Pith. sign in

REVIEW 3 major objections 5 minor 44 references

Neumann eigenmaps for landmark embedding

T0 review · 3 major / 5 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 NeuMaps are a landmark embedding whose Euclidean distances equal the reflecting walk's diffusion distance.

desk verdict The isometry theorem that carries the paper's main claim is wrong — the discarded first left eigenvector is not constant and the appendix misuses V^T D^{-1} V = I — but the underlying NeuMap construction and Nyström observation are worth a referee's time. read the letter →

arxiv 2502.06689 v1 pith:QGPK3SA3 submitted 2025-02-10 math.ST cs.LGcs.NAmath.NAstat.MLstat.TH

Shashank Sule , Wojciech Czaja This is my paper · ORCID
classification math.STcs.LGcs.NAmath.NAstat.MLstat.TH MSC 05C5060J10
keywords NeumanneigenmapslandmarkembeddingdiffusionmapsreflectingrandomwalkLaplacianNyströmextensionmanifoldlearningspectralgraphtheory
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

Neumann eigenmaps (NeuMaps) are a landmark-based manifold-embedding method. The paper's central claim is that, when a subset of data points is treated as a subgraph, the Neumann eigenvectors of that subgraph give coordinates whose Euclidean distances reproduce the diffusion distance of a reflecting random walk on the subgraph. If the claim holds, a small landmark set can be embedded with the same geometric content as a full diffusion map while the discarded points contribute through reflections at the boundary. This matters because landmarking is the standard way to scale spectral embedding to large datasets, and NeuMap adds two properties: the removed points inform the embedding instead of being ignored, and out-of-sample points receive coordinates through a Nyström-type formula.

What carries the argument

The central object is the renormalized Neumann Laplacian $L_S^N = L_S^D - B_S^T (T_S^\delta)^{-1} B_S$, which combines the Dirichlet Laplacian on the landmark subgraph with a boundary term that enforces a vanishing discrete normal derivative at the removed vertices. The paper shows that $T_S^{-1/2}(I - L_S^N)T_S^{-1/2}$ is similar to a row-stochastic matrix $R$, the reflecting random walk transition matrix. The identity doing the work is the isometry formula in Theorem III.3, which equates the squared NeuMap distance with a degree-weighted sum of squared differences of reflecting-walk transition probabilities; the Neumann boundary condition also gives the closed-form Nyström-type extension used for out-of-sample points.

What would settle it

Compute both sides of the identity in Theorem III.3 for a small non-regular graph, for instance a path of three landmarks with one boundary vertex attached to the middle landmark. If the squared NeuMap distance does not equal $\sum_j (p_{ji}^t-p_{jk}^t)^2/d_j$, the claimed isometry fails.

Watch

Extended reading notes

Core claim

The paper defines the Neumann Laplacian of a landmark subgraph $S$ as $L_S^N = L_S^D - B_S^T (T_S^\delta)^{-1} B_S$, where $B_S$ records edges from $S$ to the removed vertices $\delta S$ and $T_S^\delta$ is their degree matrix. Its normalized eigenvectors define a row-stochastic transition matrix $R$ for the reflecting random walk on $S$, in which a walker moves from one landmark to another either directly or through a reflection off a removed vertex. The Neumann map $g_t^d(x_i)$ lists the scaled $t$-step coordinates of the left eigenvectors of $R$ at vertex $i$, starting from the second eigenvector. Theorem III.3 claims that, with $p_{ji}^t$ the reflecting-walk transition probabilities, the identity $\sum_j (p_{ji}^t - p_{jk}^t)^2/d_j = \|g_t(x_i)-g_t(x_k)\|^2$ holds, so the embedding is isometric for the reflecting-walk diffusion distance. The same boundary condition provides an extension formula to removed points that is a multiple of the Nyström extension.

Load-bearing premise

The load-bearing premise is that throwing away the first Neumann eigenvector, as the method does, removes only a constant coordinate; with the paper's left-eigenvector normalization that coordinate is proportional to the vertex degree, which is constant only when all landmark vertices have equal degree.

Editorial extensions

If this is right

  • NeuMap coordinates recover the diffusion distance of the reflecting random walk on the landmark subgraph, so the embedding is isometric in that probability metric.
  • The eigendecomposition is performed only on the landmark subgraph, giving a computational speedup that grows as the landmark set shrinks.
  • The Neumann boundary condition yields a Nyström-type extension, so removed data points receive coordinates without a second eigendecomposition.
  • In the digit example, NeuMap achieves higher cluster accuracy and normalized mutual information than the Roseland baseline on the same landmarks.
  • In the butane example, the first NeuMap eigenfunction tracks the dihedral angle with lower standard error than diffusion maps and stays stable when metastable-state or uniformly subsampled landmarks are removed.

Reading between the lines

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

  • One testable extension is to include the discarded first coordinate explicitly: with the paper's left-eigenvector normalization that coordinate is degree-proportional, so adding it back may restore exact isometry on non-regular graphs.
  • The reflecting-walk interpretation suggests that NeuMap's stability should be largest when removed points are high-degree or metastable points, precisely the points whose removal distorts a plain diffusion map; varying the landmark-selection rule would test this.
  • The same renormalized Neumann construction could be applied with interior operators other than the Dirichlet Laplacian, defining reflecting-walk distances for alternative dynamics such as Schrödinger or transport operators.
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

3 major / 5 minor

Summary. The paper proposes NeuMaps, a landmark-based embedding computed from the eigenvectors of a renormalized Neumann Laplacian on an induced subgraph. The central theoretical claim is Theorem III.3: the Euclidean distance between NeuMap coordinates g_t(x_i) and g_t(x_k) equals the weighted sum of reflecting-random-walk transition probabilities, sum_j (p_{j,i} - p_{j,k})^2 / d_j. The paper also presents numerical experiments on digit classification and molecular dynamics, comparing NeuMaps with Roseland and standard diffusion maps.

Significance. If the isometry of Theorem III.3 were correct, NeuMaps would provide a principled landmark embedding that provably recovers a diffusion distance while retaining the computational advantages of subsampling. The paper also connects the Neumann boundary condition to Nyström extension and includes a code repository. However, the main theoretical result is invalid as stated: the embedding uses left eigenvectors of the transition matrix, and the claimed isometry fails already at t=0 on non-regular graphs. The numerical comparisons are therefore not backed by the asserted geometric guarantee, and the contribution currently lacks a demonstrated theoretical foundation.

major comments (3)
  1. [Section III, Definition III.3 and Theorem III.3] Theorem III.3 is false for the embedding defined in Definition III.3. The coordinates are g_t(x_i) built from y_l = T^{1/2}u_l, which are the left eigenvectors of the reflecting random walk matrix R. The first left eigenvector is proportional to the degree vector d, not to the constant vector. Consequently, dropping the first eigenvector removes a nonconstant component, contrary to standard diffusion-map practice. For t=0 the left side of (III.4) is 1/d_i + 1/d_k, while the right side equals d_i + d_k - (y_1(i) - y_1(k))^2, and (y_1(i) - y_1(k))^2 is generally nonzero when d_i ≠ d_k. The identity already fails on a three-vertex path, so the error is structural.
  2. [Appendix D, proof of Theorem III.3] The algebraic proof contains a load-bearing error. With V = T^{-1/2}U and R^t = T^{-1/2}U Σ^t U^T T^{1/2}, the expression (R^t)^T T^{-1} R^t equals T^{1/2} U Σ^t U^T T^{-2} U Σ^t U^T T^{1/2}, not the simplified form obtained in the displayed calculation. The underbrace T^{1/2}T^{-1}T^{1/2}=I is applied to the wrong factors; the actual combination is U^T T^{-2} U, which is not the identity. This invalid cancellation is essential to the claimed isometry.
  3. [Section IV, numerical experiments] The numerical experiments are run with the embedding of Definition III.3, which does not satisfy the isometry asserted in Theorem III.3. As a result, the reported NMI, ACC, and standard-error numbers do not validate the theoretical claim. Regenerating the experiments with the corrected embedding (right eigenvectors v_l = T^{-1/2}u_l) is necessary before the experimental claims can be assessed, since the algorithm itself would change.
minor comments (5)
  1. [Section IV, paragraph 2, digit classification] The word 'conisder' in the discussion of the subsampling-uniformly-in-time scheme is a typo; it should read 'consider'.
  2. [General, figures] Figures 1 and 2 are referenced in the text but are not included in the provided manuscript version; please ensure they are present in the final submission.
  3. [Appendix D, proof notation] The proof of Theorem III.3 introduces symbols C, B, W, and Ψ without definition; the notation should be aligned with the main text (U, V, Y, Σ) to make the argument readable.
  4. [References] References [2] and [5] appear to refer to the same paper (Chung and Yau, 1997, 'Eigenvalue inequalities for graphs and convex subgraphs'); if so, the duplicate should be removed or the citations updated.
  5. [Section III, Remark III.1] The sentence contrasting the reflecting random walk with 'the one in Roseland' would benefit from an explicit citation to the Roseland algorithm, as the current reference list does not clearly identify it.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the NeuMap isometry is a designed spectral identity, and the numerical benchmarks are external; self-citations are not load-bearing.

full rationale

The central theoretical claim, Theorem III.3, asserts an isometry between the Euclidean distance of NeuMap coordinates and a weighted sum of reflecting-random-walk transition probabilities. By construction, NeuMap coordinates are defined in Definition III.3 as the left-eigenvector coordinates sigma_j^t y_j^T(x_i) of the transition matrix R = T^{-1/2}(I-N)T^{1/2} = V Sigma Y^T. The diffusion distance on the left side of (III.4) is also defined from the same matrix R via the transition probabilities p^t_{ji}. The theorem then expands both sides in the spectral decomposition R^t = V Sigma^t Y^T and identifies the Euclidean norm of the coordinate differences with the weighted transition-probability distance. This is the standard diffusion-map mechanism: the embedding is deliberately chosen from the eigenfunctions of the Markov operator, so the equality is a mathematical derivation rather than an empirical prediction or a fit. No data-dependent parameters are fitted to make the identity hold, and no labels or external quantities are used in the derivation. The numerical experiments are evaluated against external benchmarks (UCI digits and butane molecular-dynamics data) and against Roseland and Dmap, so the demonstrations do not reduce to the theory's own inputs. The paper's self-citations (e.g., [7], [23], [24], [45]) appear in remarks about alternative operators and Nystrom extension; none is load-bearing for the isometry result. A possible algebraic error in Appendix D or the non-constant first left eigenvector would be a correctness issue, not circularity: the theorem is not true merely by definition, and the paper does not define the diffusion distance in terms of the NeuMap coordinates. Therefore, no circular step is exhibited, and the circularity score is minimal apart from the presence of non-essential self-citations.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

No new physical entities are introduced. The method recombines known graph operators (Dirichlet Laplacian, boundary operator, Schur complement) into a renormalized Neumann Laplacian; the main user-chosen parameters are bandwidth, dimension, and cluster count.

free parameters (3)
  • Gaussian kernel bandwidth epsilon = chosen via max-min criterion in datafold [25]
    Controls graph affinity construction; used for both Roseland and NeuMap in all experiments.
  • Embedding dimension d = 2 for digits, 1 for MD
    Number of Neumann eigenvectors retained; set by the user for the task.
  • Number of clusters k = 6 or 7 (text inconsistent)
    k-means cluster count in digit classification; text says digits 0 to 6 but describes 6 clusters.
assumptions (4)
  • standard math Neumann spectral theory for subgraphs
    Definitions II.3-II.6 and Theorem II.1 rely on Chung's spectral graph theory results.
  • domain assumption Gaussian affinity graph approximates data manifold
    Standard diffusion-map assumption, used to build G from samples.
  • domain assumption Fokker-Planck eigenfunctions are optimal collective variables
    Basis for using NeuMap in MD; cited from [43],[44].
  • standard math Discrete normal derivative extension equals Nystrom extension
    Remark III.2 derives formula (III.5); assumes boundary vertices connect only to S and boundary variation is minimized.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Neumann eigenmaps for landmark embedding." pith.science (2026). https://pith.science/paper/QGPK3SA3

@misc{pith2026250206689,
  author       = {Pith},
  title        = {Pith review of: Neumann eigenmaps for landmark embedding},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QGPK3SA3}},
  note         = {Machine review of arXiv:2502.06689}
}
read the original abstract

We present Neumann eigenmaps (NeuMaps), a novel approach for enhancing the standard diffusion map embedding using landmarks, i.e distinguished samples within the dataset. By interpreting these landmarks as a subgraph of the larger data graph, NeuMaps are obtained via the eigendecomposition of a renormalized Neumann Laplacian. We show that NeuMaps offer two key advantages: (1) they provide a computationally efficient embedding that accurately recovers the diffusion distance associated with the reflecting random walk on the subgraph, and (2) they naturally incorporate the Nystr\"om extension within the diffusion map framework through the discrete Neumann boundary condition. Through examples in digit classification and molecular dynamics, we demonstrate that NeuMaps not only improve upon existing landmark-based embedding methods but also enhance the stability of diffusion map embeddings to the removal of highly significant points.

Figures

Figures reproduced from arXiv: 2502.06689 by the authors.

Figure 1
Figure 1. Left: Embedding UCI digits with 25% random landmarking via Roseland. Right: Embedding UCI digits with Neumann [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Top to bottom: FP eigenfunction embeddings of [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

44 extracted references · 43 canonical work pages

  1. [1]

    Spectral graph theory

    Chung, Fan RK. Spectral graph theory. V ol. 92. American Mathematical Soc., 1997

    work page 1997

  2. [3]

    and Yau, S.T., 1996

    Chung, F.R.K., Graham, R.L. and Yau, S.T., 1996. On sampling with Markov chains. Random Structures & Algorithms, 9(1-2), pp.55-77

    work page 1996

  3. [4]

    and Yau, S.T., 1994

    Chung, F.R. and Yau, S.T., 1994. A Harnack inequality for homogeneous graphs and subgraphs. Communications in Analysis and Geometry, 2(4), pp.627-640

    work page 1994

  4. [5]

    and Yau, S.T., 1997

    Chung, F.R. and Yau, S.T., 1997. Eigenvalue inequalities for graphs and convex subgraphs. Communications in Analysis and Geometry, 5(4), pp.575-623

    work page 1997

  5. [6]

    Eigenvalue comparison theorems of Neumann laplacian for graphs

    Tan, J., 1999. Eigenvalue comparison theorems of Neumann laplacian for graphs. Interdisciplinary information sciences, 5(2), pp.157-159

    work page 1999

  6. [7]

    and Ehler, M., 2012

    Czaja, W. and Ehler, M., 2012. Schroedinger eigenmaps for the analysis of biomedical data. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35(5), pp.1274-1280

    work page 2012

  7. [8]

    and Lafon, S., 2006

    Coifman, R.R. and Lafon, S., 2006. Diffusion maps. Applied and computational harmonic analysis, 21(1), pp.5-30

    work page 2006

  8. [9]

    and Tenenbaum, J.B., 2004

    De Silva, V . and Tenenbaum, J.B., 2004. Sparse multidimensional scaling using landmark points (V ol. 120). technical report, Stanford University

    work page 2004

Show all 44 references
  1. [10]

    and Tenenbaum, J., 2002

    Silva, V . and Tenenbaum, J., 2002. Global versus local methods in nonlinear dimensionality reduction. Advances in neural information processing systems, 15

    2002

  2. [11]

    and Sheets, H.D., 2010

    Webster, M.A.R.K. and Sheets, H.D., 2010. A practical introduction to landmark-based geometric morphometrics. The paleontological society papers, 16, pp.163-188

    2010

  3. [12]

    and Carreira-Perpin ´an, M

    Vladymyrov, M. and Carreira-Perpin ´an, M. ´A., 2013. Locally linear landmarks for large-scale manifold learning. In Machine Learning and Knowledge Discovery in Databases: European Conference, ECML PKDD 2013, Prague, Czech Republic, September 23-27, 2013, Proceed- ings, Part I...

    2013

  4. [13]

    and Masuda, N., 2024

    Thongprayoon, C. and Masuda, N., 2024. Online landmark replacement for out-of-sample dimensionality reduction methods. Proceedings of the Royal Society A, 480(2300), p.20230966

    2024

  5. [14]

    and Ferguson, A.L., 2019

    Long, A.W. and Ferguson, A.L., 2019. Landmark diffusion maps (L- dMaps): Accelerated manifold learning out-of-sample extension. Ap- plied and Computational Harmonic Analysis, 47(1), pp.190-211

    2019

  6. [15]

    and Kimmel, R., 2019, January

    Pai, G., Talmon, R., Bronstein, A. and Kimmel, R., 2019, January. Di- mal: Deep isometric manifold learning using sparse geodesic sampling. In 2019 IEEE Winter Conference on Applications of Computer Vision (W ACV) (pp. 819-828). IEEE

    2019

  7. [16]

    and Tiwary, P., 2023

    Evans, L., Cameron, M.K. and Tiwary, P., 2023. Computing committors in collective variables via Mahalanobis diffusion maps. Applied and Computational Harmonic Analysis, 64, pp.62-101

    2023

  8. [17]

    and Leli `evre, T., 2020

    Trstanova, Z., Leimkuhler, B. and Leli `evre, T., 2020. Local and global perspectives on diffusion maps in the analysis of molecular systems. Proceedings of the Royal Society A, 476(2233), p.20190036

    2020

  9. [18]

    and Cameron, M., 2023

    Sule, S., Evans, L. and Cameron, M., 2023. Sharp error estimates for target measure diffusion maps with applications to the committor problem. arXiv preprint arXiv:2312.14418

    2023 arXiv

  10. [19]

    Diffusion maps for group-invariant manifolds

    Hoyos P, Kileel J. Diffusion maps for group-invariant manifolds. arXiv preprint arXiv:2303.16169. 2023 Mar 28

    2023 arXiv

  11. [20]

    and Tsui, M.P., 2024

    Yeh, S.Y ., Wu, H.T., Talmon, R. and Tsui, M.P., 2024. Landmark Alternating Diffusion. arXiv preprint arXiv:2404.19649

    2024 arXiv

  12. [21]

    and Wu, H.T., 2022

    Shen, C. and Wu, H.T., 2022. Scalability and robustness of spectral em- bedding: landmark diffusion is all you need. Information and Inference: A Journal of the IMA, 11(4), pp.1527-1595

    2022

  13. [22]

    and Wu, H.T., 2022

    Shen, C., Lin, Y .T. and Wu, H.T., 2022. Robust and scalable manifold learning via landmark diffusion for long-term medical signal processing. Journal of Machine Learning Research, 23(86), pp.1-30

    2022

  14. [23]

    and Messinger, D.W., 2014, June

    Cahill, N.D., Czaja, W. and Messinger, D.W., 2014, June. Schroedinger eigenmaps with nondiagonal potentials for spatial-spectral clustering of hyperspectral imagery. In Algorithms and technologies for multispectral, hyperspectral, and ultraspectral imagery XX (V ol. 9088, pp. ...

    2014

  15. [24]

    and Ndjakou Njeunje, F.O., 2021

    Czaja, W., Dong, D., Jabin, P.E. and Ndjakou Njeunje, F.O., 2021. Transport model for feature extraction. SIAM Journal on Mathematics of Data Science, 3(1), pp.321-341

    2021

  16. [25]

    datafold: data-driven models for point clouds and time series on manifolds

    Lehmberg et al., (2020). datafold: data-driven models for point clouds and time series on manifolds. Journal of Open Source Software, 5(51), 2283, https://doi.org/10.21105/joss.02283

    2020 doi

  17. [26]

    and Parrinello, M., 2023

    Bonati, L., Trizio, E., Rizzi, A. and Parrinello, M., 2023. A unified framework for machine learning collective variables for enhanced sam- pling simulations: mlcolvar. The Journal of Chemical Physics, 159(1)

    2023

  18. [27]

    and Lelievre, T., 2010

    Legoll, F. and Lelievre, T., 2010. Effective dynamics using conditional expectations. Nonlinearity, 23(9), p.2131

    2010

  19. [28]

    and Olla, S., 2017

    Legoll, F., Lelievre, T. and Olla, S., 2017. Pathwise estimates for an effective dynamics. Stochastic Processes and their Applications, 127(9), pp.2841-2863

    2017

  20. [29]

    and Lelievre, T., 2011, August

    Legoll, F. and Lelievre, T., 2011, August. Some remarks on free energy and coarse-graining. In Numerical Analysis of Multiscale Computations: Proceedings of a Winter Workshop at the Banff International Research Station 2009 (pp. 279-329). Berlin, Heidelberg: Springer Berlin He...

    2011

  21. [30]

    and Pietrucci, F., 2022

    Palacio-Rodriguez, K. and Pietrucci, F., 2022. Free energy landscapes, diffusion coefficients, and kinetic rates from transition paths. Journal of chemical theory and computation, 18(8), pp.4639-4648

    2022

  22. [31]

    and Pietrucci, F., 2023

    Mouaffac, L., Palacio-Rodriguez, K. and Pietrucci, F., 2023. Optimal reaction coordinates and kinetic rates from the projected dynamics of transition paths. Journal of Chemical Theory and Computation, 19(17), pp.5701-5711

    2023

  23. [32]

    and Gao, Y .Q., 2019

    Yang, Y .I., Shao, Q., Zhang, J., Yang, L. and Gao, Y .Q., 2019. Enhanced sampling in molecular dynamics. The Journal of chemical physics, 151(7)

    2019

  24. [33]

    and Yang, L., 2006

    Gao, Y .Q. and Yang, L., 2006. On the enhanced sampling over energy barriers in molecular dynamics simulations. The Journal of chemical physics, 125(11)

    2006

  25. [34]

    and Samaey, G., 2020

    Legoll, F., Leli `evre, T., Myerscough, K. and Samaey, G., 2020. Parareal computation of stochastic differential equations with time-scale separa- tion: a numerical convergence study. Computing and Visualization in Science, 23, pp.1-18

    2020

  26. [35]

    and Coifman, R.R., 2009

    Singer, A., Erban, R., Kevrekidis, I.G. and Coifman, R.R., 2009. Detecting intrinsic slow variables in stochastic dynamical systems by anisotropic diffusion maps. Proceedings of the National Academy of Sciences, 106(38), pp.16090-16095

    2009

  27. [36]

    and Kozinsky, B., 2022

    Batzner, S., Musaelian, A., Sun, L., Geiger, M., Mailoa, J.P., Kornbluth, M., Molinari, N., Smidt, T.E. and Kozinsky, B., 2022. E (3)-equivariant graph neural networks for data-efficient and accurate interatomic poten- tials. Nature communications, 13(1), p.2453

    2022

  28. [37]

    and Tiwary, P., 2018

    Ribeiro, J.M.L., Bravo, P., Wang, Y . and Tiwary, P., 2018. Reweighted autoencoded variational Bayes for enhanced sampling (RA VE). The Journal of chemical physics, 149(7)

    2018

  29. [38]

    and Tiwary, P., 2023

    Vani, B.P., Aranganathan, A., Wang, D. and Tiwary, P., 2023. Alphafold2-rave: From sequence to boltzmann ranking. Journal of chemical theory and computation, 19(14), pp.4351-4354

    2023

  30. [39]

    and Tiwary, P., 2024

    Wang, D., Wang, Y ., Evans, L. and Tiwary, P., 2024. From latent dynamics to meaningful representations. Journal of Chemical Theory and Computation, 20(9), pp.3503-3513

    2024

  31. [40]

    and Stoltz, G., 2021

    Belkacemi, Z., Gkeka, P., Leli `evre, T. and Stoltz, G., 2021. Chasing collective variables using autoencoders and biased trajectories. Journal of chemical theory and computation, 18(1), pp.59-78

    2021

  32. [41]

    and Zhang, W., 2024

    Leli `evre, T., Pigeon, T., Stoltz, G. and Zhang, W., 2024. Analyzing multimodal probability measures with autoencoders. The Journal of Physical Chemistry B, 128(11), pp.2607-2631

    2024

  33. [42]

    and Tuckerman, M.E., 2019

    Rogal, J., Schneider, E. and Tuckerman, M.E., 2019. Neural-network- based path collective variables for enhanced sampling of phase trans- formations. Physical Review Letters, 123(24), p.245701

    2019

  34. [43]

    and Kevrekidis, I.G., 2006

    Nadler, B., Lafon, S., Coifman, R.R. and Kevrekidis, I.G., 2006. Dif- fusion maps, spectral clustering and reaction coordinates of dynamical systems. Applied and Computational Harmonic Analysis, 21(1), pp.113- 127

    2006

  35. [44]

    and Nadler, B., 2008

    Coifman, R.R., Kevrekidis, I.G., Lafon, S., Maggioni, M. and Nadler, B., 2008. Diffusion maps, reduction coordinates, and low dimensional representation of stochastic systems. Multiscale Modeling & Simulation, 7(2), pp.842-864

    2008

  36. [45]

    and Halevy, A

    Czaja, W., Doster, T. and Halevy, A. ”An overview of numerical acceleration techniques for nonlinear dimension reduction.” Recent Applications of Harmonic Analysis to Function Spaces, Differential Equations, and Data Science: Novel Methods in Harmonic Analysis, V olume 2 (2017...

    2017

Pith tools

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