Pith. sign in

REVIEW 4 major objections 8 minor 40 references

Intrinsic Green's Learning: Supervised Learning on Manifolds via Inverse PDE

T0 review · 4 major / 8 minor · reviewed 2026-07-09 · glm-5.2

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

Pith's one-line read Learn the source, not the solution: inverse-PDE learning on manifolds

desk verdict Novel inverse-PDE framework with sound factorization math, but validation is thin and the low-rank source assumption is unverified on the hardest experiment. read the letter →

arxiv 2607.07034 v1 pith:VR46IEOM submitted 2026-07-08 cs.LG cs.AI

Alexandre Quemy This is my paper · ORCID
classification cs.LGcs.AI
keywords intrinsicsourcecoordinatedimensiongreenlearninglinearmanifold
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 proposes that instead of directly fitting a target function on a high-dimensional manifold, one should learn a source term whose convolution against a Green's kernel reproduces the target. The source and kernel are both expressed as low-rank tensor decompositions in coordinates discovered by a neural encoder, which collapses a d-dimensional integral into d independent one-dimensional integrals with cost linear in the intrinsic dimension d. A two-stage algorithm—alternating between coordinate discovery (outer, nonlinear) and source fitting (inner, exact linear least squares)—prevents the dimensional collapse that plagues joint optimization, and learnable gates on each coordinate automatically zero out unnecessary dimensions, recovering the manifold's intrinsic dimension. On MNIST, the method achieves 0.991 linear-probe accuracy while automatically selecting 12 of 64 latent dimensions, consistent with prior estimates of MNIST's intrinsic dimension.

What carries the argument

Three components: (1) an encoder Ψ: R^D → R^d discovering coordinates where the source admits CP tensor decomposition of rank R; (2) a rank-K separable approximation of the Green's function via the exponential sum trick, which converts the additive eigenvalue coupling 1/(Σ λ_j) into a product of exponentials; (3) a two-stage variable-projection optimizer that solves the inner linear system (source weights w via least squares) to optimality before each outer encoder update. Learnable gates g_j ∈ [0,1] with Group Lasso discover effective dimension; rank gates ρ_r discover effective source rank. Polynomial augmentation handles null-space modes (constants, linear trends) that would otherwise要求高秩

What would settle it

Construct a target function on a manifold that does not admit low-rank tensor structure in any discoverable coordinate chart (e.g., a function with irreducible cross-coordinate coupling), and show that IGL's effective rank grows unboundedly or its accuracy degrades relative to direct approximation methods regardless of encoder capacity or training procedure.

Watch

Extended reading notes

Core claim

The central mechanism is the separation of geometry from fitting. By parameterizing the target as u = G * f (Green's kernel convolved with a learned source), the fitting problem for source weights becomes linear given fixed coordinates and kernel scales. Solving this linear problem to optimality at each encoder step (variable projection) forces the encoder gradient to reflect only coordinate quality, not source-capacity shortcuts. This is what prevents dimensional collapse: the encoder cannot compensate for bad coordinates by inflating source rank, because the source is always optimally fit. The tensor factorization of both source and kernel then makes the integral tractable via Fubini's定理,O

Load-bearing premise

The source term must admit a low-rank tensor decomposition in the coordinates the encoder discovers. The paper acknowledges this is conditional on the encoder finding coordinates where the target has bounded mixed derivatives, but provides no theorem guaranteeing convergence to such coordinates—the evidence is empirical, limited to Swiss Roll and MNIST.

Editorial extensions

If this is right

  • Any supervised learning problem where data lies on a low-dimensional manifold could benefit from replacing direct function approximation with source-term learning, gaining automatic dimension discovery and topology preservation as byproducts.
  • The two-stage variable-projection principle—solving the linear inner problem to optimality before updating the nonlinear outer parameters—could be applied to other architectures with linear heads over nonlinear embeddings, wherever joint training causes dimensional collapse.
  • Operator choice (Laplacian, Helmholtz, harmonic oscillator, fractional) becomes a tunable inductive bias knob: smoothness, locality, oscillation, or long-range correlations, each selected per problem class.
  • The framework naturally extends to meta-learning: amortize Stage 1 (coordinate discovery) across a task distribution, adapting only the lightweight Stage 2 (O(KR) parameters) per new task.

Reading between the lines

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

  • If the two-stage principle generalizes, it suggests that dimensional collapse in representation learning is primarily an optimization-architecture mismatch, not a fundamental capacity limitation—joint training allows the encoder to find degenerate shortcuts that exact inner solves eliminate.
  • The gauge symmetry (invariance of the integral under coordinate diffeomorphisms with inverse source transformation) implies the encoder is free to discover any chart that diagonalizes tensor rank, not necessarily an isometric embedding—this is a weaker and potentially more achievable goal than manifold isometry.
  • The reliance on bounded mixed derivatives for low-rank tensor approximation suggests IGL will struggle on targets with strong cross-coordinate coupling (e.g., radial functions in Cartesian coordinates) unless the encoder finds the diagonalizing chart—empirically demonstrated on simple manifolds but unproven for complex topologies.
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 / 8 minor

Summary. The paper introduces Intrinsic Green's Learning (IGL), a framework for supervised learning on manifolds that models the target function as the solution to a linear PDE with a learned source term. The key technical contribution is a two-stage algorithm: Stage 1 discovers a coordinate chart via an encoder where the source and Green's kernel admit low-rank tensor decompositions, and Stage 2 solves for source weights via a near-convex linear solve. The tensor structure allows a d-dimensional integral to be factored into d independent 1D integrals via Fubini's theorem, yielding O(KRd) complexity. The paper validates the approach on synthetic manifolds (Swiss Roll, rotated planes) and MNIST, demonstrating topology preservation, sample efficiency, and automatic intrinsic dimension discovery.

Significance. The paper presents a novel and mathematically grounded approach to learning on manifolds. The core insight—inverting the PDE perspective to learn a source term rather than the solution directly, and using tensor decomposition to make the resulting integral tractable—is elegant and well-motivated. The two-stage variable projection algorithm is a principled solution to the dimensional collapse problem, and the envelope theorem argument (Appendix C.1) provides a clean theoretical justification for why the separation prevents collapse. The Fubini factorization proofs (Appendix A.1-A.3) are algebraically correct. The MNIST results, showing 0.991 linear probe accuracy with 12 of 64 gates active, are promising if the underlying assumptions hold. The framework's modularity (operator choice, encoder architecture) is a practical strength.

major comments (4)
  1. The central computational claim of O(KRd) complexity depends entirely on the source term f admitting a low-rank CP tensor decomposition (Eq. 2, §3.2). The paper acknowledges this is 'conditional on coordinates' (Appendix B.1) and invokes bounded mixed derivatives (Eq. 19) as the theoretical justification. However, the paper's most complex experiment—MNIST (Table 3, §4)—does not directly verify that the learned source is actually low-rank. Table 5 lists the rank importance parameter ρ_r as 'Learned' via softplus gating, but no values of the effective rank R_eff are reported for MNIST. The experiment uses R=128 anchors but does not report how many are active, nor does it report classification accuracy as a function of R. Without this, the reader cannot distinguish between (a) the source genuinely admitting low-rank structure (small R_eff, graceful degradation when R is reduced) and (b) the
  2. polynomial/null-space augmentation terms (Eq. 5) or the CNN encoder doing the real work, with the Green's integral contributing marginally. The Swiss Roll ablation (Table 2) shows rank-dependent MSE improvement, but only on a 2D manifold with a smooth target. The authors should report R_eff and accuracy-vs-R curves for MNIST to substantiate that the low-rank assumption holds in the paper's hardest setting.
  3. The claim that Stage 2 is 'near-convex' (§3.1, §5) requires qualification. The linearity holds only for fixed encoder parameters θ_Ψ and fixed kernel parameters {γ_k, σ_k,j} (Appendix C.1). However, Table 5 and Algorithm 1 indicate that the anchor positions μ_{r,j} and kernel widths σ_{k,j} are updated in the outer loop (Stage 1), meaning the design matrix Φ (Eq. 24) depends nonlinearly on these parameters. The paper should clarify which parameters are truly fixed in Stage 2 (only w_r and c_β?) versus which are jointly optimized in Stage 1, and soften the 'near-convex' language accordingly. As stated, the claim overstates the convexity of the actual optimization problem being solved.
  4. The two-stage vs. joint training comparison (Figure 3, §4) attributes the performance gap to optimization structure alone, stating 'architectures are identical.' However, the two-stage method solves Stage 2 exactly via least squares at each outer step, which is computationally equivalent to an infinite number of inner gradient steps. The comparison is therefore between joint training with a finite (and unspecified) number of inner steps versus two-stage training with exact inner solves. To isolate the effect of optimization structure from the effect of inner solve quality, the authors should either (a) report the number of inner gradient steps used for joint training, or (b) include a baseline where joint training uses a large number of inner steps to approximate the exact solve. Without this, the 'phase transition' in Figure 3 may reflect under-optimized joint training rather than afund
minor comments (8)
  1. §3.1: The loss function includes a Group Lasso penalty λ ||w_r γ||_2, but the role of γ_k in this penalty is unclear. Should this be ||w_r||_2 or ||w_r γ_k||_2? The notation is ambiguous.
  2. Table 2: The d_eff column reports values of 3 for all operators, but the text refers to 'true d=2' with 'embedding curvature overhead.' It would help to clarify whether d_eff=3 is a consistent finding or an artifact of the greedy knockout procedure.
  3. Figure 2: The convergence curves (columns 3-4) are difficult to read due to axis labeling. Consider using log-scale for the y-axis and clarifying what 'MSE' refers to (training or test).
  4. Appendix B.1: The statement 'functions with bounded mixed derivatives do admit efficient tensor approximations' cites Khoromskij (2012), but the connection between bounded mixed derivatives and the specific CP decomposition used in Eq. 2 could be made more explicit. The current treatment is somewhat hand-wavy.
  5. §4, Exp. 7: The MNIST experiment uses a 'shared CNN encoder' but the encoder architecture is not specified. The paper should state the number of layers, hidden dimensions, and whether the encoder is pre-trained or trained from scratch.
  6. Table 3: The 'Smooth' metric is not defined in the main text. The reader must infer its meaning from context. A brief definition would improve readability.
  7. Appendix C.2, Eq. 28: The distinction between the particular solution (computed by IGL) and the homogeneous solution (free parameters) is clear, but the paper should note that for the Laplacian on a bounded domain, the null space includes constants and linear functions, which are absorbed by the polynomial augmentation (Eq. 5). This connection is mentioned but not made explicit enough.
  8. The paper uses both 'EDP' (Figure 1 caption) and 'PDE' (main text). 'EDP' appears to be a typo or non-standard abbreviation and should be corrected to 'PDE'.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found. The derivation chain is self-contained and conditional on clearly stated assumptions.

full rationale

The paper's core mathematical derivation proceeds as follows: (1) The target u is modeled as a Green's convolution u = G*f (Eq. 1), a standard PDE representation. (2) The source f is given a CP tensor decomposition (Eq. 2) and the Green's function a rank-K tensor approximation (Eq. 3) — both explicitly presented as ansätze, not derived from conclusions. (3) Fubini's theorem then factors the d-dimensional integral into d 1D integrals (Eq. 4), yielding O(KRd) complexity. This is a straightforward conditional result: IF the source and kernel admit low-rank tensor structure, THEN the integral factors. The paper is transparent that the source low-rank property is 'conditional on coordinates' (Appendix B.1) and invokes bounded mixed derivatives as the theoretical justification — an assumption, not a conclusion smuggled back as input. The two-stage Variable Projection algorithm uses the envelope theorem to argue that solving Stage 2 to optimality makes encoder gradients reflect coordinate quality; this is standard optimization theory, not circular. The compensation mechanism (Appendix B.2) where f̂ absorbs the metric √|g| is acknowledged as a representational choice with stated limitations ('This is not a problem if we are not interested in preserving the geometric structure of the manifold'). The MNIST intrinsic dimension discovery (12 of 64 gates) is validated against external citations (Pope et al., 2021; Ansuini et al., 2019) by different author groups — no self-citation exists (the paper is single-authored by Alexandre Quemy with zero self-citations in the reference list). The skeptic's concern that MNIST does not report effective source rank R_eff is a valid empirical completeness issue, but it is not circularity: the paper does not claim to verify low-rank structure on MNIST, and the dimension discovery result is an output of optimization, not a fitted input renamed as prediction. No step in the derivation chain reduces to its own inputs by construction.

Assumptions & free parameters 12 free parameters · 6 assumptions · 3 invented entities

The framework introduces 12 free parameters (6 learned, 6 hyperparameters) and relies on 6 axioms (3 domain assumptions, 1 standard math, 2 ad-hoc-to-paper). The most consequential ad-hoc axiom is that the two-stage split prevents dimensional collapse; this is supported empirically but not proven. The low-rank source assumption is the most fragile domain assumption. The invented entities (compensated source, dimension gates, ResIGL) are well-motivated, with gates and ResIGL having falsifiable handles.

free parameters (12)
  • Encoder weights θ_Ψ
    Learned network parameters for coordinate discovery; architecture and dimension are hyperparameters.
  • Dimension gates g_j
    Learned [0,1] gates that discover intrinsic dimension; 12 active on MNIST.
  • Source weights w_r
    Linear coefficients of the CP tensor decomposition; solved via least squares in Stage 2.
  • Anchor positions μ_{r,j}
    Positions of RBF basis functions in the source decomposition; learned in outer loop.
  • Kernel scale weights γ_k
    Weights for the K-term Green's function approximation; learned.
  • Kernel widths σ_{k,j}
    Per-coordinate kernel widths; learned in outer loop.
  • Polynomial coefficients c_β
    Null-space augmentation coefficients; learned via least squares.
  • Regularization λ
    Group Lasso penalty weight; hyperparameter.
  • Max dimension d_max
    Upper bound on intrinsic dimension; hyperparameter (64 for MNIST).
  • Max source rank R_max
    Upper bound on tensor rank; hyperparameter (128 for MNIST).
  • Green's rank K
    Number of kernel components; hyperparameter (4 length scales for MNIST).
  • Null-space degree p
    Polynomial augmentation degree; hyperparameter.
assumptions (6)
  • domain assumption The target function u is the solution to a linear PDE Lu=f with learned source f.
    Stated in §1 and §3.1 as an 'architectural choice, not a physical claim.' This is the foundational assumption of the framework.
  • domain assumption The source term f admits a low-rank CP tensor decomposition in the learned coordinates.
    Eq. 2, §3.2. Appendix B.1 acknowledges this is 'conditional on coordinates' and relies on bounded mixed derivatives, which the encoder must discover.
  • standard math The Green's function admits a rank-K separable approximation via the exponential sum trick.
    Eq. 3, §3.2; justified in Appendix B.1 citing Hackbusch & Khoromskij (2006) and Beylkin & Monzón (2010). Rigorous for asymptotically smooth kernels with K=O(log(1/ε)).
  • domain assumption The metric volume element sqrt(|g|) can be absorbed into the compensated source without breaking factorization.
    Appendix A.3 and B.2. The paper argues this is representational (learned end-to-end) but notes it incurs a multiplicative rank cost.
  • ad hoc to paper The two-stage variable projection prevents dimensional collapse by solving Stage 2 to optimality.
    §3.1 and Appendix C.1. The envelope theorem argument is sound, but the claim that this prevents collapse is empirical (Figure 2), not proven.
  • ad hoc to paper Joint optimization of encoder and source is prone to dimensional collapse.
    §1 and §3.1. Used to motivate the two-stage algorithm. Empirically supported by Figure 2 and Table 3 (Joint vs. Two-stage on MNIST).
invented entities (3)
  • Compensated source ˆf = ˜f · sqrt(|g|)
    purpose: Absorbs the metric volume element to maintain separability of the integral on curved manifolds.
    Introduced in Appendix A.3/B.2. The paper argues it is learned end-to-end with zero computational overhead, but provides no external validation that the absorbed rank cost is manageable for complex manifolds.
  • Dimension gates g_j ∈ [0,1] independent evidence
    purpose: Learnable scalar gates that discover intrinsic dimension by driving inactive coordinates to zero via Group Lasso.
    The paper provides falsifiable evidence: 12 of 64 gates active on MNIST, consistent with prior estimates of 10-15 (Pope et al., 2021). The mechanism is testable on other datasets.
  • ResIGL (hybrid MLP + IGL architecture) independent evidence
    purpose: Combines IGL's topology-preserving smooth branch with an MLP residual for sharp features.
    Experiment 5 (Figure 6) shows iso-height line preservation vs. pure MLP collapse on Swiss Roll with discontinuous target. Falsifiable on other discontinuous-target manifold tasks.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Intrinsic Green's Learning: Supervised Learning on Manifolds via Inverse PDE." pith.science (2026). https://pith.science/paper/VR46IEOM

@misc{pith2026260707034,
  author       = {Pith},
  title        = {Pith review of: Intrinsic Green's Learning: Supervised Learning on Manifolds via Inverse PDE},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VR46IEOM}},
  note         = {Machine review of arXiv:2607.07034}
}
read the original abstract

We introduce Intrinsic Green's Learning (IGL), a framework that models a target function on a manifold as the solution to a linear PDE whose source term is learned from data. Rather than approximating the target directly, IGL learns a source and integrates it against a Green's kernel. An encoder discovers a low-dimensional coordinate chart on the manifold where both the source and the kernel decompose as low-rank tensors, collapsing a high-dimensional integral into independent one-dimensional integrals with cost linear in the intrinsic dimension. A two-stage algorithm separates coordinate discovery from source fitting, a near-convex linear solve, preventing the dimensional collapse of joint training. Learnable gates on each coordinate automatically discover the intrinsic dimension of the manifold. We validate IGL on synthetic manifolds and on MNIST, where it simultaneously achieves near-optimal classification and automatic recovery of the intrinsic dimension.

Figures

Figures reproduced from arXiv: 2607.07034 by the authors.

Figure 1
Figure 1. IGL pipeline as a two-stage algorithm. Stage 1 (coordinate discovery): the encoder Ψ maps data from a manifold M ⊂ R D (left) to intrinsic coordinates ξ ∈ R d where the source decomposes. The tensor source ˆf(ζ) = P r wr Q j ϕr,j (ζ j ) is defined by anchor points µr,j (green dots). Green’s convolution G ∗ ˆf yields the solution u(ξ). Stage 2 (source fitting): source weights wr are solved via gradient descent or lea… view at source ↗
Figure 2
Figure 2. Two-stage training prevents dimensional collapse on the Swiss Roll. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Sample efficiency on a rotated plane (d=2, D=100). Three Green’s kernels (Spatial, Helmholtz, Spectral) are each trained jointly and two-stage, alongside MLP, Linear, and Kernel Ridge baselines. At N≤100, joint training degrades more gracefully; between N=200 and N=500, two-stage undergoes a phase transition with all three kernels exceeding R2>0.9. At N=2000, Spectral (Two-Stage) reaches R2=0.998, while all jointly-… view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Scalability of IGL on log-log axes, decomposed into encoder (Stage 1) and source-fitting [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: Gabor wavelets learn sharp, non-axis-aligned decision boundaries. Three datasets: [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]
Figure 6
Figure 6. Figure 6: Topological preservation in hybrid architectures. Swiss Roll with discontinuous target [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: Spectral IGL regression on a randomly rotated 2D manifold in [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: t-SNE visualization of autoencoder latent spaces on MNIST (10 classes). From left to [PITH_FULL_IMAGE:figures/full_fig_p009_8.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 40 canonical work pages

  1. [1]

    Journal of Computational Physics , volume=

    Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations , author=. Journal of Computational Physics , volume=. 2019 , publisher=

    work page 2019

  2. [2]

    Nature Machine Intelligence , volume=

    Learning nonlinear operators via DeepONet based on the universal approximation theorem of operators , author=. Nature Machine Intelligence , volume=. 2021 , publisher=

    work page 2021

  3. [3]

    Fourier Neural Operator for Parametric Partial Differential Equations

    Fourier neural operator for parametric partial differential equations , author=. arXiv preprint arXiv:2010.08895 , year=

    work page Pith review arXiv 2010

  4. [4]

    2002 , publisher=

    Learning with kernels: support vector machines, regularization, optimization, and beyond , author=. 2002 , publisher=

    work page 2002

  5. [5]

    SIAM Review , volume=

    Tensor decompositions and applications , author=. SIAM Review , volume=. 2009 , publisher=

    work page 2009

  6. [6]

    KAN: Kolmogorov-Arnold Networks

    KAN: Kolmogorov-Arnold Networks , author=. arXiv preprint arXiv:2404.19756 , year=

    work page Pith review arXiv

  7. [7]

    Proceedings of the Fourth Eurographics Symposium on Geometry Processing , volume=

    Poisson surface reconstruction , author=. Proceedings of the Fourth Eurographics Symposium on Geometry Processing , volume=

    work page

  8. [8]

    Predicting Structured Data , volume=

    A tutorial on energy-based learning , author=. Predicting Structured Data , volume=

    work page

Show all 40 references
  1. [9]

    arXiv preprint arXiv:2011.13456 , year=

    Score-based generative modeling through stochastic differential equations , author=. arXiv preprint arXiv:2011.13456 , year=

    2011 arXiv

  2. [10]

    Journal of Computational Physics , volume=

    A fast algorithm for particle simulations , author=. Journal of Computational Physics , volume=. 1987 , publisher=

    1987

  3. [11]

    Part I: Introduction to H-matrices , author=

    A sparse matrix arithmetic based on H-matrices. Part I: Introduction to H-matrices , author=. Computing , volume=. 1999 , publisher=

    1999

  4. [12]

    Royal Signals and Radar Establishment Malvern (United Kingdom) , year=

    Radial basis functions, multi-variable functional interpolation and adaptive networks , author=. Royal Signals and Radar Establishment Malvern (United Kingdom) , year=

  5. [13]

    ICML Workshop on Theoretical Foundations and Applications of Deep Generative Models , year=

    Neural processes , author=. ICML Workshop on Theoretical Foundations and Applications of Deep Generative Models , year=

  6. [14]

    Advances in Neural Information Processing Systems , pages=

    Spatial transformer networks , author=. Advances in Neural Information Processing Systems , pages=

  7. [15]

    International Conference on Machine Learning , pages=

    Variational inference with normalizing flows , author=. International Conference on Machine Learning , pages=. 2015 , organization=

    2015

  8. [16]

    European Conference on Computer Vision , pages=

    Nerf: Representing scenes as neural radiance fields for view synthesis , author=. European Conference on Computer Vision , pages=. 2020 , organization=

    2020

  9. [17]

    Journal of Machine Learning Research , volume=

    Learning Green's functions associated with time-dependent partial differential equations , author=. Journal of Machine Learning Research , volume=

  10. [18]

    Advances in Neural Information Processing Systems , volume=

    Implicit neural representations with periodic activation functions , author=. Advances in Neural Information Processing Systems , volume=

  11. [19]

    Deep Learning , author=

  12. [20]

    Science , volume=

    Reducing the dimensionality of data with neural networks , author=. Science , volume=. 2006 , publisher=

    2006

  13. [21]

    Scaling Learning Algorithms Towards

    Bengio, Yoshua and LeCun, Yann , booktitle =. Scaling Learning Algorithms Towards

  14. [22]

    2015 , publisher=

    Hierarchical Matrices: Algorithms and Analysis , author=. 2015 , publisher=

    2015

  15. [23]

    Applied and Computational Harmonic Analysis , volume=

    Approximation by exponential sums revisited , author=. Applied and Computational Harmonic Analysis , volume=. 2010 , publisher=

    2010

  16. [24]

    Journal of the ACM , volume=

    Iterative procedures for nonlinear integral equations , author=. Journal of the ACM , volume=

  17. [25]

    SIAM Journal on Numerical Analysis , volume=

    Anderson acceleration for fixed-point iterations , author=. SIAM Journal on Numerical Analysis , volume=

  18. [26]

    SIAM Journal on Numerical Analysis , volume=

    Convergence analysis for Anderson acceleration , author=. SIAM Journal on Numerical Analysis , volume=

  19. [27]

    Low-rank

    Hackbusch, Wolfgang and Khoromskij, Boris N , journal=. Low-rank. 2006 , publisher=

    2006

  20. [28]

    Chemometrics and Intelligent Laboratory Systems , volume=

    Tensor-structured numerical methods in scientific computing: Survey on recent advances , author=. Chemometrics and Intelligent Laboratory Systems , volume=. 2012 , publisher=

    2012

  21. [29]

    Learning deep architectures for

    Bengio, Yoshua , journal=. Learning deep architectures for. 2009 , publisher=

    2009

  22. [30]

    1995 , publisher=

    The Quantum Theory of Fields , author=. 1995 , publisher=

    1995

  23. [31]

    SIAM Journal on Numerical Analysis , volume=

    The differentiation of pseudo-inverses and nonlinear least squares problems whose variables separate , author=. SIAM Journal on Numerical Analysis , volume=. 1973 , publisher=

    1973

  24. [32]

    NeurIPS , year=

    Learning shape correspondence with anisotropic convolutional neural networks , author=. NeurIPS , year=

  25. [33]

    Chamberlain, Benjamin and Rowbottom, James and Gorinova, Maria and Bronstein, Michael and Webb, Stefan and Rossi, Emanuele , booktitle=

  26. [34]

    International Conference on Learning Representations , year=

    The Intrinsic Dimension of Images and Its Impact on Learning , author=. International Conference on Learning Representations , year=

  27. [35]

    Advances in Neural Information Processing Systems , year=

    Intrinsic dimension of data representations in deep neural networks , author=. Advances in Neural Information Processing Systems , year=

  28. [36]

    Proceedings of the National Academy of Sciences , volume=

    The training process of many deep networks explores the same low-dimensional manifold , author=. Proceedings of the National Academy of Sciences , volume=. 2024 , publisher=

    2024

  29. [37]

    International Conference on Machine Learning , pages=

    Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks , author=. International Conference on Machine Learning , pages=. 2017 , organization=

    2017

  30. [38]

    International Conference on Learning Representations , year=

    Neural Eigenfunctions Are Structured Representation Learners , author=. International Conference on Learning Representations , year=

  31. [39]

    2004 , publisher=

    Scattered Data Approximation , author=. 2004 , publisher=

    2004

  32. [40]

    Meshfree Approximation Methods with

    Fasshauer, Gregory E , year=. Meshfree Approximation Methods with

Pith tools

Reviewed July 9, 2026 · model on record in the stance chip above.