Pith. sign in

REVIEW 2 major objections 5 minor 52 references

Joint estimation of high-dimensional spiked covariance matrices via a partially shared subspace

T0 review · 2 major / 5 minor · reviewed 2026-07-10 · grok-4.5

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

Pith's one-line read When two high-dimensional covariances share only part of their spiked subspace, you can recover that shared part, optimally pool it, and improve the target covariance estimate.

desk verdict Solid high-dim RMT paper that generalizes partial subspace sharing and gives a usable pooling estimator; the main soft spot is the unproved leap from oracle consistency to the iterative algorithm they actually run. read the letter →

arxiv 2607.08123 v1 pith:JIU5QSKO submitted 2026-07-09 stat.ME stat.APstat.ML

Changwon Yoon , Minwoo Kim , Sungkyu Jung , Jeongyoun Ahn This is my paper
classification stat.MEstat.APstat.ML MSC 62H2562H1260B20
keywords commonprincipalcomponentscontrastivedimensionreductionhigh-dimensionalcovarianceestimationrandommatrixtheorytransferlearningspikedmodelpartiallysharedsubspace
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

High-dimensional covariance estimation is usually starved for samples, yet related auxiliary data often exist. This paper argues that the useful overlap is geometric: the two spiked covariances share a low-dimensional subspace of unknown rank that can sit anywhere in the spectrum, while each keeps its own distinct spiked directions. Under proportional growth and random-matrix asymptotics, the sample principal angles between the two spiked subspaces converge to deterministic limits; those limits let you estimate both the shared rank and the indices of the eigenvectors that span it. Once the shared projection is identified, a closed-form weight optimally pools the two sample projections to minimize asymptotic Frobenius loss, and the resulting plug-in estimator for the target covariance improves on any target-only spiked estimator. The same shared-versus-distinct split supplies a high-dimensional contrastive dimension-reduction method with asymptotic guarantees, something existing contrastive PCA-style procedures lack. Applications to pandemic portfolio construction and tumor-versus-normal gene expression illustrate the practical payoff.

What carries the argument

The partially shared subspace (PSS) model together with the almost-sure limits of sample principal angles (Lemma 2 / Theorem 3): shared singular values converge to a product of attenuation factors times a rotation, distinct singular values converge separately, and a debiased cutoff separates them, driving the iterative rank-and-index estimator and the optimal pooling weight.

What would settle it

Generate data from the PSS model with known shared rank and mixed index sets under proportional growth; check whether Algorithm 1 recovers the true shared rank and index sets with probability approaching one, and whether the pooled covariance's Frobenius error falls below the target-only spiked estimator at the rate predicted by the asymptotic loss formula.

Watch

Extended reading notes

Core claim

Under the partially shared subspace (PSS) model, the shared rank and the index sets of the shared spiked eigenvectors can be recovered consistently from the singular values and principal angles of the two sample spiked subspaces; the shared projection can then be estimated by a weighted sum of the two sample projections whose optimal weight has a closed form that depends on relative sample sizes and relative spike strengths, yielding an asymptotic efficiency gain over target-only estimation and a consistent high-dimensional contrastive subspace.

Load-bearing premise

The shared directions must produce sample singular values that stay strictly larger than those of the distinct directions, and the alternating algorithm that uses those cutoffs is assumed to inherit the consistency proved only for its oracle versions.

Editorial extensions

If this is right

  • Target covariance estimates in high dimensions can be improved by any related background dataset that shares even a non-leading subspace, without requiring the whole eigenbasis or parameter proximity.
  • Negative transfer is self-limiting: if no shared structure exists the estimated shared rank collapses to zero.
  • Contrastive dimension reduction gains its first high-dimensional asymptotic guarantees via the PSS distinct subspace.
  • Portfolio risk estimates during regime shifts can exploit pre-shift returns as background while isolating crisis-specific factors.
  • Tumor-versus-normal gene-expression analysis can separate organ-level shared variation from disease-specific directions with quantified error.

Reading between the lines

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

  • The same principal-angle separation idea could extend to more than two groups if a multi-subspace angle or joint-and-individual variation measure replaces the two-matrix singular values.
  • Combining the optimal shared-projection pooling with existing optimal eigenvalue shrinkage would likely produce a still tighter covariance estimator for downstream tasks such as LDA or GMV portfolios.
  • The method itself can serve as a diagnostic: a stable zero shared-rank estimate is evidence against a shared-subspace hypothesis, something many common-PCA procedures cannot detect.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. The paper introduces a partially shared subspace (PSS) spiked covariance model for two high-dimensional datasets that share an r_S-dimensional subspace of unknown rank and arbitrary spectral position while retaining distinct spikes. Under proportional growth (Assumptions 1–2) and a shared/distinct singular-value separation condition (Assumption 3), it derives almost-sure limits for principal angles between sample eigenspaces (Lemma 2, Theorem 3), constructs oracle estimators for shared rank and index sets that are a.s. consistent (Theorems 5–6), and proposes an iterative Algorithm 1 that alternates between them. A weighted pooled projection estimator of the shared subspace is shown to have asymptotic Frobenius loss L(α) with closed-form optimal weight α* (Theorem 7), yielding a plug-in target covariance estimator and a high-dimensional contrastive dimension-reduction estimator from the distinct subspace. Simulations under the PSS model, misspecification, and degenerate endpoints, plus two real-data illustrations (COVID-era GMV portfolios and LGG gene expression), support the claims.

Significance. If the results hold, the paper supplies a clean geometric alternative to proximity-based transfer learning for high-dimensional covariances and the first contrastive dimension-reduction procedure with asymptotic guarantees in the proportional-growth regime. The RMT foundations (extensions of Johnstone/Paul/BBP) are carefully derived, the optimal pooling weight is closed-form and interpretable, and the framework strictly generalizes existing common-PC and multi-group models (Remark 1). The simulations are extensive (including misspecification and degenerate endpoints) and the real-data examples are coherent. The main practical deliverable—an implementable joint estimator of shared rank, indices, and the pooled covariance—would be useful for portfolio construction, multi-study genomics, and related settings where background data are abundant but only partially aligned.

major comments (2)
  1. End of §3.2 and Theorems 5–6: the paper itself states that oracle consistency of ˆr*_S and ˆΨ*_X, ˆΨ*_Y “do not automatically guarantee the consistency of the empirical estimator ˆr_S output by Algorithm 1,” and that termination is not universally guaranteed (only that the finite-state map eventually cycles or fixes). The subsequent asymptotic loss L(α), optimal weight α*, and efficiency claims for ˆΣ_X and the distinct-subspace CDR estimator all attach to the procedure that is actually run. Either a consistency argument for the alternating map under Assumptions 1–3, or a high-probability bound that it recovers the oracle fixed point, is needed before the central estimation claims can be regarded as fully established.
  2. Assumption 3 and Discussion §6 / Appendix A4.2: the separation condition ϕ_cX(λ_rS,S)ϕ_cY(γ_rS,S) > σ_1(Φ_X,D Q_D^⊤ U_D Φ_Y,D) is load-bearing for Theorems 5–6 and for the cutoff used in Algorithm 1. Finite-sample underestimation of the cutoff is only patched by an ad-hoc margin ε whose three candidates (0, 1/√(n_X+n_Y), parametric bootstrap) have no uniformly best choice. The paper should either supply a data-driven, theoretically justified rule for ε or quantify the probability that the unadjusted cutoff recovers r_S under the stated assumptions, so that practitioners know when the procedure is reliable.
minor comments (5)
  1. Notation for debiased spikes (˜λ, ˜γ) and the debiasing map d(ℓ,c) appears in §3.2 without an explicit display of the inverse of the BBP map; a short displayed equation would help readers.
  2. Figure 1 and Table 1: the mixed-position, small-sample cells show substantial underestimation of r_S by PSS; a brief discussion of when the method is expected to struggle would improve interpretability.
  3. Appendix A1 Algorithm 1: the initialization ˆΨ_X = {r_X}, ˆΨ_Y = {r_Y} is natural but could be motivated more explicitly (why the largest indices rather than, e.g., a random or top-k start).
  4. Related-work paragraph on transfer learning: a short comparison of the geometric PSS condition versus the usual Frobenius/sparsity proximity conditions would clarify the novelty claim.
  5. Typos / polish: “efficiency” (ligature issues), “difficulty”, and a few missing spaces around math operators appear in the main text and appendix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: asymptotic claims rest on external RMT limits (Paul, Baik–Silverstein, Johnstone) and standard singular-value continuity; the iterative Algorithm 1 is an unproved practical bridge, not a definitional reduction of any claimed result.

full rationale

The paper’s load-bearing derivations (Lemma 2, Theorems 3–7) start from classical single-sample spiked asymptotics (Proposition 1, citing Paul 2007 / Baik–Silverstein) and extend them via elementary almost-sure arguments (inner-product decompositions, block-diagonalization of singular values, continuity of singular values, and the CS decomposition). The oracle estimators ˆr*_S and ˆΨ* are shown consistent under Assumptions 1–3 by direct comparison of limiting singular-value multisets; the approximate loss L†(α) and optimal weight α* follow by expanding the Frobenius distance of the weighted projection sum and collecting the known ϕc factors. No parameter is fitted to data and then re-used as a “prediction” of a related quantity; no uniqueness theorem or ansatz is imported from the authors’ own prior work; the model strictly generalizes (rather than renames) earlier common-subspace models. The only interdependence is the chicken-and-egg between rank and index sets inside Algorithm 1, which the paper itself flags as unproved for the empirical (non-oracle) estimator. That is a gap in the consistency theory, not a circular derivation of any stated theorem. Hence the derivation chain is self-contained against external RMT benchmarks and scores 0.

Assumptions & free parameters 3 free parameters · 4 assumptions · 2 invented entities

The claim rests on classical spiked RMT plus three modeling assumptions (Gaussianity/spiked form, proportional growth with BBP-supercritical spikes, and shared/distinct singular-value separation). Free parameters are mainly estimation knobs (ε margin, bulk-estimated ranks, plug-in α). The PSS decomposition is the main invented structure; it is a modeling choice with empirical motivation, not an unfalsifiable particle.

free parameters (3)
  • cutoff margin ε
    Finite-sample relaxation of the asymptotic singular-value threshold; Appendix A4.2 compares 0, 1/√(nX+nY), and bootstrap SD with no uniformly best choice.
  • spike counts rX, rY and bulk levels
    Treated as known after bulk eigenvalue matching (Ke et al. 2023); errors there propagate into shared-rank estimation.
  • pooling weight α (plug-in ˆα)
    Optimal α* depends on unknown shared spikes and T; estimated via debiased eigenvalues and a diagonal approximation of T.
assumptions (4)
  • domain assumption Data are i.i.d. Gaussian with spiked covariances ΣX, ΣY (unit bulk after rescaling).
    Section 2.1 setup; enables classical spiked RMT eigenstructure limits.
  • domain assumption Proportional growth p/nX→cX, p/nY→cY with all spikes above BBP thresholds (Assumptions 1–2).
    Required for Proposition 1 and Lemma 2; without it sample eigenvectors carry no signal.
  • ad hoc to paper Assumption 3: ϕcX(λrS,S)ϕcY(γrS,S) > σ1(ΦX,D Q⊤D UD ΦY,D).
    Separation needed for threshold recovery of rS; may fail if distinct directions align strongly or shared spikes are weak.
  • standard math Standard singular-value / projection perturbation facts (Stewart–Sun, Vershynin concentration lemmas).
    Used throughout Appendix A2 proofs of Theorems 3, 6, 7.
invented entities (2)
  • Partially shared subspace (PSS) model independent evidence
    purpose: Decompose spiked eigenspaces into shared subspace of unknown rank/position and dataset-specific distinct spikes.
    Modeling construct that generalizes prior common-subspace models; falsifiable via estimated rS→0 when no sharing exists.
  • Oracle-then-iterative shared rank/index estimators (ˆr∗S, ˆΨ∗, Algorithm 1)
    purpose: Identify which sample eigenvectors span the shared subspace without knowing rS a priori.
    Core methodological object; oracle consistency proved, iterative map only empirically validated.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Joint estimation of high-dimensional spiked covariance matrices via a partially shared subspace." pith.science (2026). https://pith.science/paper/JIU5QSKO

@misc{pith2026260708123,
  author       = {Pith},
  title        = {Pith review of: Joint estimation of high-dimensional spiked covariance matrices via a partially shared subspace},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JIU5QSKO}},
  note         = {Machine review of arXiv:2607.08123}
}
read the original abstract

Statistical analysis of high-dimensional data is often hampered by limited sample sizes, yet auxiliary datasets from related sources are often readily available. When two such datasets share part of their covariance structure, but not all of it, exploiting the shared part can substantially improve estimation. We propose a spiked covariance model that explicitly captures this partial sharing: two datasets share a subspace of unknown rank and arbitrary position in the spectrum, while each retains its own distinct spiked directions. The model treats the two datasets symmetrically and strictly generalizes existing models for shared covariance structure. We develop a complete estimation procedure that includes joint estimation of the shared subspace and its rank, a closed-form pooling weight for combining the two datasets, and asymptotic guarantees derived from random matrix theory in the proportional-growth regime. The framework also resolves a gap in contrastive dimension reduction by providing a principled estimator for high-dimensional settings. We illustrate the methodology on portfolio construction during the early COVID-19 pandemic and on contrastive analysis of brain tumor gene expression.

Figures

Figures reproduced from arXiv: 2607.08123 by the authors.

Figure 1
Figure 1. Frobenius error ‖Σ̂𝑋 − Σ𝑋‖𝐹 for target covariance estimation under the PSS model, across the four configurations (shared-index position × orthogonality of distinct subspaces) and two (𝑝, 𝑛𝑋, 𝑛𝑌 ) regimes. Boxplots over 100 replications. estimates of the target covariance: the sample covariance matrix 𝑆𝑋, the spiked covariance estimate 𝑄̂Λ̂𝑄̂⊤+𝐼𝑝 , and the naïvely pooled sample covariance (𝑛𝑋+𝑛𝑌 ) −1(𝑛𝑋𝑆𝑋+𝑛𝑌 𝑆𝑌 ). Th… view at source ↗
Figure 2
Figure 2. Estimation errors for target covariance estimation under four departures from [PITH_FULL_IMAGE:figures/full_fig_p025_2.png] view at source ↗
Figure 3
Figure 3. (Left) Out-of-sample standard deviation of GMV portfolio returns over March 2020. The S&P 500 index and the equal-weighted portfolio (EWP) are included as non￾optimized benchmarks. (Top-right) Loadings of the first shared eigenvector ̂𝑞𝑆,1. (Bottom￾right) Loadings of the first distinct eigenvector 𝐷,1 ̂𝑞 . The left panel in [PITH_FULL_IMAGE:figures/full_fig_p026_3.png] view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: LGG samples projected onto: (a) Eigenspace from PCA with tumor data, (b) the shared and (c) the distinct subspaces estimated by PSS with normal brain tissue as background, and the contrastive subspaces from (d) pcPCA and (e) gcPCA. Astrocytoma and oligodendroglioma sub…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

  1. [1]

    2018 , publisher=

    Johnstone, Iain M and Paul, Debashis , journal=. 2018 , publisher=

    work page 2018

  2. [2]

    Annals of Statistics , volume=

    Optimal shrinkage of eigenvalues in the spiked covariance model , author=. Annals of Statistics , volume=

    work page

  3. [3]

    Journal of Machine Learning Research , volume=

    Shared subspace models for multi-group covariance estimation , author=. Journal of Machine Learning Research , volume=

    work page

  4. [4]

    2020 , howpublished =

    Franks, Alexander , title =. 2020 , howpublished =

    work page 2020

  5. [5]

    Annals of Statistics , volume=

    On the distribution of the largest eigenvalue in principal components analysis , author=. Annals of Statistics , volume=. 2001 , publisher=

    work page 2001

  6. [6]

    Journal of the American Statistical Association , volume=

    Estimation of the number of spiked eigenvalues in a covariance matrix by bulk eigenvalue matching analysis , author=. Journal of the American Statistical Association , volume=. 2023 , publisher=

    work page 2023

  7. [7]

    SIAM Journal on Matrix Analysis and Applications , volume=

    Schubert varieties and distances between subspaces of different dimensions , author=. SIAM Journal on Matrix Analysis and Applications , volume=. 2016 , publisher=

    work page 2016

  8. [8]

    Statistica Sinica , volume=

    Asymptotics of sample eigenstructure for a large dimensional spiked covariance model , author=. Statistica Sinica , volume=

    work page

Show all 52 references
  1. [9]

    Journal of Multivariate Analysis , volume=

    Eigenvalues of large sample covariance matrices of spiked population models , author=. Journal of Multivariate Analysis , volume=. 2006 , publisher=

    2006

  2. [10]

    Vershynin, Roman , title =

  3. [11]

    2019 , publisher=

    Probability: Theory and examples , author=. 2019 , publisher=

    2019

  4. [12]

    Nature Communications , volume=

    Exploring patterns enriched in a dataset with contrastive principal component analysis , author=. Nature Communications , volume=. 2018 , publisher=

    2018

  5. [13]

    The Annals of Applied Statistics , volume=

    Probabilistic contrastive dimension reduction for case-control study data , author=. The Annals of Applied Statistics , volume=. 2024 , publisher=

    2024

  6. [14]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Unsupervised learning with contrastive latent variable models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  7. [15]

    Advances in Neural Information Processing Systems , volume=

    Contrastive dimension reduction: When and how? , author=. Advances in Neural Information Processing Systems , volume=

  8. [16]

    Common principal components in

    Flury, Bernhard N , journal=. Common principal components in. 1984 , publisher=

    1984

  9. [17]

    Biometrika , volume=

    Two generalizations of the common principal component model , author=. Biometrika , volume=. 1987 , publisher=

    1987

  10. [18]

    Deflation methods for sparse

    Mackey, Lester , journal=. Deflation methods for sparse

  11. [19]

    and Sun, Ji-guang , publisher =

    Stewart, Gilbert W. and Sun, Ji-guang , publisher =. Matrix perturbation theory , year =

  12. [20]

    2007 , publisher=

    Applied multivariate statistical analysis , author=. 2007 , publisher=

    2007

  13. [21]

    Electronic Journal of Statistics , volume =

    Estimating structured high-dimensional covariance and precision matrices: Optimal rates and adaptive estimation , author=. Electronic Journal of Statistics , volume =

  14. [22]

    arXiv preprint arXiv:2510.11847 , year=

    Contrastive dimension reduction: A systematic review , author=. arXiv preprint arXiv:2510.11847 , year=

  15. [23]

    Journal of the American Statistical Association , volume=

    Transfer learning under high-dimensional generalized linear models , author=. Journal of the American Statistical Association , volume=. 2023 , publisher=

    2023

  16. [24]

    Proceedings of the International Conference on Artificial Intelligence and Statistics , pages=

    Transfusion: Covariate-shift robust transfer learning for high-dimensional regression , author=. Proceedings of the International Conference on Artificial Intelligence and Statistics , pages=. 2024 , organization=

    2024

  17. [25]

    arXiv preprint arXiv:2501.02411 , year=

    Transfer learning via regularized linear discriminant analysis , author=. arXiv preprint arXiv:2501.02411 , year=

    arXiv

  18. [26]

    2023 , school=

    Transfer learning in classification and regression with summary statistics , author=. 2023 , school=

    2023

  19. [27]

    Duan, Junting and Pelger, Markus and Xiong, Ruoxuan , journal=. Target. 2024 , publisher=

    2024

  20. [28]

    arXiv preprint arXiv:2403.07431 , year=

    Knowledge transfer across multiple principal component analysis studies , author=. arXiv preprint arXiv:2403.07431 , year=

    arXiv

  21. [29]

    2024 , institution=

    Transfer learning for covariance matrix estimation: Optimality and Adaptivity , author=. 2024 , institution=

    2024

  22. [30]

    Journal of the American Statistical Association , volume=

    Between-groups comparison of principal components , author=. Journal of the American Statistical Association , volume=. 1979 , publisher=

    1979

  23. [31]

    Biometrika , volume=

    Common principal component subspaces in two groups , author=. Biometrika , volume=. 1988 , publisher=

    1988

  24. [32]

    Biometrika , volume=

    Some tests for common principal component subspaces in several groups , author=. Biometrika , volume=. 1991 , publisher=

    1991

  25. [33]

    Biometrics , volume=

    Semiparametric partial common principal component analysis for covariance matrices , author=. Biometrics , volume=. 2021 , publisher=

    2021

  26. [34]

    Personalized

    Shi, Naichen and Al Kontar, Raed , journal=. Personalized

  27. [35]

    2022 , howpublished =

    Shi, Naichen and Al Kontar, Raed , title =. 2022 , howpublished =

    2022

  28. [36]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Contrastive functional principal component analysis , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  29. [37]

    and Gasalberti, D

    Karsonovich, T. and Gasalberti, D. P. and Rayi, A. , title =. StatPearls [Internet] , publisher =. 2025 , month = jan, url =

    2025

  30. [38]

    Visualizing and interpreting cancer genomics data via the

    Goldman, Mary J and Craft, Brian and Hastie, Mim and Repe. Visualizing and interpreting cancer genomics data via the. Nature Biotechnology , volume=. 2020 , publisher=

    2020

  31. [39]

    Dimension estimation in noisy

    Ulfarsson, Magnus O., and Solo, Victor , journal=. Dimension estimation in noisy. 2008 , publisher=

    2008

  32. [40]

    2009 , publisher=

    Modern Portfolio Theory and Investment Analysis , author=. 2009 , publisher=

    2009

  33. [41]

    The Journal of Finance , volume=

    Risk reduction in large portfolios: Why imposing the wrong constraints helps , author=. The Journal of Finance , volume=. 2003 , publisher=

    2003

  34. [42]

    The examination of

    Horvath, Dominik and Wang, Yung-Lin , journal=. The examination of. 2021 , publisher=

    2021

  35. [43]

    Analysis of risk correlations among stock markets during the

    Wu, JunFeng and Zhang, Chao and Chen, Yun , journal=. Analysis of risk correlations among stock markets during the. 2022 , publisher=

    2022

  36. [44]

    Scientific Reports , volume=

    Principal components analysis and the reported low intrinsic dimensionality of gene expression microarray data , author=. Scientific Reports , volume=. 2016 , publisher=

    2016

  37. [45]

    Nature Biotechnology , volume=

    A global map of human gene expression , author=. Nature Biotechnology , volume=. 2010 , publisher=

    2010

  38. [46]

    Identifying patterns differing between high-dimensional datasets with generalized contrastive

    de Oliveira, Eliezyer Fermino and Garg, Pranjal and Hjerling-Leffler, Jens and Batista-Brito, Renata and Sjulson, Lucas , journal=. Identifying patterns differing between high-dimensional datasets with generalized contrastive. 2025 , publisher=

    2025

  39. [47]

    Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=

    Probabilistic principal component analysis , author=. Journal of the Royal Statistical Society Series B: Statistical Methodology , volume=. 1999 , publisher=

    1999

  40. [48]

    Journal of Multivariate Analysis , volume=

    Angle-based joint and individual variation explained , author=. Journal of Multivariate Analysis , volume=. 2018 , publisher=

    2018

  41. [49]

    Journal of Machine Learning Research , volume=

    High-dimensional linear discriminant analysis classifier for spiked covariance model , author=. Journal of Machine Learning Research , volume=

  42. [50]

    Tony and Li, Hongzhe , title =

    Li, Sai and Cai, T. Tony and Li, Hongzhe , title =. Journal of the American Statistical Association , volume =. 2023 , publisher =

    2023

  43. [51]

    Journal of Machine Learning Research , volume =

    Ren, Mingyang and Zhen, Yaoming and Wang, Junhui , title =. Journal of Machine Learning Research , volume =

  44. [52]

    Hendy, Sharon and Dar, Yehuda , journal=

Pith tools

Reviewed July 10, 2026 · model on record in the stance chip above.