Pith. sign in

REVIEW 5 major objections 6 minor 4 cited by

Optimal Estimation of Shared Singular Subspaces across Multiple Noisy Matrices

T0 review · 5 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 This paper establishes that when several noisy low-rank matrices share a left singular subspace, the Stack-SVD estimator—the SVD of the concatenated matrices—is minimax rate-optimal, and that when the shared subspace is only partial…

desk verdict A serious minimax-theory contribution on Stack-SVD with a real but bridgeable gap between the oracle results and the claimed guarantees for the practical algorithm. read the letter →

arxiv 2411.17054 v2 pith:OW2RF4NC submitted 2024-11-26 math.ST stat.MEstat.MLstat.TH

Zhengchi Ma , Rong Ma This is my paper · ORCID
classification math.STstat.MEstat.MLstat.TH MSC 62H1262C2062H25
keywords sharedsingularsubspacelow-rankmatrixdenoisingStack-SVDminimaxoptimalityvectortracingmulti-viewdataintegrationspectralmethods
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks how accurately one can recover the shared left singular subspace of several low-rank signal matrices from their noisy observations. Its central answer is that Stack-SVD, which computes the top singular vectors of the concatenated noisy matrix, is minimax rate-optimal when the matrices share the same singular subspace, meaning no estimator can have smaller worst-case error up to constants, with matching upper and lower bounds of order $n(\gamma^2 + p_1 + p_2)/\gamma^4$ in squared spectral distance. The paper then characterizes when Stack-SVD stays optimal under partial sharing: it remains rate-optimal when unshared signals are weak, while a shifted version that selects the shared directions of the stacked SVD is needed when unshared signals dominate. It also proposes an algorithm that traces which singular vectors of the stacked matrix are shared, and proves that under orthogonality of the unshared vectors it recovers the correct set with probability tending to one. The result is a benchmark for multi-view data integration: simple stacking is provably as good as any method in the fully shared case, and a modest tracing step restores optimality when parts of the subspace are matrix-specific.

What carries the argument

The argument is carried by two structural facts and a separation index. Proposition 1 says that when all singular vectors of the two signal matrices are pairwise orthogonal, every singular vector of $X_1$ and $X_2$ appears in the SVD of the stacked matrix $(X_1\, X_2)$, possibly reordered, so estimating the shared subspace reduces to locating the correct columns of the stacked SVD. Proposition 2 is a one-sided perturbation bound for a selected block of $r$ singular vectors, giving $\mathbb{E}\|\sin\Theta(\hat U_r, U_r)\|^2 \le c p_1(\sigma_r^2(X)+p_2)/\sigma_r^4(X)\wedge 1$; it replaces the uniform two-sided perturbation bound that would make the rate depend on the wrong side of the matrix. Around these sit the parameter spaces $H_{r,t}$ and $S_{r,t}$, defined by requiring an eigen-gap $g^2 > c\sigma_{s+1}^2$ at each vector-type switch and a minimum switch gap $t^2$; the index set $J$ of shared singular vectors in the stacked matrix is what the oracle estimator and the tracing algorithm are built to recover.

What would settle it

Simulate the fully shared two-matrix model over a grid of $n$, $p_1=p_2$, and per-direction signal $\gamma^2$ chosen so the predicted minimax rate $n(\gamma^2+p_1+p_2)/\gamma^4$ lies strictly between 0 and 1, and compare the empirical worst-case $\|\sin\Theta(U,\hat U)\|^2$ of Stack-SVD with that rate; if the observed errors decay at a strictly faster rate as $n$ grows, the matching lower bound cannot be correct.

Watch

Extended reading notes

Core claim

The core discovery is that the minimax risk for estimating a fully shared left singular subspace from noisy matrices is, up to constants, $n(\gamma^2 + p_1 + p_2)/\gamma^4$ in squared spectral distance, with an extra factor $r$ in Frobenius distance, where $\gamma$ is the minimum over shared directions of the sum of squared singular values across matrices; Stack-SVD attains this rate and no estimator can improve on it when the matrix dimensions are comparable or the combined signal is large. In the partial-sharing model where each matrix has shared vectors $U_r$ plus unshared vectors $U_{1*}$ and $U_{2*}$ with $U_{1*}^\top U_{2*}=0$, the optimal rate is governed by the minimum eigen-gap $t$ at the points where the stacked singular vectors switch type: an oracle that selects the shared positions achieves $n(t^2 + p_1 + p_2)/t^4$, with matching lower bounds. The naive top-$r$ Stack-SVD selector becomes inconsistent when unshared signals are strong, but the paper shows that selecting the singular vectors at the correct positions, for example the $(d+1)$-th through $(d+r)$-th when $d$ unshared vectors lead in the stacked spectrum, restores the minimax rate. Non-orthogonality of the unshared subspaces is shown not to change this picture for the oracle estimator, because the stacking SVD rotates the unshared block while leaving the shared subspace and its singular values intact.

Load-bearing premise

The paper's fast algorithm for telling shared from unshared singular vectors is proven to work only when the unshared vectors point in mutually orthogonal directions; if they do not, the bookkeeping can include too many vectors, and the full procedure's optimality is not established.

Editorial extensions

If this is right

  • When all $k$ matrices share the same left singular subspace, Stack-SVD is minimax rate-optimal, so any averaging or principal-angle alternative can do no better under comparable dimensions or strong signals.
  • Stacking can identify shared directions that are individually non-identifiable in every single matrix, because the squared singular value of a shared direction in the stacked matrix is the sum of its squared signals across matrices.
  • When unshared signals dominate the shared ones, taking the top $r$ singular vectors of the stacked matrix is inconsistent, but selecting the singular vectors at the shared positions restores the minimax rate.
  • The tracing algorithm separating shared from unshared singular vectors is consistent when the unshared vectors are mutually orthogonal and the singular values are well separated, making the oracle estimator practically implementable.
  • For the oracle estimator, non-orthogonal unshared subspaces do not change the minimax rate, since the stacked SVD preserves the shared subspace through the rotation of the unshared block.

Reading between the lines

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

  • Editorial inference: a natural extension is to adapt the tracing algorithm to quantify mild non-orthogonality of unshared vectors, using the gap between within-matrix and cross-matrix $\sin\Theta$ distances; the paper's over-inclusive-set observation suggests the size of that gap carries information about the angle between unshared subspaces.
  • Editorial inference: the phase-transition threshold $\min\{\sigma^2_{(i)}(X_1)+\sigma^2_{(i)}(X_2)\}/\tau^2 \asymp \sqrt{n(n+p_1+p_2)}$ gives a practical diagnostic for whether stacking will help, and suggests that adaptive procedures trading off individual SVD and stacked SVD could interpolate smoothly across the critical region.
  • Editorial inference: the one-sided perturbation bound should extend to the right singular subspace by swapping the roles of $n$ and $p_i$, and to higher-order analogues such as stacked tensors, yielding similar minimax benchmarks for multi-view problems outside the matrix case.
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

5 major / 6 minor

Summary. This paper studies the problem of estimating a shared left singular subspace Ur from multiple noisy matrices Yi = Xi + Zi in a low-rank matrix denoising framework. The authors analyze Stack-SVD, which takes the top left singular vectors of the stacked matrix (Y1 Y2), and establish minimax upper and lower bounds for sin-Theta risk when the signal matrices share an identical singular subspace (Theorem 2.1), with extensions to k matrices (Corollary 2.3). For partially shared subspaces, they introduce parameter spaces with orthogonal unshared subspaces (H_{r,t}) and with non-orthogonal unshared subspaces (S_{r,t}), and prove matching upper and lower bounds for an oracle estimator that knows the true index set J of shared vectors in the stacked SVD (Theorems 3.1-3.5). They then propose Algorithms 1 and 2 to identify J, prove their consistency under an additional orthogonality and separation condition (Theorems 4.1 and 4.2), and support the theory with simulations and a single-cell data application.

Significance. Should the deferred proofs be correct, the paper makes a substantial theoretical contribution: it provides minimax rates for shared singular subspace estimation across multiple noisy matrices, identifies dimension- and SNR-dependent phase transitions, and demonstrates that Stack-SVD is rate-optimal under full sharing while a popular Average-SVD alternative can be suboptimal. The matching rate forms in Theorems 2.1 and 3.1, together with simulation results that track the predicted phase boundaries, are encouraging internal evidence. The proposed tracing algorithm is simple and shows promising empirical performance on simulated and single-cell data. However, the central minimax-optimality claim under partial sharing is established only for an oracle estimator that knows J, not for the algorithm that estimates J, and the consistency guarantees for the algorithm are proven under assumptions stronger than those defining the minimax parameter spaces. These gaps are acknowledged in Section 7 but not resolved, and they materially limit the force of the abstract's claims.

major comments (5)
  1. [Section 4 / Theorem 3.1 / Theorem 4.1] The minimax optimality results for the partial-sharing models, Theorems 3.1 and 3.5, are stated for the oracle estimator \hat U_r^J, which uses the true index set J of shared singular vectors in the stacked SVD. The practical procedure of Section 4 replaces J by the output \hat J of Algorithm 1, but no theorem in the paper bounds the risk of \hat U_r^{\hat J}. Theorem 4.1 only gives P(\hat J = J) → 1 under conditions that are stronger than the defining conditions of H_{r,t} and S_{r,t}, and it does not combine this consistency event with the risk bounds (9)-(12). Consequently, the abstract's claim that the proposed methods are proven minimax rate-optimal under partial sharing is not supported for the estimator a user would actually run.
  2. [Section 4, Theorem 4.1] The assumptions of Theorem 4.1 require every singular value of the stacked signal matrix to be separated (σ_k ≥ (1+δ)σ_{k+1}) and require the minimum matrix signal strengths to satisfy (α^{c1} ∧ β^{c1}) ≥ Cn and (α∧β)^2 ≥ C2(p1∨p2). These conditions are global, while the minimax parameter spaces H_{r,t} and S_{r,t} impose only local gap conditions at type-switch positions and a lower bound on the minimum gap over switches. The theorem therefore proves consistency of the tracing algorithm only on a sub-regime of the parameter spaces used for the minimax results, leaving open whether the algorithm's success probability is high enough to preserve the rate (9) when the signal is near the boundary of H_{r,t} or S_{r,t}.
  3. [Section 3.3 / Section 7] Theorem 3.5 extends the oracle minimax bounds to non-orthogonal unshared subspaces (parameter space S_{r,t}), but the tracing algorithm's consistency (Theorems 4.1 and 4.2) is proved only under U_{1*}^T U_{2*} = 0. The paper's own Section 7 states that for pronounced non-orthogonality the algorithm may return an over-inclusive set with J ⊂ \hat J and that extending the algorithm is future work. Thus the paper does not establish minimax optimality for the practical estimator outside the orthogonal-unshared setting, and the abstract's broad phrase 'under partial sharing' overstates the proven scope.
  4. [Section 2.1, Theorem 2.1] The lower bounds (7)-(8) are stated under the additional conditions p1 ≍ p2 or γ ≳ τ^2(p1+p2), whereas the upper bounds (5)-(6) hold on the full class F_{r,γ}. The manuscript does not provide a lower bound for the regime with p1 and p2 of different orders and moderate γ, so the phrase 'Stack-SVD achieves minimax rate-optimality when the true singular subspaces are identical' in the abstract is not true on the full parameter space defined in Section 2.1. Section 7 acknowledges the gap, but the abstract and the summary bullet points should be qualified accordingly.
  5. [Supplementary Material / Throughout] All proofs are deferred to a supplementary file that is not included with the submission; the main text refers to 'Section S1.4' after Theorem 2.1, to 'Section S1.1' after Theorem 3.1, and Theorem 4.1 is stated without a proof sketch. Because the contributions are primarily theoretical, the absence of the supplement leaves the central lower-bound and consistency arguments unverifiable in the submitted version. The authors should provide the supplement for review.
minor comments (6)
  1. [Section 4, Algorithm 1] In step 8, the phrase 'Take the last r1−k1 and r2−k2 index sets' is inconsistent with the subsequent description that these are the indices of the smallest values of d1i and d2j; please clarify whether the selection is based on the smallest or the largest distances.
  2. [Section 4, Theorem 4.1] The symbols α, β, γ are reused with different meanings: here α = σ_min(X1), β = σ_min(X2), and γ = α ∧ β, whereas γ is the signal-strength parameter in F_{r,γ} and in the discussion following Theorem 2.1; this notational collision should be fixed.
  3. [Section 3.3, Theorem 3.4] The term 'column singular matrix' is not standard and is not defined in the text; also, the displayed SVD expressions would be much easier to check if the dimensions of S, U*, Σ*, and V* were stated explicitly.
  4. [Section 2.1, after Theorem 2.1] The text announces a novel lower-bound argument but gives no outline of it in the main text; given that the lower bound is a central technical novelty, the authors should include an informal description of the construction in Section 2 or in the Introduction.
  5. [Throughout] There are several typos: 'staked' and 'the staked matrix' in Example 3 and Section 4; 'A interesting direction' in Section 7; 'the the singular value' in the paragraph after Theorem 3.1; and in the references, [25] contains a stray '6' in 'The Annals of statistics 47 6 3009-3031'.
  6. [Figure 1] The axes and constants in the phase diagrams are not defined in the caption or the main text; please specify what is plotted (e.g., log SNR versus dimension ratios) so that the claimed regions can be checked.

Circularity Check

0 steps flagged · score 0.0 of 10

The paper's minimax upper and lower bounds are derived over the same parameter spaces with independent arguments; the only noted gaps are scope limitations in Section 7, not circular reductions.

full rationale

The derivation chain is self-contained with respect to the claimed rates. Theorem 2.1's upper bound is obtained by applying an external perturbation bound (Cai and Zhang, 2018) to the stacked signal matrix, whose r-th singular value is controlled by the parameter gamma^2 = min_i {sigma_(i)^2(X1)+sigma_(i)^2(X2)}; the minimax lower bound is constructed over the same parameter space F_{r,gamma} by a separate argument, not by reusing the Stack-SVD estimator's risk. The partial-sharing results in Theorems 3.1-3.3 and 3.5 likewise pair upper bounds for oracle or modified Stack-SVD estimators with independent lower bounds over H_{r,t}, H_(r,t)^(1), H_(r,t)^(2), and S_{r,t}; the rate n(t^2+p1+p2)/t^4 is driven by the parameter t (minimum eigen-gap) rather than by any fitted constant, so no input-output identity is present. The practical tracing algorithms in Section 4 are analyzed as consistency results for recovering J, with the orthogonality and singular-value-separation assumptions stated explicitly. Section 7 concedes that the algorithm is analyzed only for orthogonal unshared vectors and that pronounced non-orthogonality may yield J subset of Jhat; this is a limitation of scope, not a circular step, because the minimax statements are for the oracle estimators and the specific estimators analyzed. Self-citations such as Cai, Li, and Ma (2021) are used as background for the matrix denoising model and do not supply the load-bearing rate-optimality arguments. No equation is defined in terms of the quantity it predicts, and no fitted parameter is renamed as a prediction.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No free parameters are fitted in the theoretical sections. The analysis is parameterized by true signal strengths and eigen-gaps; these are properties of the unknown signal matrices, not calibration constants. The main extra assumptions beyond the model are orthogonality of unshared subspaces, eigen-gap separation, and known ranks. No new model entities such as forces, particles, or dimensions are introduced; 'vector type switch' and 'eigen-gap' are formal definitions, not invented entities.

assumptions (5)
  • domain assumption Noise entries are independent, zero-mean sub-Gaussian with variance proxy tau (model (1), class G_tau).
    Every theorem is stated for this noise class; rates scale with tau only through universal constants. Other noise structures such as heteroskedastic or dependent entries are outside scope.
  • domain assumption The shared and unshared left singular subspaces satisfy stated orthogonality: U1*^T U2* = 0 in H_{r,t}, H^(1), H^(2), and for the algorithm in Section 4.
    Proposition 1 and Theorems 3.1-3.3 rely on this; the non-orthogonal case is handled only for the oracle estimator (Theorems 3.4-3.5), and the algorithm's extension is heuristic.
  • domain assumption At every vector-type switch in the stacked matrix, the eigen-gap satisfies g^2_s(X) > c sigma^2_{s+1}(X), and the minimum gap is at least t^2.
    The upper and lower bounds in Section 3 and the consistency of the tracing algorithm require this separation; the paper notes non-identifiability when gaps are small (Section S1.1).
  • domain assumption The ranks r, r1*, r2* are known, and the index set J is known for the oracle results; Algorithm 2 estimates k1 and k2 only under strong signal conditions.
    Oracle optimality is conditional on knowing where shared vectors sit; the practical algorithm needs ranks as input, and no rank-selection guarantee is given.
  • standard math Standard singular-subspace perturbation bounds, e.g., Cai and Zhang (2018), are valid and used to derive Proposition 2.
    The upper bounds inherit the assumptions of these external bounds, including spectral gaps and sub-Gaussian noise.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Optimal Estimation of Shared Singular Subspaces across Multiple Noisy Matrices." pith.science (2026). https://pith.science/paper/OW2RF4NC

@misc{pith2026241117054,
  author       = {Pith},
  title        = {Pith review of: Optimal Estimation of Shared Singular Subspaces across Multiple Noisy Matrices},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/OW2RF4NC}},
  note         = {Machine review of arXiv:2411.17054}
}
read the original abstract

Estimating singular subspaces from noisy matrices is a fundamental problem with wide-ranging applications across various fields. Driven by the challenges of data integration and multi-view analysis, this study focuses on estimating shared singular subspaces across multiple matrices within a low-rank matrix denoising framework. A common approach for this task is to perform singular value decomposition on the stacked matrix (Stack-SVD), which concatenates all the matrices. We establish that Stack-SVD achieves minimax rate-optimality when the true singular subspaces of the noisy matrices are identical, whereas a popular alternative approach based on SVD of concatenated singular vector matrices (Average-SVD) can be sub-optimal. We then tackle the more complex scenario where the true singular subspaces are only partially shared across matrices. For various cases of partial sharing, we rigorously characterize the conditions under which Stack-SVD remains effective, achieves minimax optimality, or fails to deliver consistent estimates, offering theoretical insights into its practical applicability. To address the limitations of Stack-SVD in scenarios with partial sharing, we propose novel estimators and an efficient algorithm designed to identify both shared and unshared singular vectors. We further prove that these methods attain minimax rate-optimality under partial sharing. Extensive simulations and real-world data applications demonstrate the advantages of our proposed approach.

Figures

Figures reproduced from arXiv: 2411.17054 by the authors.

Figure 1
Figure 1. Phase diagrams when p1 ≍ p2 ≫ n (Left) and when p1 ≫ p2 ≫ n (Right). Each region is labeled by the name of consistent estimators: "Xi" indicates Uˆ(i) is consistent, "Stack" indicates Uˆ is consistent, "Stack(opt)" indicates Uˆ is minimax rate-optimal, and "impossible" indicates no consistent estimator exists. 2.3. Extension to multiple matrices. So far, our discussion has been primarily focusing on two matrices sha… view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 4 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Stacked SVD or SVD stacked? A Random Matrix Theory perspective on data integration

    stat.ML 2025-07 accept novelty 8.0 of 10

    In the proportional high-dimensional limit, the paper derives exact squared-overlap formulas and phase transitions for Stack-SVD and SVD-Stack, and proves optimally weighted Stack-SVD always beats optimally weighted S...

  2. Transfer Learning in High-Dimensional Clustering: Minimax Thresholds and Applications in Single-Cell Data

    math.ST 2026-07 conditional novelty 7.5 of 10

    In high-d two-community GMMs, consistent target clustering via transfer is possible iff either the target SNR clears the usual (d/n)^{1/4} barrier or the source is strong and aligned enough that µ∆_T, ∆_S, and µ∆_S∆_T...

  3. Spectral Joint Subspace Estimation for Heterogeneous Multi-View Data: Geometry and Reweighting

    math.ST 2025-12 conditional novelty 7.0 of 10

    Geometry of the individual components, not the number of views, determines whether AJIVE's error vanishes at the K^{-1/2} rate; a weighted variant handles heterogeneous views.

  4. Estimating shared subspace with AJIVE: the power and limitation of multiple data matrices

    stat.ML 2025-01 conditional novelty 7.0 of 10

    In high-SNR settings AJIVE's shared-subspace error is minimax-optimal and decays like 1/√K in the number of matrices, while in low-SNR settings a non-diminishing error floor appears even for an oracle-aided spectral e...

Reference graph

Works this paper leans on

72 extracted references · 66 canonical work pages · cited by 4 Pith papers

  1. [1]

    and W ANG , K

    A BBE , E., F AN, J. and W ANG , K. (2022). An ℓp theory of PCA and spectral clustering. The Annals of Statistics 50 2359–2385

    work page 2022

  2. [2]

    and P RIEBE , C

    A GTERBERG , J., L UBBERTS , Z. and P RIEBE , C. E. (2022). Entrywise estimation of singular vectors of low-rank matrices with heteroskedasticity and dependence. IEEE Transactions on Information Theory 68 4618–4650

    work page 2022

  3. [3]

    A PARICIO , L., B ORDYUH , M., B LUMBERG , A. J. and R ABADAN , R. (2020). A random matrix theory approach to denoise single-cell data. Patterns 1

    work page 2020

  4. [4]

    A RGELAGUET , R., A RNOL , D., B REDIKHIN , D., D ELORO , Y., V ELTEN , B., M ARIONI , J. C. and S TE- GLE , O. (2020). MOFA+: a statistical framework for comprehensive integration of multi-modal single- cell data. Genome biology 21 1–17

    work page 2020

  5. [5]

    and Z HOU , W

    B AO, Z., H U, J., P AN, G. and Z HOU , W. (2019). Canonical correlation coefficients of high-dimensional Gaussian vectors: Finite rank case. The Annals of Statistics 47 612–640

    work page 2019

  6. [6]

    and N ADAKUDITI , R

    B ENAYCH -GEORGES , F. and N ADAKUDITI , R. R. (2012). The singular values and vectors of low rank perturbations of large rectangular random matrices. Journal of Multivariate Analysis 111 120–135

    work page 2012

  7. [7]

    and P FEIFFER , R

    B URA , E. and P FEIFFER , R. (2008). On the distribution of the left singular vectors of a random matrix and its applications. Statistics & Probability Letters 78 2275–2280

    work page 2008

  8. [8]

    and G ORIN , V

    B YKHOVSKAYA , A. and G ORIN , V. (2023). High-dimensional canonical correlation analysis. arXiv preprint arXiv:2306.16393

    arXiv 2023

Show all 72 references
  1. [9]

    C AI, C., L I, G., C HI, Y., P OOR , H. V. and C HEN , Y. (2021). Subspace estimation from unbalanced and incomplete data matrices: L2 → ∞statistical guarantees. The Annals of Statistics 49 944–967

    2021

  2. [10]

    and M A, R

    C AI, T., LI, H. and M A, R. (2021). Optimal Structured Principal Subspace Estimation: Metric Entropy and Minimax Rates. Journal of Machine Learning Research 22 1-45

    2021

  3. [11]

    C AI, T. T. and Z HANG , A. (2018). Rate-optimal perturbation bounds for singular subspaces with applica- tions to high-dimensional statistics. The Annals of Statistics 46 60 – 89

    2018

  4. [12]

    and H ARABASZ , J

    C ALI ´NSKI , T. and H ARABASZ , J. (1974). A dendrite method for cluster analysis. Communications in Statistics-theory and Methods 3 1–27

    1974

  5. [13]

    J., S ING -L ONG , C

    C ANDÈS , E. J., S ING -L ONG , C. A. and T RZASKO , J. D. (2013). Unbiased Risk Estimates for Singular Value Thresholding and Spectral Estimators.IEEE Transactions on Signal Processing 61 4643-4657

    2013

  6. [14]

    and P RIEBE , C

    C APE , J., T ANG , M. and P RIEBE , C. E. (2019). The two-to-infinity norm and singular subspace geometry with applications to high-dimensional statistics. The Annals of Statistics 47 2405–2439

    2019

  7. [15]

    and F ÉRAL , D

    C APITAINE , M., D ONATI -MARTIN , C. and F ÉRAL , D. (2009). The largest eigenvalues of finite rank de- formation of large Wigner matrices: Convergence and nonuniversality of the fluctuations. Annals of probability: An official journal of the Institute of Mathematical Statist...

    2009

  8. [16]

    D., L I, H

    C HEN , X., L EE, J. D., L I, H. and YANG , Y. (2022). Distributed estimation for principal component analy- sis: An enlarged eigenspace analysis.Journal of the American Statistical Association117 1775–1786

    2022

  9. [17]

    C HEN , Y., CHI, Y., FAN, J., M A, C. et al. (2021). Spectral methods for data science: A statistical perspec- tive. Foundations and Trends® in Machine Learning14 566–806. 24

    2021

  10. [18]

    C HOI , S. G. and J UNG , S. (2022). Integrative decomposition of multi-source data by identifying partially- joint score subspaces. arXiv preprint arXiv:2203.14041

    2022 arXiv

  11. [19]

    and K AHAN , W

    D AVIS, C. and K AHAN , W. M. (1970). The rotation of eigenvectors by a perturbation. III. SIAM Journal on Numerical Analysis 7 1–46

    1970

  12. [20]

    K., K OWALCZYK , M

    D ING , J., A DICONIS , X., S IMMONS , S. K., K OWALCZYK , M. S., H ESSION , C. C., M ARJANOVIC , N. D., HUGHES , T. K., W ADSWORTH , M. H., B URKS , T., N GUYEN , L. T. et al. (2019). Systematic com- parative analysis of single cell RNA-sequencing methods. BioRxiv 632216

    2019

  13. [21]

    K., K OWALCZYK , M

    D ING , J., A DICONIS , X., S IMMONS , S. K., K OWALCZYK , M. S., H ESSION , C. C., M ARJANOVIC , N. D., HUGHES , T. K., WADSWORTH , M. H., B URKS , T., NGUYEN , L. T. et al. (2020). Systematic compar- ison of single-cell and single-nucleus RNA-sequencing methods. Nature biote...

    2020

  14. [22]

    and WANG , Z

    D ING , X., H U, Y. and WANG , Z. (2024). Two sample test for covariance matrices in ultra-high dimension. Journal of the American Statistical Association just-accepted 1–22

    2024

  15. [23]

    D ONOHO , D. L. and G AVISH , M. (2013). Minimax risk of matrix denoising by singular value thresholding. The Annals of Statistics 42 2413-2440

    2013

  16. [24]

    M., M IRCEA , M., M UELLER , N

    E RASLAN , G., S IMON , L. M., M IRCEA , M., M UELLER , N. S. and T HEIS , F. J. (2019). Single-cell RNA- seq denoising using a deep count autoencoder. Nature communications 10 390

    2019

  17. [25]

    and Z HU, Z

    F AN, J., W ANG , D., W ANG , K. and Z HU, Z. (2017). Distributed estimation of principal eigenspaces. The Annals of statistics 47 6 3009-3031

    2017

  18. [26]

    and MING ZHANG , C

    F AN, L., M ENG , R., G UO, Q., S HI, M. and MING ZHANG , C. (2019). Image denoising by low-rank approximation with estimation of noise energy distribution in SVD domain. IET Image Process. 13 680-691

    2019

  19. [27]

    and J OHNSTONE , I

    F AN, Z. and J OHNSTONE , I. M. (2022). Tracy–Widom at each edge of real covariance and MANOV A estimators. The Annals of Applied Probability 32

    2022

  20. [28]

    and M ARRON , J

    F ENG , Q., J IANG , M., H ANNIG , J. and M ARRON , J. (2018). Angle-based joint and individual variation explained. Journal of multivariate analysis 166 241–265

    2018

  21. [29]

    A., H O, Y.-L., O STROUCHOV , G., X U, Z., S HEN , S., XIONG , X., G RECO , K

    G AN, Z., Z HOU , D., R USH , E., P ANICKAN , V. A., H O, Y.-L., O STROUCHOV , G., X U, Z., S HEN , S., XIONG , X., G RECO , K. F. et al. (2023). Arch: Large-scale knowledge graph via aggregated narrative codified health records analysis. medRxiv

    2023

  22. [30]

    G ARTHWAITE , P. H. (1994). An interpretation of partial least squares. Journal of the American Statistical Association 89 122–127

    1994

  23. [31]

    and D ONOHO , D

    G AVISH , M. and D ONOHO , D. L. (2014). The Optimal Hard Threshold for Singular Values is 4/ √

    2014

  24. [32]

    IEEE Transactions on Information Theory 60 5040-5053

  25. [33]

    and L I, G

    G AYNANOVA , I. and L I, G. (2017). Structural learning and integrative decomposition of multi-view data. Biometrics 75 1121 - 1132

    2017

  26. [34]

    and Z HANG , A

    H AN, R., W ILLETT , R. and Z HANG , A. R. (2022). An optimal statistical and computational framework for generalized tensor estimation. The Annals of Statistics 50 1–29

    2022

  27. [35]

    L., B RYSON , B

    H IE, B. L., B RYSON , B. D. and B ERGER , B. (2019). Efficient integration of heterogeneous single-cell transcriptomes using Scanorama. Nature Biotechnology 37 685 - 691

    2019

  28. [36]

    and L U, M

    J IN, S., Z HANG , Y. and L U, M. (2022). Heterogeneous Adaptive Denoising Networks for Recommenda- tion. In International Conference on Neural Computing for Advanced Applications 30–43. Springer

    2022

  29. [37]

    J OHNSTONE , I. M. (2008). Multivariate analysis and Jacobi ensembles: Largest eigenvalue, Tracy–Widom limits and rates of convergence. Annals of statistics 36 2638

    2008

  30. [38]

    J OHNSTONE , I. M. and O NATSKI , A. (2020). Testing in High-Dimensional Spiked Models. The Annals of Statistics 48 1231–1254

    2020

  31. [39]

    and K RISHNASWAMY , S

    L EONE , S., S UN, X., P ERLMUTTER , M. and K RISHNASWAMY , S. (2024). Bayesian Spectral Graph De- noising with Smoothness Prior. 2024 58th Annual Conference on Information Sciences and Systems (CISS) 1-6

    2024

  32. [40]

    F., H OADLEY , K

    L OCK , E. F., H OADLEY , K. A., M ARRON , J. S. and N OBEL , A. B. (2013). Joint and individual variation explained (JIVE) for integrated analysis of multiple data types.The annals of applied statistics 7 523

    2013

  33. [41]

    L ÖFFLER , M., Z HANG , A. Y. and Z HOU , H. H. (2021). Optimality of spectral clustering in the Gaussian mixture model. The Annals of Statistics 49 2506–2530

    2021

  34. [42]

    and L I, H

    M A, R., T ONY CAI, T. and L I, H. (2021). Optimal permutation recovery in permuted monotone matrix model. Journal of the American Statistical Association 116 1358–1372

    2021

  35. [43]

    and YANG , F

    M A, Z. and YANG , F. (2023). Sample canonical correlation coefficients of high-dimensional random vectors with finite rank correlations. Bernoulli 29 1905–1932

    2023

  36. [44]

    D., R AGHAVAN, P

    M ANNING , C. D., R AGHAVAN, P. and S CHÜTZE , H. (2008). Dictionaries and tolerant retrieval In Intro- duction to Information Retrieval 45–60. Cambridge University Press

    2008

  37. [45]

    J., Z HANG , Z., G UO, Y

    M URDEN , R. J., Z HANG , Z., G UO, Y. and RISK , B. B. (2022). Interpretive JIVE: Connections with CCA and an application to brain connectivity. Frontiers in Neuroscience16 969510. OPTIMAL ESTIMATION OF SHARED SINGULAR SUBSPACES 25

    2022

  38. [46]

    and W ANG , K

    O’R OURKE , S., V U, V. and W ANG , K. (2018). Random perturbation of low rank matrices: Improving classical bounds. Linear Algebra and its Applications 540 26-59

    2018

  39. [47]

    P ALZER , E. F. (2022). Multi-source Data Decomposition and Prediction for Various Data Types, PhD thesis, University of Minnesota

    2022

  40. [48]

    F., W ENDT , C

    P ALZER , E. F., W ENDT , C. H., B OWLER , R. P., H ERSH , C. P., S AFO , S. E. and L OCK , E. F. (2022). sjive: Supervised joint and individual variation explained. Computational statistics & data analysis 175 107547

    2022

  41. [49]

    and G HOSH , A

    P ONZI , E., T HORESEN , M. and G HOSH , A. (2021). RaJIVE: Robust Angle Based JIVE for Integrating Noisy Multi-Source Data. arXiv preprint arXiv:2101.09110

    2021 arXiv

  42. [50]

    and M ARRON , J

    P ROTHERO , J., J IANG , M., H ANNIG , J., T RAN -DINH , Q., A CKERMAN , A. and M ARRON , J. (2024). Data integration via analysis of subspaces (DIV AS).TEST 1–42

    2024

  43. [51]

    R OUSSEEUW , P. J. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics 20 53-65

    1987

  44. [52]

    and V ILLAVICENCIO , A

    S ALLE , A., I DIART , M. and V ILLAVICENCIO , A. (2016). Matrix factorization using window sampling and negative sampling for improved word representations. arXiv preprint arXiv:1606.00819

    2016 arXiv

  45. [53]

    and GAYNANOVA , I

    S ERGAZINOV , R., TAEB , A. and GAYNANOVA , I. (2024). A spectral method for multi-view subspace learn- ing using the product of projections. arXiv preprint arXiv:2410.19125

    2024 arXiv

  46. [54]

    S HABALIN , A. A. and N OBEL , A. B. (2013). Reconstruction of a low-rank matrix in the presence of Gaussian noise. Journal of Multivariate Analysis 118 67-76

    2013

  47. [55]

    and M O, Q

    S HEN , R., W ANG , S. and M O, Q. (2013). Sparse integrative clustering of multiple omics data sets. The annals of applied statistics 7 269

    2013

  48. [56]

    S HU, H. (2024). D-CDLF: Decomposition of Common and Distinctive Latent Factors for Multi-view High- dimensional Data. arXiv preprint arXiv:2407.00730

    2024 arXiv

  49. [57]

    S TEWART, G. (1990). Perturbation theory for the singular value decomposition. svd and signal processing, II: algorithms. Analysis and Applications 99-109

    1990

  50. [58]

    S TEWART, M. (2006). Perturbation of the SVD in the presence of small singular values.Linear Algebra and its Applications 419 53-77

    2006

  51. [59]

    V U, V. (2011). Singular vectors under random perturbation.Random Structures & Algorithms 39 526–538

    2011

  52. [60]

    and Z HANG , F

    W ANG , H., C EN, Y., H E, Z., H E, Z., Z HAO, R. and Z HANG , F. (2018). Reweighted Low-Rank Matrix Analysis With Structural Smoothness for Image Denoising. IEEE Transactions on Image Processing 27 1777-1792

    2018

  53. [61]

    and C HUA, T.-S

    W ANG , W., F ENG , F., H E, X., N IE, L. and C HUA, T.-S. (2021). Denoising implicit feedback for recom- mendation. In Proceedings of the 14th ACM international conference on web search and data mining 373–381

    2021

  54. [62]

    W EDIN , P.-Å. (1972). Perturbation bounds in connection with singular value decomposition.BIT Numerical Mathematics 12 99–111

    1972

  55. [63]

    W EYL , H. V. (1912). Das asymptotische Verteilungsgesetz der Eigenwerte linearer partieller Differential- gleichungen (mit einer Anwendung auf die Theorie der Hohlraumstrahlung). Mathematische Annalen 71 441-479

    1912

  56. [64]

    and ZHANG , H

    W U, W., WU, Z. and ZHANG , H. (2024). Weighted bilinear factorization of low-rank matrix with structural smoothness for image denoising. Multim. Syst. 30 41

    2024

  57. [65]

    and FAN, J

    Y AN, Y., C HEN , Y. and FAN, J. (2024). Inference for heteroskedastic PCA with missing data. The Annals of Statistics 52 729–756

    2024

  58. [66]

    Y I, S., W ONG , R. K. W. and G AYNANOVA , I. (2022). Hierarchical nuclear norm penalization for multi- view data integration. Biometrics 79 2933 - 2946

    2022

  59. [67]

    and S AMWORTH , R

    Y U, Y., W ANG , T. and S AMWORTH , R. J. (2015). A useful variant of the Davis—Kahan theorem for statisticians. Biometrika 102 315–323

    2015

  60. [68]

    and H AN, R

    Z HANG , A. and H AN, R. (2019). Optimal Sparse Singular Value Decomposition for High-Dimensional High-Order Data. Journal of the American Statistical Association 114 1708

    2019

  61. [69]

    and X IA, D

    Z HANG , A. and X IA, D. (2018). Tensor SVD: Statistical and computational limits. IEEE Transactions on Information Theory 64 7311–7338

    2018

  62. [70]

    R., C AI, T

    Z HANG , A. R., C AI, T. T. and W U, Y. (2022). Heteroskedastic PCA: Algorithm, optimality, and applica- tions. The Annals of Statistics 50 53–80

    2022

  63. [71]

    and TANG , M

    Z HENG , R. and TANG , M. (2022). Limit results for distributed estimation of invariant subspaces in multiple networks inference and PCA. arXiv preprint arXiv:2206.04306

    2022

  64. [72]

    and S AMWORTH , R

    Z HU, Z., W ANG , T. and S AMWORTH , R. J. (2022). High-dimensional principal component analysis with heterogeneous missingness. Journal of the Royal Statistical Society Series B: Statistical Methodology 84 2000–2031

    2022

Pith tools

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