wu, and Nan Wu. Spectral convergence of graph laplacian and heat kernel reconstruction in l∞ from random samples. Applied and Computational Harmonic Analysis, 55:282–336, 2021

    2021

  6. [13]

    Spectral convergence of diffusion maps: Improved error bounds and an alternative normalization

    Caroline L Wormell and Sebastian Reich. Spectral convergence of diffusion maps: Improved error bounds and an alternative normalization. SIAM Journal on Numerical Analysis , 59(3):1687–1734, 2021

    2021

  7. [14]

    Large sample spectral analysis of graph-based multi-manifold clustering

    Nicolas Garcia Trillos, Pengfei He, and Chenghui Li. Large sample spectral analysis of graph-based multi-manifold clustering. Journal of Machine Learning Research , 24(143):1–71, 2023

    2023

  8. [15]

    Lipschitz regularity of graph laplacians on random data clouds

    Jeff Calder, Nicolas Garcia Trillos, and Marta Lewicka. Lipschitz regularity of graph laplacians on random data clouds. SIAM Journal on Mathematical Analysis , 54(1):1169–1222, 2022. 33

    2022

  9. [16]

    Refined shared nearest neighbors graph for combining multiple data clusterings

    Hanan Ayad and Mohamed Kamel. Refined shared nearest neighbors graph for combining multiple data clusterings. In International Symposium on Intelligent Data Analysis , pages 307–318. Springer, 2003

    2003

  10. [17]

    Spectral neural networks: Approximation theory and optimization landscape

    Chenghui Li, Rishi Sonthalia, and Nicolas Garcia Trillos. Spectral neural networks: Approximation theory and optimization landscape. arXiv preprint arXiv:2310.00729 , 2023

    2023 arXiv

  11. [18]

    Small relu networks are powerful memorizers: a tight analysis of memorization capacity

    Chulhee Yun, Suvrit Sra, and Ali Jadbabaie. Small relu networks are powerful memorizers: a tight analysis of memorization capacity. Advances in Neural Information Processing Systems , 32, 2019

    2019

  12. [19]

    Probability inequalities for sums of bounded random variables

    Wassily Hoeffding. Probability inequalities for sums of bounded random variables. The collected works of Wassily Hoeffding , pages 409–426, 1994

    1994

  13. [20]

    Riemannian geometry, volume 6

    Manfredo Perdigao Do Carmo and J Flaherty Francis. Riemannian geometry, volume 6. Springer, 1992

    1992

  14. [21]

    Sobolev spaces on Riemannian manifolds , volume 1635

    Emmanuel Hebey. Sobolev spaces on Riemannian manifolds , volume 1635. Springer Science & Business Media, 1996

    1996

  15. [22]

    A tutorial on spectral clustering

    Ulrike Von Luxburg. A tutorial on spectral clustering. Statistics and computing , 17(4):395–416, 2007

    2007

  16. [23]

    The approximation of one matrix by another of lower rank.Psychometrika, 1(3):211–218, 1936

    Carl Eckart and Gale Young. The approximation of one matrix by another of lower rank.Psychometrika, 1(3):211–218, 1936

    1936

  17. [24]

    Exploring simple siamese representation learning

    Xinlei Chen and Kaiming He. Exploring simple siamese representation learning. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages 15750–15758, 2021

    2021

  18. [25]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky, Geoffrey Hinton, et al. Learning multiple layers of features from tiny images. 2009

    2009

  19. [26]

    Big self- supervised models are strong semi-supervised learners

    Ting Chen, Simon Kornblith, Kevin Swersky, Mohammad Norouzi, and Geoffrey E Hinton. Big self- supervised models are strong semi-supervised learners. Advances in neural information processing sys- tems, 33:22243–22255, 2020

    2020

  20. [27]

    A simple framework for contrastive learning of visual representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations. In International conference on machine learning , pages 1597–1607. PMLR, 2020

    2020

  21. [29]

    The game theoretic p-laplacian and semi-supervised learning with few labels

    Jeff Calder. The game theoretic p-laplacian and semi-supervised learning with few labels. Nonlinearity, 32(1):301, 2018

    2018

  22. [30]

    Large deviations for sums of partly dependent random variables

    Svante Janson. Large deviations for sums of partly dependent random variables. Random Structures & Algorithms, 24(3):234–248, 2004

    2004

  23. [31]

    Differential topology, volume 370

    Victor Guillemin and Alan Pollack. Differential topology, volume 370. American Mathematical Soc., 2010

    2010

  24. [32]

    Lectures on classical differential geometry

    Dirk Jan Struik. Lectures on classical differential geometry . Courier Corporation, 1961

    1961

  25. [33]

    first, second, third, fourth, fifth, sixth

    Lawrence C Evans. Partial differential equations , volume 19. American Mathematical Society, 2022. 34 A Relevant concentration results We present the following two known concentration lemmas. Lemma A.1. [11, Lemma 3.1 and its proof] (see also [29, Lemma 1]) Let M satisfy the a...

    2022

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.