Pith. sign in

REVIEW 4 major objections 6 minor 46 references

Generation of synthetic gait data: application to multiple sclerosis patients' gait patterns

T0 review · 4 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

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

Pith's one-line read The paper claims that synthetic gait data faithful to quaternion geometry can be generated from a small clinical sample of 27 MS patients.

desk verdict A genuinely novel pipeline for synthetic unit quaternion gait data, but the geometry-preservation claim rests on metrics that are partly built into the construction. read the letter →

arxiv 2411.10377 v2 pith:B4V3ZRQK submitted 2024-11-15 cs.CV stat.AP

Klervi Le Gall , Lise Bellanger , David Laplaud , Aymeric Stamm This is my paper · ORCID
classification cs.CVstat.AP MSC 62H2562R30
keywords syntheticdataquaterniontimeseriesfunctionalprincipalcomponentanalysisgaitmultiplesclerosisnearestneighborweightingDirichletdistributionprivacy
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

Multiple sclerosis gait studies often have too few patients to build stable statistical tools, and sharing clinical data is privacy-sensitive. This paper claims that unit quaternion time series recording hip rotation during a step can be transformed into functional PCA scores, synthesized in that tabular space with a nearest-neighbor weighting method called SynGait, and mapped back into unit-quaternion curves that stay on the same geometric manifold as the original data. Applied to 27 MS patients, the generated synthetic gait patterns score well on fidelity metrics and preserve the data's geometry better than copula-based and GAN-based tabular synthesizers. If correct, small clinical samples can be expanded into privacy-aware synthetic datasets for developing and testing clustering and other gait-analysis tools.

What carries the argument

The central object is the unit quaternion time series (QTS), which represents hip rotation during a gait cycle; the key identity is the logarithmic/exponential map pair linking the unit-quaternion space $ mathcal{H}_u$ to its tangent space $\mathbb{R}^3$. The machinery combines the log map, cubic B-spline functionalization, multivariate functional PCA that produces eigenfunctions and a score matrix, and the flexible avatar method: synthetic scores are convex combinations of the scores of the $\gamma$ nearest neighbors, with Dirichlet weights whose concentrations are inversely proportional to neighbor distances. Equation (6) recombines synthetic scores with the original principal functions and the mean QTS to produce synthetic QTS.

What would settle it

Compute geodesic distances between reconstructed synthetic QTS and real QTS on the unit-quaternion sphere and compare them with the range seen among real gait cycles, or ask clinicians to label synthetic IGPs as plausible or implausible gait; if the synthetic curves fall far outside the real gait manifold or are flagged as implausible, the geometry-preservation claim fails.

Watch

Extended reading notes

Core claim

The central claim is that the SynGait framework produces high-fidelity synthetic unit quaternion time series for small and private datasets. The pipeline centers each QTS at the Fréchet mean, applies the logarithmic map to the tangent space, interpolates cubic B-splines, runs multivariate functional PCA to obtain a score matrix, synthesizes new scores by taking Dirichlet-weighted averages of each observation's nearest neighbors, and reconstructs QTS with the exponential map plus the mean. On the 27-patient MS dataset, the synthetic data achieves a mean RV coefficient of 0.84, Frobenius distances between k-nearest-neighbor graphs that are smaller than the copula and CTGAN baselines, and a hidden rate of 85%, supporting the claim that the generated curves are new, faithful, and geometry-preserving.

Load-bearing premise

The load-bearing premise is that closeness of synthetic scores to the nearest neighbors' real scores in PCA space implies the reconstructed quaternion curves lie on the true gait manifold and are clinically plausible, which the paper does not independently test.

Editorial extensions

If this is right

  • Small-sample gait studies can generate many synthetic QTS datasets for bootstrapping and stability analysis of clustering, without collecting more patients.
  • Because the framework reduces to tabular score generation, any existing tabular synthesizer can be plugged in, while SynGait's nearest-neighbor weighting offers better geometry preservation.
  • The method provides a practical privacy workflow: hyperparameters can be tuned so synthetic points are new and the hidden rate is high, supporting sharing of synthetic gait data.
  • The same pipeline applies to any motion describable as unit QTS, not just MS hip rotation, and to additional pathologies.
  • Default hyperparameter settings (components covering about 95% of inertia, concentration around 5, and a number of neighbors near one tenth of the sample size) make the method usable on other unit-QTS datasets.

Reading between the lines

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

  • Because synthetic scores are convex combinations of real nearest-neighbor scores, geometric closeness in score space is partly enforced by construction; the paper does not independently verify that the reconstructed QTS are clinically plausible gait curves, so a clinician review or geodesic-distance check would strengthen the claim.
  • The approach could serve as a privacy-preserving data-sharing mechanism for multi-site MS cohorts: each site can release synthetic QTS built from its own patients, and the hidden-rate and local-cloaking metrics suggest re-identification is nontrivial.
  • Neighboring problems such as synthetic accelerometer or EMG time series could reuse the same tangent-space fPCA plus nearest-neighbor score synthesis, with the same geometry-preservation claim being testable.
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 / 6 minor

Summary. The paper proposes SynGait, a two-part framework for generating synthetic unit quaternion time series (QTS) gait data. The first part transforms QTS into a tabular score matrix via Fréchet centering, logarithm map to the tangent space, B-spline smoothing, and multivariate functional PCA. The second part synthesizes new scores using a flexible avatar method: for each observation, the γ nearest neighbors are found in the first τ principal-component scores, weights are sampled from a Dirichlet distribution whose concentration parameters depend on distances to those neighbors, and the new score is a weighted average of the neighbors' scores (Eq. 8). Synthetic QTS are reconstructed by exponentiating the linear combination of principal functions and adding back the mean QTS (Eq. 6). The method is applied to 27 multiple sclerosis patients' gait data and compared with copula and CTGAN baselines using k-nearest-neighbor graph Frobenius distances, the RV coefficient, SDV fidelity metrics, and local cloaking/hidden rate; all evaluation metrics are computed on the score matrix.

Significance. If the claims are supported, the framework would be a valuable contribution: it adapts manifold-aware functional data analysis to the synthesis of rotation-valued time series, works with very small samples, and could enable privacy-preserving sharing and stability analysis of gait data. The geometric treatment of unit quaternions and the use of fPCA on the tangent space are appropriate and well motivated. The method is simple, does not require large training datasets, and the comparison with standard tabular synthesizers is a useful benchmark. However, the current evaluation is not sufficient to establish the central claims of high-fidelity QTS generation and geometry preservation, because the metrics are computed in the same score space in which the synthetic data are constructed as convex combinations of real neighbors.

major comments (4)
  1. [§4.2 and §5, Eq. (8)] All fidelity and geometry metrics are computed on the functional score matrix F, as stated at the start of Section 4.2. However, Eq. (8) generates each synthetic score vector as a convex combination of the γ nearest real score vectors. Consequently, the k-NNG Frobenius distances (Section 4.2.1) and the RV coefficient (Section 4.2.2) measure similarity in the very space where the generative process is defined, so high scores are partly entailed by construction and do not independently confirm that the reconstructed QTS (Eq. 6) preserve the geometry of the original gait manifold. Please add quantitative validation directly on the reconstructed QTS space, for example by computing geodesic distances between original and synthetic QTS, or by comparing clinically meaningful gait descriptors derived from the QTS (e.g., range of motion, timing of gait phases).
  2. [§3.4 and §5] The hyperparameters α0, γ, and τ are selected by the procedure in Section 3.4 using the same 27-patient dataset that is later used to report all fidelity and geometry metrics in Section 5. This selection-on-test-set protocol risks overfitting the reported performance to the particular sample. Please provide a resampling-based evaluation, such as leave-one-out or a train/validation split, to demonstrate that the method's performance generalizes beyond the tuning sample.
  3. [§5, Figure 5] The paper acknowledges in Section 5 that the synthetic curves 'seem to be living on a smaller space,' which is a direct consequence of the convex averaging in Eq. (8). This shrinkage implies that the synthetic QTS may not cover the full variability of the original gait patterns, but no quantitative QTS-space metric is provided to assess the impact of this shrinkage on fidelity. The claim that SynGait produces 'high-fidelity synthetic QTS data' is therefore supported only by score-space metrics, not by any direct evaluation of the reconstructed unit quaternion time series.
  4. [§6 and Abstract] The stated motivation for the work includes using synthetic data to study the stability of clustering methods on the IGP data (Abstract and Section 6), but no clustering stability experiment is reported. Adding such an analysis, for example by clustering the synthetic QTS and comparing the resulting groups with those obtained on the original data, would provide a utility-based validation that the synthetic data preserve clinically relevant structure.
minor comments (6)
  1. [§2.3.2] The text says quaternions were formalized by Sir William Hamilton in 1943; the correct year is 1843.
  2. [§1] There is a typo in the second paragraph: 'n this article' should read 'In this article'.
  3. [§3.3] The description of the nearest-neighbor weighting step would be clearer if the Dirichlet sampling were written explicitly, e.g., w_i ∼ Dir(α_i) with α_i defined as in Eq. (7), and if the notation for the weights w_ij were defined consistently with Eq. (8).
  4. [§5] The abbreviation 'IGP' is used in the results section but is not defined there; it appears earlier in the text, but a reminder at first use in Section 5 would help readability.
  5. [§5] The statement 'the mean R V coefficient is 0.84' reports a single point estimate; please include the variability across the 100 synthetic datasets, such as a standard deviation or a boxplot.
  6. [§3.4] The recommended threshold for d_min is set to 10% of the smallest distance between observations in the original dataset, but the rationale for this particular percentage is not explained.

Circularity Check

1 steps flagged · score 6.0 of 10

SynGait's geometry-preservation evidence is computed on the same score matrix in which synthetic rows are convex combinations of real neighbor rows (Eq. 8); the headline fidelity claim is therefore partly entailed by construction.

  1. self definitional [Section 3.3 Eq. (8); Section 4.2 metrics; Section 5 results (RV, k-NNG, Figure 6)]
    "We then produce synthetic scores for observation i as a weighed average of the scores of its γ nearest neighbors: f(s)i = Σγ j=1 wij fℓij ∈ R n−1, i ∈ [[1,n]]. [...] In the remainder of the paper, we mainly focus on fidelity measures as well as one privacy metric. All of these metrics are computed on the functional score matrix F (original and synthetic)."

    By Eq. (8), every synthetic score row is a convex combination of the γ nearest original score rows in the same score matrix F. Section 4.2 then evaluates all fidelity and privacy metrics on F: k-NNG Frobenius distances, the RV coefficient, SDV column similarities, and local cloaking distances. Thus the reported geometry preservation (e.g., mean RV = 0.84 and low k-NNG Frobenius distances) is largely entailed by the generative construction rather than independently established: the synthetic score cloud is forced into the convex hull of the original score cloud. Section 5 even concedes that the synthetic curves 'seem to be living on a smaller space,' the expected shrinkage from convex averaging. No experiment validates the reconstructed QTS from Eq.

full rationale

The mathematical pipeline itself is not circular: fPCA and the exponential-map reconstruction (Eq. 6) are well-defined, and the Dirichlet-weighted nearest-neighbor scheme is a concrete generative algorithm rather than a restatement of the target. The circularity is in the evaluation. Because SynGait produces synthetic scores as convex combinations of neighboring original scores in the same score space where all geometry/fidelity metrics are computed, the high scores on those metrics are partly guaranteed by the method's definition, not by an independent check of whether the synthetic unit quaternion time series lie on the true gait manifold. The comparison with copula and CTGAN synthesizers inside the same fPCA framework provides some independent benchmarking value, and the paper is transparent about the expected shrinkage; however, none of the reported quantitative metrics test the reconstructed QTS themselves. There is no load-bearing self-citation issue here; the concern is a self-referential validation loop. Score 6 reflects that a central claimed result (respecting the initial geometry) is partially constructed into the metric rather than fully demonstrated.

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

The method introduces no new physical entities. Its free parameters are hyperparameters and thresholds selected on the same data used for evaluation, which raises a circularity concern.

free parameters (4)
  • alpha_0 (Dirichlet concentration sum) = 4.52
    Selected by the proposed d_min/d_max optimization on the 27-patient MYO dataset; controls variability and privacy of synthetic scores.
  • gamma (number of nearest neighbors) = 2
    Selected by the same optimization; controls how many real score rows are averaged to make one synthetic row.
  • tau (number of principal components used in neighbor search) = 9
    Selected by the same optimization; controls the score-space dimension used for distance computations.
  • d_min threshold = 0.024
    User-chosen threshold set to 10% of the smallest distance between original score rows; used to filter hyperparameter combinations.
assumptions (4)
  • domain assumption Unit quaternions form a Lie group with well-defined log and exp maps, and the tangent space at the Frechet mean captures the gait manifold adequately
    Section 3.1-3.2 project QTS to R3 via the log map; if hip rotations are not concentrated enough, the tangent-space projection will distort the geometry.
  • domain assumption The first n-1 multivariate functional principal components capture the essential gait information, so synthesizing scores in this basis yields plausible QTS
    Section 3.1.2 and Eq. (6): principal functions are kept fixed and only scores are synthesized; information discarded by fPCA cannot be recovered.
  • domain assumption Synthetic score rows created as convex combinations of nearest-neighbor rows remain anatomically valid gait patterns
    Eq. (8) in Section 3.3; no clinical or kinematic validation is provided for interpolated scores.
  • standard math Functional PCA eigen-decomposition and multivariate functional data theory (MFPCA package) are standard and correct
    Section 3.1.2 relies on standard fPCA results from Ramsay and Silverman and the MFPCA implementation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Generation of synthetic gait data: application to multiple sclerosis patients' gait patterns." pith.science (2026). https://pith.science/paper/B4V3ZRQK

@misc{pith2026241110377,
  author       = {Pith},
  title        = {Pith review of: Generation of synthetic gait data: application to multiple sclerosis patients' gait patterns},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/B4V3ZRQK}},
  note         = {Machine review of arXiv:2411.10377}
}
read the original abstract

Multiple sclerosis (MS) is the leading cause of severe non-traumatic disability in young adults and its incidence is increasing worldwide. The variability of gait impairment in MS necessitates the development of a non-invasive, sensitive, and cost-effective tool for quantitative gait evaluation. The eGait movement sensor, designed to characterize human gait through unit quaternion time series (QTS) representing hip rotations, is a promising approach. However, the small sample sizes typical of clinical studies pose challenges for the stability of gait data analysis tools. To address these challenges, this article presents two key scientific contributions. First, a comprehensive framework is proposed for transforming QTS data into a form that preserves the essential geometric properties of gait while enabling the use of any tabular synthetic data generation method. Second, a synthetic data generation method is introduced, based on nearest neighbors weighting, which produces high-fidelity synthetic QTS data suitable for small datasets and private data environments. The effectiveness of the proposed method, is demonstrated through its application to MS gait data, showing very good fidelity and respect of the initial geometry of the data. Thanks to this work, we are able to produce synthetic data sets and work on the stability of clustering methods.

Figures

Figures reproduced from arXiv: 2411.10377 by the authors.

Figure 1
Figure 1. SynGait. Schematic overview of the proposed comprehensive framework for unit QTS synthetic data generation. Nearest neighbor weights. We next generate weights for each neighbor of each ob￾servation. These weights belong to the 𝛾-simplex and we therefore need to resort to a statistical distribution with such a support. The Dirichlet distribution is a natural choice [30]. For the 𝑖-th observation, this distribution ha… view at source ↗
Figure 2
Figure 2. Overview of how a GAN model is trained. the generator model is updated iteratively until the discriminator cannot distinguish real from fake data. The generator has no direct access to real data, so in practice the discriminator is usually trained first then the generator is trained by trying to fool the discriminator. You can find more about how GANs are trained in [38]. The architecture of the specific model for c… view at source ↗
Figure 3
Figure 3. Manifold approximation via k-NNG. A cloud of 𝑛 = 100 points sampled in the space between the two red circles. Three k-NNGs are used to approximate the corresponding manifold: with 𝑘 = 10 (left panel), with 𝑘 = 30 (middle panel) and with 𝑘 = 50 (right panel). It illustrates that a value of 𝑘 close to 30 is best in this case to approximate the space between the red circles. Since the manifold onto which the functional… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Scatterplot of the first two principal components scores for the log-QFD of the [PITH_FULL_IMAGE:figures/full_fig_p018_4.png]
Figure 5
Figure 5. Figure 5: provides a representation of the original IGP data, and synthetic data ob￾tained using the proposed framework with synthetic scores obtained by the SynGait method with the parameters previously chosen, the Copula synthesizer, and the CTGAN synthesizer. What strikes fir…
Figure 6
Figure 6. Figure 6: Distances between original manifold and synthesized ones. Distribution of the Frobenius distances between the 𝑘-NNG built on the original IGPs and 𝑘-NNGs built from 100 synthesized IGPs using the SynGait method (purple), the copula method (darkblue) and the CTGAN metho…
Figure 7
Figure 7. Figure 7: illustrates the performances of the synthetic gait generation methods based on the metrics proposed by the Synthetic Data Vault. Notably, all three methods demon￾strate strong performance results with most metrics results over 0.75. Overall the CTGAN is the one that pe…
Figure 8
Figure 8. Figure 8: Graph of the minimum distances between synthetic and original individuals [PITH_FULL_IMAGE:figures/full_fig_p024_8.png]
Figure 9
Figure 9. Figure 9: Graph of the maximum distances between synthetic individuals [PITH_FULL_IMAGE:figures/full_fig_p025_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 42 canonical work pages

  1. [1]

    Bacon, Eric Chamot, Amber R

    Ilya Kister, Tamar E. Bacon, Eric Chamot, Amber R. Salter, Gary R. Cutter, Jen- nifer T. Kalina, and Joseph Herbert. Natural History of Multiple Sclerosis Symp- toms. International Journal of MS Care , 15(3):146–156, 10 2013

    work page 2013

  2. [2]

    Nicholas G. LaRocca. Impact of walking impairment in multiple sclerosis. The Patient: Patient-Centered Outcomes Research , 4(3):189–201, September 2011

    work page 2011

  3. [3]

    L Rocher, J. M. Hendrickx, and Y de Montjoye. Estimating the success of re- identifications in incomplete datasets using generative models. Nature Communica- tions, 10, 2019

    work page 2019

  4. [4]

    Rothstein

    Mark A. Rothstein. Is deidentification sufficient to protect health privacy in research? The American Journal of Bioethics , 10(9):3–11, 2010

    work page 2010

  5. [5]

    No silver bullet: De-identification still doesn’t work

    Arvind Narayanan and Edward W Felten. No silver bullet: De-identification still doesn’t work. White Paper, 8, 2014

    work page 2014

  6. [6]

    High-fidelity synthetic data applications for data augmentation

    Zhenchen Wang, Barbara Draghi, Ylenia Rotalinti, Darren Lunn, and Puja Myles. High-fidelity synthetic data applications for data augmentation. In Manuel Dom ´ ınguez-Morales, Javier Civit-Masot, Luis Mu˜ noz-Saavedra, and Rober- tas Damaˇ seviˇ cius, editors,Deep Learning, chapter 7. IntechOpen, Rijeka, 2024

    work page 2024

  7. [7]

    Donald B. Rubin. Statistical disclosure limitation: discussion. Journal of Official Statistics, 9:2461–468, 1993. 25

    work page 1993

  8. [8]

    Multiple im- putation for statistical disclosure limitation

    Trivellore E Raghunathan, Jerome P Reiter, and Donald B Rubin. Multiple im- putation for statistical disclosure limitation. Journal of official statistics , 19(1):1, 2003

    work page 2003

Show all 46 references
  1. [9]

    A large-scale synthetic gait dataset towards in-the-wild sim- ulation and comparison study

    Pengyi Zhang, Huanzhang Dou, Wenhu Zhang, Yuhan Zhao, Zequn Qin, Dongping Hu, Yi Fang, and Xi Li. A large-scale synthetic gait dataset towards in-the-wild sim- ulation and comparison study. ACM Trans. Multimedia Comput. Commun. Appl. , 19(1), 1 2023

    2023

  2. [10]

    A vision-based system for stage classification of parkinsonian gait using machine learning and synthetic data

    Jorge Marquez Chavez and Wei Tang. A vision-based system for stage classification of parkinsonian gait using machine learning and synthetic data. Sensors, 22(12), 2022

    2022

  3. [11]

    Hargrove

    Minjae Kim and Levi J. Hargrove. Generating synthetic gait patterns based on benchmark datasets for controlling prosthetic legs. Journal of NeuroEngineering and Rehabilitation, 20(1), 9 2023

    2023

  4. [12]

    International classification of functioning, disability, and health: Icf 2001, 2001

    World Health Organization. International classification of functioning, disability, and health: Icf 2001, 2001

    2001

  5. [13]

    Nandikolla, Robin Bochen, Steven Meza, and Allan Garcia

    Vidya K. Nandikolla, Robin Bochen, Steven Meza, and Allan Garcia. Experimental gait analysis to study stress distribution of the human foot. Journal of Medical Engineering, 2017(1):3432074, 2017

    2017

  6. [14]

    Individual recognition using gait energy image

    Ju Han and Bir Bhanu. Individual recognition using gait energy image. IEEE Trans Pattern Anal Mach Intell , 28(2):316–22, 2006

    2006

  7. [15]

    Evaluation of calibrated kinect gait kinematics using a vicon motion capture system

    H Lamine, S Bennour, M Laribi, L Romdhane, and S Zaghloul. Evaluation of calibrated kinect gait kinematics using a vicon motion capture system. Computer Methods in Biomechanics and Biomedical Engineering , 20(sup1):S111–S112, 2017

    2017

  8. [16]

    Gait analysis using wear- able sensors

    Weijun Tao, Tao Liu, Rencheng Zheng, and Hutian Feng. Gait analysis using wear- able sensors. Sensors, 12(2):2255–2283, 2012

    2012

  9. [17]

    Kieseier and Carlo Pozzilli

    Bernd C. Kieseier and Carlo Pozzilli. Assessing walking disability in multiple scle- rosis. Multiple Sclerosis Journal , 18(7):914–924, 2012

    2012

  10. [18]

    John F. Kurtzke. Rating neurologic impairment in multiple sclerosis: an expanded disability status scale (edss). Neurology, 33(11), 1983

    1983

  11. [19]

    Gait impairment monitoring in multiple sclerosis using a wearable motion sensor

    Pierre Drouin, Aymeric Stamm, Laurent Chevreuil, Vincent Graillot, Laetitia Barbin, Philippe Nicolas, et al. Gait impairment monitoring in multiple sclerosis using a wearable motion sensor. Medical Case reports and Reviews , 5:1–5, 2022

    2022

  12. [20]

    Quaternion Algebras

    John Voight. Quaternion Algebras. Springer Nature, 2005

    2005

  13. [21]

    Dijkhuizen

    Mathijs S. Dijkhuizen. The double covering of the quantum group soq(3). In Proceed- ings of the Winter School ”Geometry and Physics”. Circolo Matematico di Palermo , volume 37, pages 47–57, 1994

    1994

  14. [22]

    Ramsay and Bernard W

    James O. Ramsay and Bernard W. Silverman. Principal components analysis for functional data , pages 147–172. Springer New York, 2005. 26

    2005

  15. [23]

    Patient-centric synthetic data generation, no reason to risk re-identification in biomedical data analysis

    Morgan Guillaudeux, Olivia Rousseau, Julien Petot, Zineb Bennis, Charles-Axel Dein, Thomas Goronflot, et al. Patient-centric synthetic data generation, no reason to risk re-identification in biomedical data analysis. npj Digital Medicine , 6, 2023

    2023

  16. [24]

    A micro lie theory for state estimation in robotics

    Joan Sola, Jeremie Deray, and Dinesh Atchuthan. A micro lie theory for state estimation in robotics. arXiv preprint arXiv:1812.01537 , 2018

    2018 arXiv

  17. [25]

    Ramsay and Bernard W

    James O. Ramsay and Bernard W. Silverman. Introduction, pages 1–18. Springer New York, 2005

    2005

  18. [26]

    K. Pearson. Liii. on lines and planes of closest fit to systems of points in space. The London, Edinburgh, and Dublin Philosophical Magazine and Journal of Science , 2(11):559–572, 1901

    1901

  19. [27]

    Analysis of a complex of statistical variables into principal com- ponents

    Harold Hotelling. Analysis of a complex of statistical variables into principal com- ponents. Journal of Educational Psychology , 24:498–520, 1933

    1933

  20. [28]

    Multivariate functional principal component analysis for data observed on different (dimensional) domains

    Clara Happ and Sonja Greven. Multivariate functional principal component analysis for data observed on different (dimensional) domains. Journal of the American Statistical Association, 113(522):649–659, 2018

    2018

  21. [29]

    MFPCA: Multivariate Functional Principal Component Analysis for Data Observed on Different Dimensional Domains , 2022

    Clara Happ-Kurz. MFPCA: Multivariate Functional Principal Component Analysis for Data Observed on Different Dimensional Domains , 2022. R package version 1.3-10

    2022

  22. [30]

    Dirichlet and Related Distri- butions: Theory, Methods and Applications

    Kai Wang Ng, Guo Liang Tian, and Man Lai TANG. Dirichlet and Related Distri- butions: Theory, Methods and Applications . Wiley-Blackwell, 2011

    2011

  23. [31]

    The synthetic data vault

    Neha Patki, Roy Wedge, and Kalyan Veeramachaneni. The synthetic data vault. In 2016 IEEE International Conference on Data Science and Advanced Analytics (DSAA), pages 399–410, 2016

    2016

  24. [32]

    Sequential models in the synthetic data vault

    Kevin Zhang, Neha Patki, and Kalyan Veeramachaneni. Sequential models in the synthetic data vault. arXiv preprint arXiv:2207.14406 , 2022

    2022 arXiv

  25. [33]

    Shih-Chieh Kao, Hoe Kyoung Kim, Cheng Liu, Xiaohui Cui, and Budhendra L. Bhaduri. Dependence-preserving approach to synthesizing household characteristics. Transportation Research Record, 2302(1):192–200, 2012

    2012

  26. [34]

    Learning vine cop- ula models for synthetic data generation

    Yi Sun, Alfredo Cuesta-Infante, and Kalyan Veeramachaneni. Learning vine cop- ula models for synthetic data generation. Proceedings of the AAAI Conference on Artificial Intelligence, 33(01):5049–5057, 7 2019

    2019

  27. [35]

    Fonctions de r´ epartition ` a n dimensions et leurs marges

    M Sklar. Fonctions de r´ epartition ` a n dimensions et leurs marges. In Annales de l’ISUP, volume 8, pages 229–231, 1959

    1959

  28. [36]

    Copula-based synthetic data generation for machine learning emulators in weather and climate: application to a simple radiation model

    David Meyer, Thomas Nagler, and Robin J Hogan. Copula-based synthetic data generation for machine learning emulators in weather and climate: application to a simple radiation model. Geoscientific Model Development Discussions , 2021:1–21, 2021. 27

    2021

  29. [37]

    Generative adversarial networks

    Ian Goodfellow, Jean Pouget-Abadie, Mehdi Mirza, Bing Xu, David Warde-Farley, Sherjil Ozair, Aaron Courville, and Yoshua Bengio. Generative adversarial networks. Communications of the ACM , 63(11):139–144, 2020

    2020

  30. [38]

    Antonia Creswell, Tom White, Vincent Dumoulin, Kai Arulkumaran, Biswa Sen- gupta, and Anil A. Bharath. Generative adversarial networks: An overview. IEEE Signal Processing Magazine, 35(1):53–65, 2018

    2018

  31. [39]

    Modeling tabular data using conditional gan

    Lei Xu, Maria Skoularidou, Alfredo Cuesta-Infante, and Kalyan Veeramachaneni. Modeling tabular data using conditional gan. Advances in neural information pro- cessing systems, 32, 2019

    2019

  32. [40]

    Computational geometry: an introduc- tion

    Franco P Preparata and Michael I Shamos. Computational geometry: an introduc- tion. Springer Science & Business Media, 2012

    2012

  33. [41]

    Marchette

    David J. Marchette. cccd: Class Cover Catch Digraphs , 2022. R package version 1.6

    2022

  34. [42]

    Josse, J

    J. Josse, J. Pag` es, and F. Husson. Testing the significance of the rv coefficient. Computational Statistics & Data Analysis , 53(1):82–91, 2008

    2008

  35. [43]

    Robert and Y

    P. Robert and Y. Escoufier. A unifying tool for linear multivariate statistical meth- ods: The rv- coefficient. Journal of the Royal Statistical Society. Series C (Applied Statistics), 25(3):257–265, 1976

    1976

  36. [44]

    Demystifying membership inference attacks in machine learning as a service

    Stacey Truex, Ling Liu, Mehmet Emre Gursoy, Lei Yu, and Wenqi Wei. Demystifying membership inference attacks in machine learning as a service. IEEE Transactions on Services Computing , 14:2073–2089, 2019

    2019

  37. [45]

    Anonymization techniques for privacy preserving data publishing: A comprehensive survey

    Abdul Majeed and Sungchang Lee. Anonymization techniques for privacy preserving data publishing: A comprehensive survey. IEEE Access, 9:8512–8545, 2021

    2021

  38. [46]

    Semi-supervised clustering of quaternion time series: Application to gait analysis in multiple sclerosis using motion sensor data

    Pierre Drouin, Aymeric Stamm, Laurent Chevreuil, Vincent Graillot, Laetitia Barbin, Pierre-Antoine Gourraud, et al. Semi-supervised clustering of quaternion time series: Application to gait analysis in multiple sclerosis using motion sensor data. Statistics in Medicine , 42(4)...

    2022

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.