Pith. sign in

REVIEW 1 major objections 5 minor 2 cited by

Federated PCA and Estimation for Spiked Covariance Matrices: Optimal Rates and Efficient Algorithm

T0 review · 1 major / 5 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 minimax optimal error of the central server in federated, differentially private PCA under a spiked covariance model is the harmonic mean of the minimax optimal errors of the local clients.

desk verdict The harmonic-mean rate formula and matrix van Trees inequality are real contributions, but the headline 'one good client' robustness claim is not backed by the proved upper bound, which requires every local client to be nontrivial. read the letter →

arxiv 2411.15660 v1 pith:2HTSA3HU submitted 2024-11-23 math.ST cs.ITmath.ITstat.MLstat.TH

Jingyang Li , T. Tony Cai , Dong Xia , Anru R. Zhang This is my paper · ORCID
classification math.STcs.ITmath.ITstat.MLstat.TH MSC 62H2562C2062G20
keywords federatedlearningdifferentialprivacyspikedcovariancemodelprincipalcomponentanalysisminimaxoptimalityharmonicmeanspectraldecompositionvanTreesinequality
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

This paper establishes the minimax optimal rates for federated PCA and covariance matrix estimation when data are spread across clients that each enforce their own differential-privacy budget. The central claim is that the server's optimal squared error is the harmonic mean of the local clients' minimax errors, namely $1/\sum_j \Psi_0^{-2}(n_j,\varepsilon_j,\delta_j)$ for the principal subspace and an analogous combination with $\Psi_1$ for eigenvalues. Because a harmonic mean is dominated by the smallest local errors and shrinks with the number of clients, the server can be consistent even if only one local client is consistent, and even if none is consistent provided there are enough clients. The paper proves achievability with a Gaussian-mechanism algorithm analyzed through three layers of spectral decomposition, and proves matching lower bounds with a matrix version of van Trees' inequality. If correct, this pins down the fundamental privacy-accuracy trade-off of distributed PCA and shows that federated aggregation itself is a source of statistical robustness.

What carries the argument

The object that carries the argument is the harmonic-mean identity $(\sum_j \Psi^{-2}(n_j,\varepsilon_j,\delta_j))^{-1}$, which replaces the best or average local rate and is what makes one good client enough. The algorithm side is a three-layer spectral decomposition: each client computes a sample spectral projector, adds Gaussian noise calibrated to the sensitivity of the empirical projector to get a differentially private projector, and the server averages these projectors with weights inversely proportional to each client's local error rate before extracting the top-$r$ subspace; eigenvalues are then estimated in a second round and combined similarly. The error analysis expands the final deviation into three composed Neumann series and controls the cross-client inner-product terms so that only the weighted harmonic-mean term survives. The lower bound is carried by a matrix version of van Trees' inequality, which bounds the average subspace error by a projection-gradient Fisher information ratio under privacy constraints.

What would settle it

Run the paper's federated setting with two clients, one whose local sample size makes it highly accurate and one whose sample size or privacy budget makes it inconsistent enough to violate the nontriviality condition $\tilde\Psi_0 < c_1\sqrt{r}$, then compare the server's subspace error to the harmonic-mean bound $1/(\Psi_0^{-2}(n_1,\varepsilon_1,\delta_1)+\Psi_0^{-2}(n_2,\varepsilon_2,\delta_2))$; if the observed error substantially exceeds the bound whenever the bad client's violation is strong, the achievability direction of the 'one good client' claim fails in that regime.

Watch

Extended reading notes

Core claim

Under the spiked covariance model $\Sigma = U\Lambda U^\top + \sigma^2 I$ with Gaussian data, the paper shows that federated $(\varepsilon,\delta)$-differentially private PCA at a central server achieves the (scaled) harmonic mean of the per-client minimax rates: $\inf_{\widehat U}\sup_\Sigma \mathbb{E}\|\widehat U\widehat U^\top - UU^\top\|_F^2 \asymp (\sum_{j=1}^m \Psi_0^{-2}(n_j,\varepsilon_j,\delta_j))^{-1} \wedge r$, up to logarithmic factors, where $\Psi_0(n_j,\varepsilon_j,\delta_j)$ is the minimax error rate of a single private client. The covariance estimator adds an eigenvalue-estimation term $\lambda^2(\sum_j \Psi_1^{-2})^{-1}$, and the two terms correspond to separate private estimation of eigenvectors and eigenvalues. The lower bound matches the upper bound, so the harmonic-mean formula is the true optimal rate. This makes the central estimator multiply robust: consistency is preserved as long as at least one local client is consistent, and the estimator can remain consistent even when all local clients are inconsistent but the number of clients grows.

Load-bearing premise

The achievability proof assumes every local client is non-trivial, meaning each local error rate satisfies $\tilde\Psi_0(n_j,\varepsilon_j,\delta_j) < c_1\sqrt{r}$; the advertised consistency with only one good client is supported by the lower bound but not by a proven upper bound when other clients violate this condition.

Editorial extensions

If this is right

  • In the homogeneous case $n_j \asymp n$, $\varepsilon_j \asymp \varepsilon$, the server matches the pooled-data oracle for the statistical term $pr/(mn)$, while the privacy-cost term $p^2r^2/(mn^2\varepsilon^2)$ shrinks as $m$ grows, so distributing data does not hurt statistical efficiency and reduces the total privacy cost.
  • Federated PCA is multiply robust: the central estimate is consistent whenever at least one local client's estimate is consistent.
  • Even when every local client is inconsistent, consistency can be recovered if the number of clients is large enough.
  • Setting $n_j=1$ recovers non-interactive local differential privacy, giving an LDP PCA rate of order $p^2/(m\varepsilon^2)$ for rank one, which the paper notes follows from its lower bound.
  • The matrix version of van Trees' inequality is a reusable tool for minimax lower bounds in other distributed private estimation problems.

Reading between the lines

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

  • The harmonic-mean phenomenon probably extends to other federated private estimation problems with inverse-variance aggregation and mean-zero cross-client noise; the paper's own discussion points to federated SVD, where the same rate formula should hold for left and right singular subspaces.
  • A testable refinement of the 'one good client' guarantee: the lower bound holds without per-client conditions, but the paper's achievability theorem requires every client to satisfy the nontriviality condition $\tilde\Psi_0(n_j,\varepsilon_j,\delta_j) < c_1\sqrt{r}$; whether the upper bound can be extended to arbitrarily bad clients is open.
  • The data-independent weights in the algorithm mean the method can be deployed without knowing the signal strength $\lambda$; a natural extension is to make the weights adaptive to an estimated $\lambda$ while preserving the same rate.
  • With total sample size fixed, the privacy term grows with $m$; this suggests a client-selection or subsampling rule that drops clients with very small $n_j\varepsilon_j^2$ could improve the finite-sample constant.
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

1 major / 5 minor

Summary. This paper studies federated PCA and covariance estimation for spiked covariance matrices under per-client differential privacy constraints. The central theoretical claim is that the central server's minimax rate is the (scaled) harmonic mean of the per-client minimax rates, with two advertised consequences: consistency of the federated estimator as long as one local estimator is consistent, and consistency even when all local estimators are inconsistent provided the number of clients is large. The authors prove matching upper and lower bounds, up to logarithmic factors and delta_j-dependence, and propose Algorithm 1, which uses three layers of spectral decomposition and Gaussian mechanisms. The lower bound is based on a matrix version of van Trees' inequality, which is a useful technical contribution in its own right. Simulations and a lung-cancer data application illustrate the algorithm's performance.

Significance. If the stated guarantees hold, the paper makes a solid contribution to the privacy-preserving federated learning literature. The harmonic-mean rate formula is elegant, the lower-bound technique is nonstandard and potentially reusable, and the three-layer spectral perturbation analysis is substantial. The paper also ships a clearly written proof appendix, and the homogeneous-case corollary (rates depending on mn and mn^2 epsilon^2) is a clean and plausible benchmark. However, two load-bearing issues need attention: the achievability theorem requires a per-client nontriviality condition that is stronger than the advertised 'one good client' robustness, and the privacy guarantee is stated only with high probability, which is not the standard worst-case (epsilon,delta)-DP guarantee.

major comments (1)
minor comments (5)
  1. [Throughout] There are several typos: 'Algoirthm 1' and 'achive' in Section 2, and 'Van Tree's inequality' should be 'van Trees' inequality'.
  2. [Section 2, Eq. (5)] The displayed definition of ePsi0 contains an awkward expression 'r sigma^2/lambda' after the first fraction; it should be checked against the intended form sqrt((sigma^2/lambda + sigma^4/lambda^2) * (...)).
  3. [Section 2, data-independent weights] The symbol omega_k is introduced in the displayed formula for the data-independent weights, but the surrounding text discusses w_k; the notation should be unified.
  4. [Section 1.1 and Section 3] The collection of estimators is denoted U_{n,epsilon,delta} in Section 1.1 and U(n,epsilon,delta) in Theorem 3; please use one notation throughout.
  5. [Section 4.1] The sentence about transmitting the top r left singular vectors of eUj eUj^T + Zj is incomplete and should be rephrased.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the harmonic-mean rate is proven by an aggregation bound plus an independent matrix van Trees lower bound; the abstract's one-good-client robustness claim exceeds Theorem 1's per-client condition, but that is a theorem-abstract gap, not circularity.

full rationale

The central rate claims (3)-(4) are not built from the target conclusion. Theorem 1 proves an upper bound of the form C2/(sum_j ePsi0^{-2}) ^ (2r) by analyzing three layers of spectral decomposition; the main work is controlling the cross-client terms E<Delta_k1,Delta_k2> in eq. (8), and the ePsi0's are local error benchmarks, not the central rate. The matching lower bound in Theorem 3 is derived from a matrix van Trees inequality whose Fisher-information analysis yields the n_j eps_j^2 privacy terms directly; it does not assume the harmonic-mean form. The local single-client rates in (2) are imported from Cai et al. (2024b), an overlapping-author preprint, and the algorithm's sensitivity bound is also taken from that work, but these are parameter-free single-client results whose assumptions do not include the federated target, so they are legitimate inputs rather than a circular reduction. The same holds for the spectral representation formula cited to Xia (2021). A separate concern, not a circularity, is that Theorem 1's condition (5) requires ePsi0(n_j,eps_j,delta_j)<c1 sqrt(r) for every j, so the abstract's 'as long as at least one local client provides consistent results' and 'even if some local estimators are inconsistent, provided there are enough clients' are not covered by the proved upper bound; the lower bound alone does not give an estimator. The paper itself flags a related SNR limitation for the LDP special case ('our proposed estimator from Algorithm 1 will require a strong signal-to-noise ratio condition as stated in Theorem 1', Section 1.2). These are robustness/scope gaps, not self-referential derivations.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The paper's central contribution, the harmonic mean rate, is built from the single-client DP-PCA rates of Cai et al. (2024b), from the spectral perturbation tools of Xia (2021), and from the Gaussian spiked covariance model with known rank and noise level. The matrix van Trees inequality is derived self-consistently under a Gaussian prior. No free parameters are fitted to data in the theoretical analysis; the algorithm's weights are data-independent or depend on known quantities, except that the covariance weights v_j in Theorem 2 depend on the unknown lambda.

assumptions (6)
  • domain assumption Spiked covariance model: Sigma = U Lambda U^T + sigma^2 I with eigenvalues all comparable to a common scale lambda.
    This is the model throughout the paper, defined in equation (1) and the parameter set Theta(lambda, sigma^2).
  • domain assumption Gaussian data distribution: each local client's observations are i.i.d. N(0, Sigma).
    Assumed in the problem formulation and used throughout the proofs for Wishart-type concentration and independence structure. The discussion section notes extension to sub-Gaussian is future work.
  • domain assumption Known rank r and noise level sigma^2.
    Stated near the algorithm description: 'For simplicity, we assume that the rank r and the nuisance noise level sigma^2 are both known.' This is a strong assumption for practice.
  • domain assumption Single-client DP-PCA minimax rates Psi_0 and Psi_1 from Cai et al. (2024b).
    The paper relies on these rates and the sensitivity bounds of spectral projectors and eigenvalues from Cai et al. (2024b) as external building blocks. Any error in those rates would propagate to the harmonic mean formulas and to the lower bound.
  • standard math Spectral representation formulas from Xia (2021) for perturbed eigenprojectors.
    The proof of Theorem 1 uses these formulas to expand the three-layer spectral decomposition into Neumann-series-like terms, equations (9), (10), and (11).
  • domain assumption Condition on privacy parameters: max_j eps_j = O(1) and (rp + sqrt(r p n_j)) delta_j^0.9 <= c_1 n_j eps_j^2 for the lower bound.
    Theorem 3 imposes this condition to control the effect of delta_j on the Fisher information bound. The authors note the exponent 0.9 can be replaced by k/(k+1), so this is a mild but real restriction.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Federated PCA and Estimation for Spiked Covariance Matrices: Optimal Rates and Efficient Algorithm." pith.science (2026). https://pith.science/paper/2HTSA3HU

@misc{pith2026241115660,
  author       = {Pith},
  title        = {Pith review of: Federated PCA and Estimation for Spiked Covariance Matrices: Optimal Rates and Efficient Algorithm},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2HTSA3HU}},
  note         = {Machine review of arXiv:2411.15660}
}
read the original abstract

Federated Learning (FL) has gained significant recent attention in machine learning for its enhanced privacy and data security, making it indispensable in fields such as healthcare, finance, and personalized services. This paper investigates federated PCA and estimation for spiked covariance matrices under distributed differential privacy constraints. We establish minimax rates of convergence, with a key finding that the central server's optimal rate is the harmonic mean of the local clients' minimax rates. This guarantees consistent estimation at the central server as long as at least one local client provides consistent results. Notably, consistency is maintained even if some local estimators are inconsistent, provided there are enough clients. These findings highlight the robustness and scalability of FL for reliable statistical inference under privacy constraints. To establish minimax lower bounds, we derive a matrix version of van Trees' inequality, which is of independent interest. Furthermore, we propose an efficient algorithm that preserves differential privacy while achieving near-optimal rates at the central server, up to a logarithmic factor. We address significant technical challenges in analyzing this algorithm, which involves a three-layer spectral decomposition. Numerical performance of the proposed algorithm is investigated using both simulated and real data.

Figures

Figures reproduced from arXiv: 2411.15660 by the authors.

Figure 1
Figure 1. Numerical simulations comparing our method with existing methods and their variations. [PITH_FULL_IMAGE:figures/full_fig_p016_1.png] view at source ↗
Figure 2
Figure 2. We compare our proposed method with the equal-weight aggregation approach and the [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Layer by layer decomposition of ∥UbUb⊤ − UU ⊤∥ 2 F , with the building blocks {G(j) , H(j) , Zj,1 Zj,2 Zj,3}M j=1 Now we analyze each terms in [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

  1. A Van Trees Lower Bound for Fully Interactive Differentially Private Federated Learning

    cs.LG 2026-05 unverdicted novelty 8.0 of 10

    Under clientwise sample-level zCDP, the Fisher information of any fully interactive public federated transcript contracts to a sum of per-client privacy-vs-sample terms, yielding matching minimax rates for mean, linea...

  2. 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.

Reference graph

Works this paper leans on

20 extracted references · 12 canonical work pages · cited by 2 Pith papers

  1. [1]

    Upper bound forJ (p)

    So we have Z tr ∇ψ(Θ) ◦ ∇ψ(Θ)∗ p(Θ)dΘ = 2(p − r)r p − r − 1 ≥ 2r. Upper bound forJ (p) . Simple calculations show, for all k1, k2 ∈ [p], that ∇ψk1k2(Θ) = ¯Θ⊥ ¯Θ⊤ ⊥(ek1e⊤ k2 + ek2e⊤ k1) ¯ΘW −1/2, ∆ψk1k2 = 2[ ¯Θ⊥ ¯Θ⊤ ⊥]k1k2 · tr(W −1) − 2(p − r)[ ¯ΘW −1 ¯Θ⊤]k1k2. Since ∇p(Θ) = (2 π)−pr/2 exp(−∥Θ∥2 F/2)(−Θ), we have ⟨∇p, ∇ψk1k2⟩ = 0 for all k1, k2 ∈ [p]. As ...

    work page 1988

  2. [3]

    (40) Meanwhile (see Appendix C.2), ∇ψ(Θ)∗ : Rp×p → Rp×r is given by ∇ψ(Θ)∗(M ) = ¯Θ⊥ ¯Θ⊤ ⊥(M + M ⊤) ¯ΘW −1/2

    We will show (see Appendix C.1 for more details) ∇ψ(Θ) : Rp×r → Rp×p is the following linear map: ∇ψ(Θ)(Y ) = ¯Θ⊥ ¯Θ⊤ ⊥Y W−1/2 ¯Θ⊤ + ¯ΘW −1/2Y ⊤ ¯Θ⊥ ¯Θ⊤ ⊥. (40) Meanwhile (see Appendix C.2), ∇ψ(Θ)∗ : Rp×p → Rp×r is given by ∇ψ(Θ)∗(M ) = ¯Θ⊥ ¯Θ⊤ ⊥(M + M ⊤) ¯ΘW −1/2. (41) 48 Lower bound for R tr ∇ψ(Θ) ◦ ∇ψ(Θ)∗ p(Θ)dΘ . Based on (40) and (41), we have for al...

    work page 1988

  3. [6]

    Local privacy and statistical minimax rates

    John C Duchi, Michael I Jordan, and Martin J Wainwright. Local privacy and statistical minimax rates. In 2013 IEEE 54th annual symposium on foundations of computer science , pages 429–438. IEEE,

    work page 2013

  4. [7]

    Calibrating noise to sensitivity in private data analysis

    21 Cynthia Dwork, Frank McSherry, Kobbi Nissim, and Adam Smith. Calibrating noise to sensitivity in private data analysis. In Theory of Cryptography: Third Theory of Cryptography Conference, TCC 2006, New York, NY, USA, March 4-7,

    work page 2006

  5. [11]

    Asymptotics and concentration bounds for bilinear forms of spectral projectors of sample covariance

    22 Vladimir Koltchinskii and Karim Lounici. Asymptotics and concentration bounds for bilinear forms of spectral projectors of sample covariance. Annales de l’Institut Henri Poincar´ e, Probabilit´ es et Statistiques, 52(4):1976 – 2013,

    work page 1976

  6. [12]

    URL https://doi.org/10

    doi: 10.1214/15-AIHP705. URL https://doi.org/10. 1214/15-AIHP705. Vladimir Koltchinskii and Karim Lounici. Concentration inequalities and moment bounds for sample covariance operators. Bernoulli, pages 110–133,

    doi:10.1214/15-aihp705

  7. [13]

    John Novembre and Matthew Stephens

    doi: 10.1214/08-AOS618. John Novembre and Matthew Stephens. Interpreting principal component analyses of spatial pop- ulation genetic variation. Nature Genetics, 40(5):646–649,

    doi:10.1214/08-aos618

  8. [15]

    Using these facts, we have E X j1,··· ,jl∈[m] wj1 · · ·wjltr(U ⊤M (s1)M (s1)⊤∆j1M (s2) · · ·M (sl)⊤∆jlM (sl+1)M (sl+1)⊤U ) · 1(E) ≤ r · X j∈I2(s) uj1 · · ·ujl

    For each summand with index j ∈ I2(s), using Cauchy-Schwarz inequality, we have |tr(U ⊤M (s1)M (s1)⊤∆j1M (s2) · · ·M (sl)⊤∆jlM (sl+1)M (sl+1)⊤U )| ≤r · uj1 · · ·ujl. Using these facts, we have E X j1,··· ,jl∈[m] wj1 · · ·wjltr(U ⊤M (s1)M (s1)⊤∆j1M (s2) · · ·M (sl)⊤∆jlM (sl+1)M (sl+1)⊤U ) · 1(E) ≤ r · X j∈I2(s) uj1 · · ·ujl. Using the inclusion relation in...

    work page 2012

Show all 20 references
  1. [18]

    Recall G(j) 1 = E⟨∇ψ(Θ)(∇ log p(Dj; Θ| bUj)), ∇ψ(Θ)(∇ log p(X (j) 1 ; Θ| bUj))⟩

    We shall first consider E|G(j) 1 |k for some large and absolute integer k >0. Recall G(j) 1 = E⟨∇ψ(Θ)(∇ log p(Dj; Θ| bUj)), ∇ψ(Θ)(∇ log p(X (j) 1 ; Θ| bUj))⟩. By definition, we get E|G(j) 1 |k = E E⟨∇ψ(Θ)(∇ log p(Dj; Θ| bUj)), ∇ψ(Θ)(∇ log p(X (j) 1 ; Θ| bUj))⟩ k ≤ E ⟨∇ψ(Θ)(Sj)...

    2017

  2. [19]

    The term Etr(W −4) can be computed using the Theorem 4.1 of Von Rosen (1988), which implies Etr(W −4) · e−c1p/2 ≤ p−2

    ≤ 4p−2 + (E∥W −2∥2)1/2 · e−c1p/2 ≤ 4p−2 + (E∥W −2∥2 F)1/2 · e−c1p/2 = 4p−2 + (Etr(W −4))1/2 · e−c1p/2. The term Etr(W −4) can be computed using the Theorem 4.1 of Von Rosen (1988), which implies Etr(W −4) · e−c1p/2 ≤ p−2. Lower bound for covariance matrix estimation.We conside...

    1988

  3. [20]

    Lemma 5 (Koltchinskii and Lounici (2017))

    and we use the moment bound for Gaussian in the last inequality. Lemma 5 (Koltchinskii and Lounici (2017)) . Let X1, · · ·, Xn be i.i.d. samples from N (0, Σ), and bΣ = 1 n Pn i=1 XiX ⊤ i . Then E∥bΣ − Σ∥ ≍ ∥Σ∥ er n ∨ r er n , where er = tr(Σ) ∥Σ∥ is the effective rank of Σ. M...

    2017

  4. [2001]

    Principal component analysis: a review and recent developments

    Ian T Jolliffe and Jorge Cadima. Principal component analysis: a review and recent developments. Philosophical transactions of the royal society A: Mathematical, Physical and Engineering Sci- ences, 374(2065):20150202,

  5. [2006]

    The algorithmic foundations of differential privacy.Foundations and Trends® in Theoretical Computer Science , 9(3–4):211–407, 2014a

    Cynthia Dwork, Aaron Roth, et al. The algorithmic foundations of differential privacy.Foundations and Trends® in Theoretical Computer Science , 9(3–4):211–407, 2014a. Cynthia Dwork, Kunal Talwar, Abhradeep Thakurta, and Li Zhang. Analyze gauss: optimal bounds for privacy-prese...

    2014

  6. [2008]

    URL https://doi.org/10.1214/ 08-AOS600

    doi: 10.1214/08-AOS600. URL https://doi.org/10.1214/ 08-AOS600. Avrim Blum, Cynthia Dwork, Frank McSherry, and Kobbi Nissim. Practical privacy: the sulq framework. In Proceedings of the twenty-fourth ACM SIGMOD-SIGACT-SIGART symposium on Principles of database systems , pages 128–138,

    doi

  7. [2010]

    URL https://doi.org/10.1214/09-AOS752

    doi: 10.1214/09-AOS752. URL https://doi.org/10.1214/09-AOS752. T. Tony Cai, Zongming Ma, and Yihong Wu. Sparse PCA: Optimal rates and adaptive estimation. The Annals of Statistics , 41(6):3074 – 3110,

    doi

  8. [2013]

    URL https: //doi.org/10.1214/13-AOS1178

    doi: 10.1214/13-AOS1178. URL https: //doi.org/10.1214/13-AOS1178. 20 T. Tony Cai, Zhao Ren, and Harrison H. Zhou. Estimating structured high-dimensional covariance and precision matrices: Optimal rates and adaptive estimation. Electronic Journal of Statistics , 10(1):1 – 59,

    doi

  9. [2014]

    ISBN 9781450329576

    Association for Computing Machinery. ISBN 9781450329576. doi: 10.1145/2660267.2660348. URL https://doi.org/10.1145/2660267.2660348. Jianqing Fan, Yingying Fan, and Jinchi Lv. High dimensional covariance matrix estimation using a factor model. Journal of Econometrics , 147(1):186–197,

  10. [2016]

    URL https://doi.org/10.1214/15-EJS1081

    doi: 10.1214/15-EJS1081. URL https://doi.org/10.1214/15-EJS1081. T Tony Cai, Abhinav Chakraborty, and Lasse Vuursteen. Optimal federated learning for nonpara- metric regression with heterogeneous distributed differential privacy constraints. arXiv preprint arXiv:2406.06755, 20...

    arXiv

  11. [2018]

    URL https://doi.org/10.1214/17-AOS1541

    doi: 10.1214/17-AOS1541. URL https://doi.org/10.1214/17-AOS1541. T. Tony Cai, Cun-Hui Zhang, and Harrison H. Zhou. Optimal rates of convergence for covariance matrix estimation. The Annals of Statistics , 38(4):2118 – 2144,

    doi

  12. [2022]

    In the proof, we set λmax = λ1 and λmin = λr

    24 A Proofs A.1 Proof of Theorem 1 We first derive the upper bound for the expectation, and then derive the high probability upper bound. In the proof, we set λmax = λ1 and λmin = λr. Upper bound for expectation.We now derive the upper bound for E∥ bU bU ⊤ − U U⊤∥2 F. We will ...

    2021

Pith tools

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