pith. machine review for the scientific record. sign in

arxiv: 2604.02535 · v1 · submitted 2026-04-02 · 💻 cs.LG · cs.HC

Recognition: 2 theorem links

· Lean Theorem

A Spectral Framework for Multi-Scale Nonlinear Dimensionality Reduction

Angelos Chatzimparmpas, Takanori Fujiwara, Thomas H\"ollt, Zeyang Huang

Pith reviewed 2026-05-13 21:14 UTC · model grok-4.3

classification 💻 cs.LG cs.HC
keywords dimensionality reductionspectral methodsmulti-scale embeddingsgraph frequency analysisnonlinear DRcross-entropy optimizationmanifold continuityvisual analytics
0
0 comments X

The pith

A spectral framework pairs linear decomposition with cross-entropy optimization to create multi-scale nonlinear dimensionality reduction embeddings that preserve both global and local structure.

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

The paper introduces a framework that starts from a linear spectral decomposition of the data graph to capture global geometry and then refines the embedding with cross-entropy optimization to add local neighborhood separation. This combination is meant to resolve the longstanding tension where local-focused methods distort overall manifold shape and global-focused methods lose fine detail. Because the starting point remains linear and explicit, the approach also lets users trace how individual spectral modes shape the final layout through a graph-frequency lens, and it adds glyph overlays on scatterplots for visual inspection. If the method works as described, it would give practitioners embeddings that are both more faithful across scales and more open to direct analysis than current nonlinear techniques.

Core claim

The paper claims that high-dimensional data can be embedded via a spectral basis derived from linear decomposition, then adjusted through cross-entropy optimization, to produce representations that simultaneously maintain global manifold geometry and local neighborhood separation. The same linear basis supports post-hoc examination of the embedding by decomposing it into contributions from different graph-frequency modes. Glyph-based augmentations on the resulting scatterplots further allow visual tracing of those mode influences.

What carries the argument

The spectral basis obtained from linear spectral decomposition of the input graph, which supplies an explicit multi-scale foundation that cross-entropy optimization then refines while preserving analytical access to frequency-mode contributions.

If this is right

  • The embedding process becomes inspectable by isolating the contribution of each spectral mode to the final coordinates.
  • Scatterplot visualizations can be augmented with glyphs that directly encode those mode contributions for qualitative checks.
  • Manifold continuity improves because the spectral starting point anchors global shape before local refinement occurs.
  • The same pipeline supports quantitative comparisons that track how changes in spectral truncation affect both scale levels.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The framework might be tested on time-varying or streaming data by updating the spectral basis incrementally rather than recomputing from scratch.
  • One could replace cross-entropy with other contrastive losses and measure whether the graph-frequency analysis remains equally informative.
  • Integration with existing visualization software could let users interactively suppress or amplify specific frequency bands to explore embedding sensitivity.

Load-bearing premise

That a linear spectral decomposition can be merged with cross-entropy optimization to improve both global and local fidelity without creating new uncontrolled distortions in the embedding.

What would settle it

Apply the framework to a standard test manifold such as the Swiss roll or a hierarchical cluster dataset, then measure global structure error against Laplacian Eigenmaps and local neighborhood error against t-SNE; if the new embeddings score worse on both metrics than the specialized baselines, the central claim is falsified.

Figures

Figures reproduced from arXiv: 2604.02535 by Angelos Chatzimparmpas, Takanori Fujiwara, Thomas H\"ollt, Zeyang Huang.

Figure 1
Figure 1. Figure 1: Comparison of data structure, UMAP [47], and our spectral decomposition snapshots on three datasets. As the number of used spectral modes S increases, low-frequency snapshots recover coarse organization first and then add finer detail toward the full-spectrum embedding. analyzed through contrastive feature analysis [23] or supervised models such as decision trees [5] and explainable boosting machines [55].… view at source ↗
Figure 2
Figure 2. Figure 2: Petal glyph. The outline length of each petal rep￾resents |un,s|. The filled length repre￾sents ∥un,s ·ps∥. Comparing these two lengths reveals how cross-entropy op￾timization amplifies or suppresses in￾dividual spectral modes. We further encode this change using color: red indicates that a mode is emphasized, while blue indicates that it is suppressed. Color intensity reflects the magnitude of this change… view at source ↗
Figure 3
Figure 3. Figure 3: Reconstruction error across datasets as subspace [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Qualitative comparison of embeddings across three representative datasets as [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Spectral response at the final stage with full spectral modes for [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: (a), already at S = 8, amphid sensory classes such as AFD, ASH, [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Local spectral explanation of the embedding for MNIST handwrit [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
read the original abstract

Dimensionality reduction (DR) is characterized by two longstanding trade-offs. First, there is a global-local preservation tension: methods such as t-SNE and UMAP prioritize local neighborhood preservation, yet may distort global manifold structure, while methods such as Laplacian Eigenmaps preserve global geometry but often yield limited local separation. Second, there is a gap between expressiveness and analytical transparency: many nonlinear DR methods produce embeddings without an explicit connection to the underlying high-dimensional structure, limiting insight into the embedding process. In this paper, we introduce a spectral framework for nonlinear DR that addresses these challenges. Our approach embeds high-dimensional data using a spectral basis combined with cross-entropy optimization, enabling multi-scale representations that bridge global and local structure. Leveraging linear spectral decomposition, the framework further supports analysis of embeddings through a graph-frequency perspective, enabling examination of how spectral modes influence the resulting embedding. We complement this analysis with glyph-based scatterplot augmentations for visual exploration. Quantitative evaluations and case studies demonstrate that our framework improves manifold continuity while enabling deeper analysis of embedding structure through spectral mode contributions.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 1 minor

Summary. The paper introduces a spectral framework for nonlinear dimensionality reduction that combines linear spectral decomposition (via graph Laplacian eigendecomposition) with cross-entropy optimization to produce multi-scale embeddings bridging global and local manifold structure, while enabling graph-frequency analysis of mode contributions and glyph-based visual augmentations.

Significance. If the central claim holds, the framework would meaningfully address the global-local tension in DR methods and the expressiveness-transparency gap by retaining analytical access to spectral modes after nonlinear optimization, offering a principled alternative to purely heuristic nonlinear techniques like t-SNE or UMAP.

major comments (1)
  1. [§3.2 and §4.1] §3.2, Eq. (4) and §4.1: the spectral basis is obtained from eigendecomposition of the graph Laplacian, yet the cross-entropy loss minimized in §4.1 operates directly on pairwise distances without any regularization or projection term that would preserve stable contributions from the original eigenmodes. No derivation is provided showing that post-optimization coordinates remain interpretable in the graph-frequency basis, which directly threatens the multi-scale bridging and frequency-analysis claims.
minor comments (1)
  1. [Abstract] The abstract states that quantitative evaluations demonstrate improved manifold continuity, but no specific metrics, baselines, or ablation results are previewed; adding a brief table reference would improve clarity.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive comments. The major comment identifies a need for explicit derivation regarding post-optimization spectral interpretability. We address this point directly below and outline the planned revision.

read point-by-point responses

  1. Referee: [§3.2 and §4.1] §3.2, Eq. (4) and §4.1: the spectral basis is obtained from eigendecomposition of the graph Laplacian, yet the cross-entropy loss minimized in §4.1 operates directly on pairwise distances without any regularization or projection term that would preserve stable contributions from the original eigenmodes. No derivation is provided showing that post-optimization coordinates remain interpretable in the graph-frequency basis, which directly threatens the multi-scale bridging and frequency-analysis claims.

    Authors: We appreciate the referee highlighting this aspect. The embedding is initialized via the spectral basis from the graph Laplacian eigendecomposition in Eq. (4) of §3.2, after which the cross-entropy loss in §4.1 refines the coordinates for improved local preservation. While the optimization lacks an explicit regularization term, the coordinates remain linear combinations of the original eigenmodes by construction. We acknowledge, however, that the manuscript does not supply a formal derivation of the stability of these frequency contributions after optimization. We will revise the paper by inserting a new subsection after §4.1 that derives the re-projection of optimized coordinates onto the graph-frequency basis and shows that mode contributions remain interpretable under standard manifold smoothness assumptions. The revision will also include a brief empirical check confirming that frequency signatures are preserved. This change directly addresses the concern and strengthens the multi-scale and frequency-analysis claims. revision: yes

Circularity Check

0 steps flagged

No circularity: derivation combines standard spectral basis with cross-entropy optimization without self-referential reduction

full rationale

The paper's core construction starts from eigendecomposition of a graph Laplacian (standard linear spectral step) and then applies cross-entropy optimization to produce the final embedding coordinates. No equation in the provided abstract or described sections shows the optimized coordinates being redefined back into the spectral basis or the optimization loss being fitted to recover the same eigenmodes by construction. The multi-scale bridging claim rests on the explicit combination of these two independent components rather than on any parameter being renamed as a prediction or on a self-citation chain that forbids alternatives. External benchmarks (quantitative evaluations and case studies) are invoked to support performance, keeping the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Review based on abstract only; no explicit free parameters, axioms, or invented entities are detailed in the provided text. The approach appears to rely on standard linear spectral decomposition and cross-entropy loss without introducing new postulated entities.

pith-pipeline@v0.9.0 · 5497 in / 1114 out tokens · 38619 ms · 2026-05-13T21:14:22.295712+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

Reference graph

Works this paper leans on

84 extracted references · 84 canonical work pages · 4 internal anchors

  1. [1]

    Balestriero and Y

    R. Balestriero and Y . LeCun. Contrastive and non-contrastive self-supervised learning recover global and local spectral em- bedding methods.Adv. Neural Inf. Process. Syst., 35:26671– 26685, 2022. https://proceedings.neurips.cc/paper_files/paper/2022/file/ aa56c74513a5e35768a11f4e82dd7ffb-Paper-Conference.pdf. 1, 3, 9

    work page 2022

  2. [2]

    Becht, L

    E. Becht, L. McInnes, J. Healy, C.-A. Dutertre, I. W. Kwok, L. G. Ng et al. Dimensionality reduction for visualizing single-cell data using UMAP. Nat. Biotechnol., 37(1):38–44, 2019. doi: 10.1038/nbt.4314 1, 2, 8

    work page doi:10.1038/nbt.4314 2019

  3. [3]

    Belkin and P

    M. Belkin and P. Niyogi. Laplacian Eigenmaps for dimensionality reduc- tion and data representation.Neural Comput., 15(6):1373–1396, 2003. doi: 10.1162/089976603321780317 1, 2, 3, 6, 7

    work page doi:10.1162/089976603321780317 2003

  4. [4]

    Benson-Putnins, M

    D. Benson-Putnins, M. Bonfardin, M. E. Magnoni, and D. Martin. Spectral clustering and visualization: A novel clustering of Fisher’s iris data set. SIAM Undergraduate Research Online, 4:1–15, 2011. https://www.siam. org/media/s12ln4i2/spectral_clustering_and_visualization.pdf. 3

    work page 2011

  5. [5]

    Bibal, V

    A. Bibal, V . Delchevalerie, and B. Frénay. DT-SNE: t-SNE discrete visualizations as decision tree structures.Neurocomputing, 529:101–112,

    work page

  6. [6]

    doi: 10.1016/j.neucom.2023.01.073 3

    work page doi:10.1016/j.neucom.2023.01.073 2023

  7. [7]

    Bibal, R

    A. Bibal, R. Marion, R. von Sachs, and B. Frénay. BIOT: Explaining multidimensional nonlinear MDS embeddings using the Best Interpretable Orthogonal Transformation.Neurocomputing, 453:109–118, 2021. doi: 10.1016/j.neucom.2021.04.088. 1, 3

    work page doi:10.1016/j.neucom.2021.04.088 2021

  8. [8]

    J. N. Böhm, P. Berens, and D. Kobak. Unsupervised visualization of image datasets using contrastive learning.arXiv:2210.09879, 2022. doi: 10.48550/arXiv.2210.09879 1

    work page doi:10.48550/arxiv.2210.09879 2022

  9. [9]

    J. N. Böhm, P. Berens, and D. Kobak. Attraction-repulsion spectrum in neighbor embeddings.J. Mach. Learn. Res., 23(95):1–32, 2022. https: //jmlr.org/papers/volume23/21-0055/21-0055.pdf. 1, 2

    work page 2022

  10. [10]

    T. T. Cai and R. Ma. Theoretical foundations of t-SNE for visualizing high-dimensional clustered data.J. Mach. Learn. Res., 23(301):1–54,

    work page

  11. [11]

    https://jmlr.org/papers/volume23/21-0524/21-0524.pdf. 3, 8

    work page

  12. [12]

    R. R. Coifman and S. Lafon. Diffusion maps.Appl. Comput. Harmon. Anal., 21(1):5–30, 2006. doi: 10.1016/j.acha.2006.04.006 3

    work page doi:10.1016/j.acha.2006.04.006 2006

  13. [13]

    D. B. Coimbra, R. M. Martins, T. T. Neves, A. C. Telea, and F. V . Paulovich. Explaining three-dimensional dimensionality reduction plots.Inf. Vis., 15(2):154–172, 2016. doi: 10.1177/1473871615600010 2

    work page doi:10.1177/1473871615600010 2016

  14. [14]

    Couplet, P

    E. Couplet, P. Lambert, M. Verleysen, D. Mulders, J. A. Lee, and C. De Bodt. Natively interpretable t-SNE. InProc. ECML PKDD, pp. 107–123. Springer, 2023. doi: 10.1007/978-3-031-74630-7_8 1, 3

    work page doi:10.1007/978-3-031-74630-7_8 2023

  15. [15]

    J. P. Cunningham and Z. Ghahramani. Linear dimensionality re- duction: Survey, insights, and generalizations.J. Mach. Learn. Res., 16(89):2859–2900, 2015. https://jmlr.org/papers/volume16/ cunningham15a/cunningham15a.pdf. 2

    work page 2015

  16. [16]

    Damrich, N

    S. Damrich, N. Böhm, F. A. Hamprecht, and D. Kobak. From t-SNE to UMAP with contrastive learning. InProc. ICLR, 2023. 44 pages. doi: abs/2206.01816 1

    work page arXiv 2023

  17. [17]

    Damrich and F

    S. Damrich and F. A. Hamprecht. On UMAP’s true loss function.Adv. Neural Inf. Process. Syst., 34:5798–5809, 2021. https://proceedings.nips. cc/paper/2021/file/2de5d16682c3c35007e4e92982f1a2ba-Paper.pdf. 1, 4, 9

    work page 2021

  18. [18]

    de Bodt, A

    C. de Bodt, A. Diaz-Papkovich, M. Bleher, K. Bunte, C. Coupette, S. Dam- rich et al. Low-dimensional embeddings of high-dimensional data, Aug

    work page

  19. [19]

    doi: 10.48550/arXiv.2508.15929 2, 3

    work page doi:10.48550/arxiv.2508.15929

  20. [20]

    Defferrard, X

    M. Defferrard, X. Bresson, and P. Vandergheynst. Convolutional neural networks on graphs with fast localized spectral filtering.Adv. Neural Inf. Process. Syst., 29, 2016. https://proceedings.neurips.cc/paper/2016/file/ 04df4d434d481c5bb723be1b6df1ee65-Paper.pdf. 1, 3, 9

    work page 2016

  21. [21]

    L. Deng. The MNIST database of handwritten digit images for machine learning research.IEEE Signal Process. Mag., 29(6):141–142, 2012. doi: 10.1109/MSP.2012.2211477 4, 5

    work page doi:10.1109/msp.2012.2211477 2012

  22. [22]

    X. Dong, D. Thanou, P. Frossard, and P. Vandergheynst. Learning Lapla- cian matrix in smooth graph signal representations.IEEE Trans. Signal Process., 64(23):6160–6173, 2016. doi: 10.1109/TSP.2016.2602809 6

    work page doi:10.1109/tsp.2016.2602809 2016

  23. [23]

    Y . Dong, P. Soga, Y . He, S. Wang, and J. Li. Graph neural networks are more than filters: Revisiting and benchmarking from a spectral perspective. InProc. ICLR, 2026. doi: abs/2412.07188 3

    work page arXiv 2026

  24. [24]

    Faust, D

    R. Faust, D. Glickenstein, and C. Scheidegger. DimReader: Axis lines that explain non-linear projections .IEEE Trans. Vis. Comput. Graph., 25(01):481–490, 2019. doi: 10.1109/TVCG.2018.2865194 2

    work page doi:10.1109/tvcg.2018.2865194 2019

  25. [25]

    J. J. Forman, P. A. Clemons, S. L. Schreiber, and S. J. Haggarty. Spectral- NET – an application for spectral graph analysis and visualization.BMC Bioinformatics, 6(1), 2005. doi: 10.1186/1471-2105-6-260 1, 3

    work page doi:10.1186/1471-2105-6-260 2005

  26. [26]

    Fujiwara, O.-H

    T. Fujiwara, O.-H. Kwon, and K.-L. Ma. Supporting analysis of dimen- sionality reduction results with contrastive learning.IEEE Trans. Vis. Comput. Graph., 26(01):45–55, 2020. doi: 10.1109/TVCG.2019.2934251 3

    work page doi:10.1109/tvcg.2019.2934251 2020

  27. [27]

    Fujiwara, X

    T. Fujiwara, X. Wei, J. Zhao, and K.-L. Ma. Interactive dimensionality reduction for comparative analysis.IEEE Trans. Vis. Comput. Graph., 28(1):758–768, 2022. doi: 10.1109/TVCG.2021.3114807 3

    work page doi:10.1109/tvcg.2021.3114807 2022

  28. [28]

    W. Fung, L. Wexler, and M. G. Heiman. Cell-type-specific promoters forC. elegansglia.J. Neurogenet., 34(3-4):335–346, 2020. doi: 10. 1080/01677063.2020.1781851 8

    work page arXiv 2020

  29. [29]

    Gisbrecht, A

    A. Gisbrecht, A. Schulz, and B. Hammer. Parametric nonlinear dimension- ality reduction using kernel t-SNE.Neurocomputing, 147:71–82, 2015. doi: 10.1016/j.neucom.2013.11.045 1

    work page doi:10.1016/j.neucom.2013.11.045 2015

  30. [30]

    X. Guo, W. Xu, W. Zhang, C. Pan, A. E. Thalacker-Mercer, H. Zheng et al. High-frequency and functional mitochondrial DNA mutations at the single-cell level.Proc. Natl. Acad. Sci. U.S.A., 120(1):e2201518120, 2023. doi: 10.1073/pnas.2201518120 6

    work page doi:10.1073/pnas.2201518120 2023

  31. [31]

    D. K. Hammond, P. Vandergheynst, and R. Gribonval. Wavelets on graphs via spectral graph theory.Appl. Comput. Harmon. Anal., 30(2):129–150,

    work page

  32. [32]

    doi: 10.1016/j.acha.2010.04.005 3

    work page doi:10.1016/j.acha.2010.04.005 2010

  33. [33]

    Heulot, M

    N. Heulot, M. Aupetit, and J.-D. Fekete. ProxiLens: Interactive Ex- ploration of High-Dimensional Data using Projections. InProc. VAMP. The Eurographics Association, 2013. 5 pages. doi: 10.2312/PE.V AMP. V AMP2013.011-015 2

    work page doi:10.2312/pe.v 2013

  34. [34]

    Höllt, N

    T. Höllt, N. Pezzotti, V . van Unen, F. Koning, E. Eisemann, B. Lelieveldt et al. Cytosplore: Interactive immune cell phenotyping for large single-cell datasets.Comput. Graph. Forum, 35(3):171–180, 2016. doi: 10.1111/cgf. 12893 8

    work page doi:10.1111/cgf 2016

  35. [36]

    Huguet, A

    G. Huguet, A. Tong, E. De Brouwer, Y . Zhang, G. Wolf, I. Adel- stein et al. A heat diffusion perspective on geodesic preserving di- mensionality reduction.Adv. Neural Inf. Process. Syst., 36:6986– 7016, 2023. https://proceedings.neurips.cc/paper_files/paper/2023/file/ 16063a1c0f0cddd4894585cf44cebb2c-Paper-Conference.pdf. 1, 3, 8, 9

    work page 2023

  36. [37]

    H. Jeon, A. Cho, J. Jang, S. Lee, J. Hyun, H.-K. Ko et al. ZADU: A python library for evaluating the reliability of dimensionality reduction embeddings. InProc. VIS, pp. 196–200, 2023. doi: 10.1109/VIS54172. 2023.00048 7

    work page doi:10.1109/vis54172 2023

  37. [38]

    Jeon, Y .-H

    H. Jeon, Y .-H. Kuo, M. Aupetit, K.-L. Ma, and J. Seo. Classes are not clusters: Improving label-based evaluation of dimensionality reduction. IEEE Trans. Vis. Comput. Graph., 30(1):781–791, 2024. doi: 10.1109/ TVCG.2023.3327187 2, 7

    work page arXiv 2024

  38. [39]

    H. Jeon, H. Lee, Y .-H. Kuo, T. Yang, D. Archambault, S. Ko et al. Un- veiling high-dimensional backstage: A survey for reliable visual analytics with dimensionality reduction. InProc. CHI, art. no. 394, 24 pages. ACM,

    work page

  39. [40]

    doi: 10.1145/3706598.3713551 1, 2, 7

    work page doi:10.1145/3706598.3713551

  40. [41]

    M. Jung, T. Fujiwara, and J. Jo. GhostUMAP2: Measuring and analyzing (r,d)-stability of UMAP.IEEE Trans. Vis. Comput. Graph., 32(1):353– 362, 2026. doi: 10.1109/TVCG.2025.3633894 2, 7

    work page doi:10.1109/tvcg.2025.3633894 2026

  41. [42]

    Kobak and G

    D. Kobak and G. C. Linderman. Initialization is critical for preserving global data structure in both t-SNE and UMAP.Nat Biotechnol, 39(2):156– 157, 2021. doi: 10.1038/s41587-020-00809-z 1, 3, 6, 8

    work page doi:10.1038/s41587-020-00809-z 2021

  42. [43]

    J. B. Kruskal. Multidimensional scaling by optimizing goodness of fit to a nonmetric hypothesis.Psychometrika, 29(1):1–27, 1964. doi: 10. 1007/BF02289565 7

    work page 1964

  43. [44]

    J. A. Lee and M. Verleysen. Quality assessment of dimensionality reduc- tion: Rank-based criteria.Neurocomputing, 72(7–9):1431–1443, 2009. doi: 10.1016/j.neucom.2008.12.017 7

    work page doi:10.1016/j.neucom.2008.12.017 2009

  44. [45]

    R. B. Lehoucq, D. C. Sorensen, and C. Yang.ARPACK Users’ Guide: Solution of Large-Scale Eigenvalue Problems with Implicitly Restarted Arnoldi Methods. SIAM, 1998. doi: 10.1137/1.9780898719628 6

    work page doi:10.1137/1.9780898719628 1998

  45. [46]

    N. Li, V . van Unen, T. Höllt, A. Thompson, J. van Bergen, N. Pezzotti et al. Mass cytometry reveals innate lymphoid cell differentiation pathways in the human fetal intestine.J. Exp. Med., 215(5):1383–1396, 2018. doi: 10.1084/jem.20171934 8

    work page doi:10.1084/jem.20171934 2018

  46. [47]

    D. Liao, C. Liu, B. W. Christensen, A. Tong, G. Huguet, G. Wolf et al. Assessing neural network representations during training using noise- resilient diffusion spectral entropy. InProc. CISS, pp. 1–6. IEEE, 2024. doi: 10.1109/CISS59072.2024.10480166 3

    work page doi:10.1109/ciss59072.2024.10480166 2024

  47. [48]

    Q. Liu, Y . Ren, Z. Zhu, D. Li, X. Ma, and Q. Li. RankAxis: Towards a systematic combination of projection and ranking in multi-attribute data exploration.IEEE Trans. Vis. Comput. Graph., 29(1):701–711, 2023. doi: 10.1109/TVCG.2022.3209463 3

    work page doi:10.1109/tvcg.2022.3209463 2023

  48. [49]

    Machado, M

    A. Machado, M. Behrisch, and A. Telea. Necessary but not suffi- cient: Limitations of projection quality metrics.Comput. Graph. Forum, 44(3):e70101, 2025. doi: 10.1111/cgf.70101 2, 7

    work page doi:10.1111/cgf.70101 2025

  49. [50]

    Ma´ckiewicz and W

    A. Ma´ckiewicz and W. Ratajczak. Principal components analysis (PCA). Comput. Geosci., 19(3):303–342, 1993. 6

    work page 1993

  50. [51]

    T. Manz, F. Lekschas, E. Greene, G. Finak, and N. Gehlenborg. A general framework for comparing embedding visualizations across class-label hierarchies.IEEE Trans. Vis. Comput. Graph., 31(1):283–293, 2025. doi: 10.1109/TVCG.2024.3456370 3

    work page doi:10.1109/tvcg.2024.3456370 2025

  51. [52]

    UMAP: Uniform Manifold Approximation and Projection for Dimension Reduction

    L. McInnes, J. Healy, and J. Melville. UMAP: Uniform manifold approxi- mation and projection for dimension reduction.arXiv:1802.03426, 2018. doi: 10.48550/arXiv.1802.03426 1, 2, 3, 4, 6, 7, 9

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426 2018

  52. [53]

    Montambault, G

    B. Montambault, G. Appleby, J. Rogers, C. D. Brumar, M. Li, and R. Chang. DimBridge: Interactive explanation of visual patterns in di- mensionality reductions with predicate logic.IEEE Trans. Vis. Comput. Graph., 31(1):207–217, 2025. doi: 10.1109/TVCG.2024.3456391 2

    work page doi:10.1109/tvcg.2024.3456391 2025

  53. [54]

    K. R. Moon, D. van Dijk, Z. Wang, S. Gigante, D. B. Burkhardt, W. S. Chen et al. Visualizing structure and transitions in high-dimensional biological data.Nat. Biotechnol., 37(12):1482–1492, Dec. 2019. doi: 10. 1038/s41587-019-0336-3 3, 5, 7, 8, 9

    work page 2019

  54. [55]

    Okami, K

    N. Okami, K. Miyake, N. Sakamoto, J. Nonaka, and T. Fujiwara. Visual analytics using tensor unified linear comparative analysis.IEEE Trans. Vis. Comput. Graph., 32(1):79–89, 2026. doi: 10.1109/TVCG.2025.3633912 3

    work page doi:10.1109/tvcg.2025.3633912 2026

  55. [56]

    J. S. Packer, Q. Zhu, C. Huynh, P. Sivaramakrishnan, E. Preston, H. Dueck et al. A lineage-resolved molecular atlas ofC. elegansembryogenesis at single-cell resolution.Science, 365(6459):eaax1971, 2019. doi: 10. 1126/science.aax1971 5, 8

    work page 2019

  56. [57]

    Rosen, M

    P. Rosen, M. Hajij, and B. Wang. Homology-preserving multi-scale graph skeletonization using mapper on graphs. InProc. TopoInVis, pp. 10–20. IEEE, 2023. doi: 10.1109/TopoInVis60193.2023.00008 3

    work page doi:10.1109/topoinvis60193.2023.00008 2023

  57. [58]

    T. L. Saidi, A. Hickok, and A. J. Blumberg. Recovering manifold structure using Ollivier Ricci curvature. InProc. ICLR, 2025. doi: 10.48550/arXiv. 2410.01149 2, 9

    work page internal anchor Pith review doi:10.48550/arxiv 2025

  58. [59]

    Sainburg, L

    T. Sainburg, L. McInnes, and T. Q. Gentner. Parametric UMAP embed- dings for representation and semisupervised learning.Neural Comput., 33(11):2881–2907, 10 2021. doi: 10.1162/neco_a_01434 1

    work page doi:10.1162/neco_a_01434 2021

  59. [60]

    Salmanian, A

    P. Salmanian, A. Chatzimparmpas, A. C. Karaca, and R. M. Martins. DimVis: Interpreting Visual Clusters in Dimensionality Reduction With Explainable Boosting Machine. InProc. MLVis. The Eurographics Asso- ciation, 2024. doi: 10.2312/mlvis.20241125 3

    work page doi:10.2312/mlvis.20241125 2024

  60. [61]

    Sarikaya and M

    A. Sarikaya and M. Gleicher. Scatterplots: Tasks, data, and designs.IEEE Trans. Vis. Comput. Graph., 24(1):402–412, 2018. doi: 10.1109/TVCG. 2017.2744184 5

    work page doi:10.1109/tvcg 2018

  61. [62]

    D. I. Shuman, S. K. Narang, P. Frossard, A. Ortega, and P. Vandergheynst. The emerging field of signal processing on graphs: Extending high- dimensional data analysis to networks and other irregular domains.IEEE Signal Process. Mag., 30(3):83–98, 2013. 1, 2, 3, 4, 6, 9

    work page 2013

  62. [63]

    S. Sidney. Nonparametric statistics for the behavioral sciences.J. Nerv. Ment. Dis., 125(3), 1957. 7

    work page 1957

  63. [64]

    Sivaramakrishnan, C

    P. Sivaramakrishnan, C. Watkins, and J. I. Murray. Transcript accu- mulation rates in the earlyCaenorhabditis elegansembryo.Sci. Adv., 9(34):eadi1270, 2023. doi: 10.1126/sciadv.adi1270 8

    work page doi:10.1126/sciadv.adi1270 2023

  64. [65]

    Spielman

    D. Spielman. Spectral graph theory. In U. Naumann and O. Schenk, eds.,Combinatorial Scientific Computing, vol. 18. CRC Press Boca Raton, Florida, 2012. doi: 10.1201/b11644 3

    work page doi:10.1201/b11644 2012

  65. [66]

    J. Tang, J. Liu, M. Zhang, and Q. Mei. Visualizing large-scale and high- dimensional data. InProc. WWW, pp. 287–297, 2016. doi: 10.1145/ 2872427.2883041 2

    work page arXiv 2016

  66. [67]

    J. B. Tenenbaum, V . d. Silva, and J. C. Langford. A global geometric frame- work for nonlinear dimensionality reduction.Science, 290(5500):2319– 2323, 2000. doi: 10.1126/science.290.5500.2319 3, 5

    work page doi:10.1126/science.290.5500.2319 2000

  67. [68]

    Thibeault, A

    V . Thibeault, A. Allard, and P. Desrosiers. The low-rank hypothesis of complex systems.Nat. Phys., 20(2):294–302, 2024. doi: 10.1038/s41567 -023-02303-0 3, 6

    work page doi:10.1038/s41567 2024

  68. [69]

    Z. Tian, X. Zhai, D. van Driel, G. van Steenpaal, M. Espadoto, and A. Telea. Using multiple attribute-based explanations of multidimensional projections to explore high-dimensional data.Comput. Graph., 98:93–104,

    work page

  69. [70]

    doi: 10.1016/j.cag.2021.04.034 2

    work page doi:10.1016/j.cag.2021.04.034 2021

  70. [71]

    Trapnell

    C. Trapnell. Constructing single-cell trajectories. https://cole-trapnell-lab. github.io/monocle3/docs/trajectories/, 2022. Accessed: 2025-09-07. 5, 8

    work page 2022

  71. [72]

    Trapnell, D

    C. Trapnell, D. Cacchiarelli, J. Grimsby, P. Pokharel, S. Li, M. Morse et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells.Nat. Biotechnol., 32(4):381–386,

    work page

  72. [73]

    doi: 10.1038/nbt.2859 6

    work page doi:10.1038/nbt.2859

  73. [74]

    Vandal, E

    A. Tsitsulin, D. Mottin, P. Karras, A. Bronstein, and E. Müller. NetLSD: Hearing the shape of a graph. InProc. KDD, pp. 2347–2356, 2018. doi: 10.1145/3219819.3219991 3

    work page doi:10.1145/3219819.3219991 2018

  74. [75]

    van der Maaten and G

    L. van der Maaten and G. Hinton. Visualizing data using t-SNE.J. Mach. Learn. Res., 9(11), 2008. https://www.jmlr.org/papers/volume9/ vandermaaten08a/vandermaaten08a.pdf. 1, 2

    work page 2008

  75. [76]

    van der Maaten, E

    L. van der Maaten, E. Postma, and J. van den Herik. Dimensionality reduc- tion: A comparative review. Technical Report TiCC-TR 2009-005, Tilburg University Technical Report, 2009. 36 pages, https://lvdmaaten.github.io/ publications/papers/TR_Dimensionality_Reduction_Review_2009.pdf. 2

    work page 2009

  76. [77]

    van Unen, T

    V . van Unen, T. Höllt, N. Pezzotti, N. Li, M. J. T. Reinders, E. Eisemann et al. Visual analysis of mass cytometry data by hierarchical stochastic neighbour embedding reveals rare cell types.Nat. Commun., 8(1), 2017. doi: 10.1038/s41467-017-01689-9 8

    work page doi:10.1038/s41467-017-01689-9 2017

  77. [78]

    Venna and S

    J. Venna and S. Kaski. Local multidimensional scaling.Neural Netw., 19(6-7):889–899, 2006. doi: 10.1016/j.neunet.2006.05.014 7

    work page doi:10.1016/j.neunet.2006.05.014 2006

  78. [79]

    D. E. Wagner and A. M. Klein. Lineage tracing meets single-cell omics: opportunities and challenges.Nat. Rev. Genet., 21(7):410–427, 2020. doi: 10.1038/s41576-020-0223-2 8

    work page doi:10.1038/s41576-020-0223-2 2020

  79. [80]

    Wang and M

    X. Wang and M. Zhang. How powerful are spectral graph neural networks. InProc. ICML, vol. 162, pp. 23341–23362. PMLR, 2022. https://proceedings.mlr.press/v162/wang22am/wang22am.pdf. 3, 9

    work page 2022

  80. [81]

    Y . Wang, Y . Sun, H. Huang, and C. Rudin. Dimension reduction with locally adjusted graphs. InProc. AAAI, vol. 39, pp. 21357–21365, 2025. doi: 10.1609/aaai.v39i20.35436 7

    work page doi:10.1609/aaai.v39i20.35436 2025

Showing first 80 references.