Pith. sign in

REVIEW 1 major objections 8 minor 47 references

Spectral algorithm estimates causal bilinear forms directly from staggered tensor panels

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · glm-5.2

2026-07-08 09:27 UTC pith:W2QUWFW7

load-bearing objection Direct estimation of bilinear forms in Tucker2 tensor panels with staggered-adoption missingness — solid theory, clean phase transition result, one genuinely restrictive assumption. the 1 major comments →

arxiv 2607.06330 v1 pith:W2QUWFW7 submitted 2026-07-07 math.ST stat.MEstat.TH

Direct and efficient estimation of bilinear forms in staggered tensor panels

Alberto Bordino , Thomas B. Berrett , Olga Klopp This is my paper
classification math.ST stat.MEstat.TH MSC 62H1262H2562F12
keywords bilinear formsTucker2 decompositionstaggered adoptiontensor completionmissing not at randomspectral methodsminimax lower boundscausal inference
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper addresses the problem of estimating specific quantities—bilinear forms—from partially observed, noisy tensor data where the missingness follows a staggered adoption pattern, as arises in causal inference when units adopt a treatment at different times. The authors model the data using a Tucker2 decomposition, which assumes shared latent factors across tensor layers (e.g., multiple outcomes or policies) while allowing each layer to have its own core structure. Rather than completing the entire missing tensor and then computing the target quantity—a plug-in approach that may be statistically and computationally wasteful—the authors propose a spectral algorithm that targets the bilinear form directly. The key methodological idea is a regress-then-denoise procedure: the algorithm learns shared subspaces by pooling information across layers via stacked singular value decompositions, then uses a spectral inverse to predict the missing block and evaluate the desired functional without ever materializing the full missing matrix. The central theoretical result is a non-asymptotic mean squared error bound that exhibits a phase transition in the number of layers K. When K is small, the error decreases as 1/K because pooling improves estimation of the shared factors. When K is large, the error saturates at an irreducible layer-specific term that reflects the difficulty of estimating the slice-specific core. This elbow behavior is proven to be rate-optimal via local minimax lower bounds that match the upper bound up to constants. The construction is extended from a basic four-block missingness pattern to general staggered adoption designs through an anchored reduction that decomposes arbitrary staircase patterns into auxiliary four-block problems.

Core claim

The central discovery is that direct estimation of bilinear forms from partially observed tensor panels achieves a mean squared error rate that exhibits a provable phase transition in the number of layers K: pooling across layers yields 1/K decay for the shared-subspace estimation components, while an irreducible layer-specific term remains that cannot be reduced by pooling. This rate is matched by local minimax lower bounds, establishing optimality. The algorithm achieves this by combining stacked SVDs for subspace learning with a clipped spectral inverse for block prediction, targeting the functional directly rather than completing the tensor.

What carries the argument

The Tucker2 tensor model with shared unit and time factors across layers and slice-specific cores; the four-block missingness pattern as the building block for staggered adoption; the Stack-SVD procedure for pooling shared subspaces across layers; the clipped spectral inverse for stabilizing the regression operator; the anchored four-block reduction for extending to general staircase missingness; and the Haar compression bounds on the Stiefel manifold that yield sharper projected subspace perturbation estimates than standard Wedin-type bounds.

Load-bearing premise

The algorithm requires that the observed portion of the data within each layer is sufficiently well-conditioned relative to the latent factor structure—specifically, that the restricted Gram matrices of the shared factors over observed entries have eigenvalues bounded away from zero. This effectively requires the missing block to be small relative to the observed block, limiting applicability when the fraction of missing entries is large.

What would settle it

If the restricted Gram matrices U_1j^T U_1j and V_1j^T V_1j become singular or near-singular—which happens when the observed block is too small or the factor structure is poorly conditioned within a layer—the target bilinear form is not identifiable and no estimator can consistently recover it.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

1 major / 8 minor

Summary. This paper studies the estimation of bilinear forms from noisy, partially observed tensor data under a Tucker2 model with structured missingness motivated by staggered adoption designs. The authors propose a spectral algorithm that pools information across tensor layers and targets the functional directly, rather than completing the entire tensor. They prove a non-asymptotic mean squared error bound (Theorem 1) exhibiting a phase transition in the number of layers K, and match it with local minimax lower bounds (Theorems 2 and 3). The four-block missingness pattern is extended to general staggered adoption via an anchored reduction (Corollary 4). Simulations and two real-data applications validate the theory.

Significance. The paper addresses a well-motivated problem at the intersection of causal inference and tensor estimation. The direct targeting of bilinear forms, rather than full tensor completion, is a meaningful contribution with clear practical relevance. The key technical innovation is the Haar compression argument (Lemma 6), which yields a projected subspace error bound with a dimension-dependent compression factor that replaces the ambient dimension N; this is what enables the SNR condition to improve with K. The matching lower bounds via van Trees inequality are correctly constructed and establish rate-optimality up to constants, rank, and logarithmic factors. The extension to staggered adoption (Corollary 4) via the anchored four-block reduction is clean. The reproducible code and real-data applications add value. The paper ships falsifiable predictions (the phase transition in K) that are confirmed in simulation.

major comments (1)
  1. Theorem 1, condition (4): This condition requires that a certain tail-probability term is bounded by c_0 * Upsilon_xy. The authors argue this is satisfiable because the LHS involves exponentially small tail probabilities while the RHS involves polynomial terms. However, the condition couples tau, the bad-event probability, and Upsilon_xy in a non-trivial way. A brief explicit verification that (4) is compatible with the regime where the SNR condition (A3) holds and tau is chosen as in Theorem 1 (tau <= c_l * N_{1k} / (2N)) would strengthen the result. Currently the reader must take on faith that the polynomial-in-K Upsilon_xy always dominates the exponentially small tail terms; a short remark or a line in the proof of Theorem 1 checking this inequality case by case would help.
minor comments (8)
  1. Section 2.1, Assumption A1: The discussion of c_l > 0 is well-supported by Propositions 18-19, but the reader would benefit from a brief note in the main text (not just the appendix) on how restrictive this is in practice. Lemma 16 shows the missing block must be small relative to the observed block; a one-sentence pointer to this in Section 2.3 would contextualize the assumption.
  2. Table 1: The notation table is comprehensive but dense. Consider splitting into two tables (one for block/signal notation, one for empirical/SVD notation) or adding a brief reading guide so readers can quickly find definitions when reading the algorithm description.
  3. Section 5.1, Algorithm 2: The construction of anchor sets RowAnc_k(j) and ColAnc_k(j) is clearly defined, but a small worked example showing the sets for a concrete staircase (e.g., o_k = 4, a = 3, b = 3) would help readers verify their understanding of the index set computations.
  4. Section 6.2: The Castle Doctrine application uses r = 3. It would be useful to report sensitivity to the choice of r (e.g., r = 2, 4) for at least one functional, to assess robustness of the conclusions to rank misspecification in a real-data setting.
  5. Section 6.3: The COVID-19 application has N = 18, T = 32, K = 2. With only 2 fully observed rows in the target layer (Norway, Iceland), Assumption A1 requires c_l > 0 for the restricted Gram matrix of 2 rows. For r = 3, this is rank-deficient. The authors should clarify whether the auxiliary four-block construction resolves this (by pooling anchor rows from the other layer), or whether the theoretical guarantees are not plausibly satisfied here.
  6. Lemma 6, proof: The Haar-measure argument is the key technical step. While correct, the proof is quite involved. A brief high-level summary at the start of the proof (2-3 sentences) explaining the strategy—(1) reduce to controlling S = U_perp^T hat_U_left H_U, (2) show S lives in col(G_2), (3) apply Haar compression—would make the result more accessible.
  7. References: Several references cite 2026 dates (e.g., Agterberg 2026, Choi and Yuan 2026, Agarwal et al. 2026, Ma and Ma 2026, Liu et al. 2026). If these are forthcoming or revised versions, please update to the correct final citations.
  8. Appendix B.1, equation (7): The table of functionals is informative but the Trend row uses a linearity assumption on the row-averaged trajectory. This assumption is stated but not discussed in the main text; a brief remark in Section 6.2 on whether it is plausible for the Castle Doctrine data would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for a careful and constructive report. We address the single major comment below.

read point-by-point responses

  1. Referee: Theorem 1, condition (4): The condition requires that a certain tail-probability term is bounded by c_0 * Upsilon_xy. The authors argue this is satisfiable because the LHS involves exponentially small tail probabilities while the RHS involves polynomial terms. However, the condition couples tau, the bad-event probability, and Upsilon_xy in a non-trivial way. A brief explicit verification that (4) is compatible with the regime where the SNR condition (A3) holds and tau is chosen as in Theorem 1 (tau <= c_l * N_{1k} / (2N)) would strengthen the result. Currently the reader must take on faith that the polynomial-in-K Upsilon_xy always dominates the exponentially small tail terms; a short remark or a line in the proof of Theorem 1 checking this inequality case by case would help.

    Authors: We agree that an explicit verification of condition (4) would strengthen the presentation, and we will add it to the revised manuscript. Below we provide the argument that will appear as a remark following the statement of Theorem 1, with the detailed case-by-case verification in the proof. Recall that condition (4) requires two inequalities to hold simultaneously: (I) gamma_max^2 * tau^{-1} * (N_{1k}/N) * (p_N^{-10} + p_T^{-10}) <= c_0 * Upsilon_xy, and (II) sigma^2 * tau^{-1} * (N_{1k} + T) * (p_N^{-5} + p_T^{-5}) <= c_0 * Upsilon_xy. We verify each in turn under the regime where (A3) holds and tau = c_l * N_{1k} / (2N). For inequality (I): Substituting tau gives a LHS of order gamma_max^2 * (N/N_{1k}) * (p_N^{-10} + p_T^{-10}). The RHS contains the term sigma^2 * (r + zeta_N) / rho_N * ||U_{2k}^T x||^2. Using (A1), we have rho_N = N_{1,p}/N >= c_l * N_{1k}/N, so (N/N_{1k}) <= c_l^{-1} * rho_N^{-1} * N_{1k}/N * rho_N is not directly the right path; instead, note that (N/N_{1k}) * p_N^{-10} <= (N/N_{1k}) * (max{N_{1,p}, T})^{-10}. Since N_{1,p} >= N_{1k} (as it is a sum including N_{1k}), we have N/N_{1k} <= N/N_{1k} and p_N >= N_{1k}, so (N/N_{1k}) * p_N^{-10} <= N_{1k}^{-1} * N * N_{1k}^{-10} which is exponentially small. More precisely, under (A2) we have N - r >= c_blk * N so N is of the same order as the ambient dimension, while p_N = max{N_{1,p}, T} >= N_{1k}. The term (N/N_{1k}) * p_N^{-10} is thus of order N * N_{1k}^{-11}, which is polynomially (in fact, nearly exponentially) small in the dimensions. The RHS term sigma^2 * (r + zeta_N) / rho_N * ||U_{2k}^T x||^2 is of order sigma^2 * r * (N/N_{1,p}) * (r/N) (using (A4)) = sigma^2 * r^2 / N_{1,p}, which is polynomial in the dimensions and the SNR parameters. Under (A3), sigma is bounded by c_0 * gamma_min / max revision: no

  2. Referee: sqrt(N), sqrt(T), sqrt(N/rho_T), sqrt(N*T/N_{1k})}, so sigma^2 is at most of order gamma_min^2 / (N + T + N/rho_T + N*T/N_{1k}), which is polynomial. The exponentially decaying p_N^{-10} factor ensures that the LHS is negligible compared to any polynomial term appearing in Upsilon_xy. For inequality (II): Substituting tau gives a LHS of order sigma^2 * (N/N_{1k}) * (N_{1k} + T) * (p_N^{-5} + p_T^{-5}). Again, p_N >= N_{1k} and p_T >= T_{1k}, and under (A2) the dimensions N, T, N_{1k}, T_{1k} are all of the same polynomial order. The factor (p_N^{-5} + p_T^{-5}) decays polynomially (with exponent 5) in the dimensions, while the RHS of (4) contains terms like sigma^2 * (r + zeta_N) / rho_N * ||U_{2k}^T x||^2, which is of order sigma^2 * r * log(N_{1,p} + T) * (N/N_{1,p}) * (r/N) = sigma^2 * r^2 * log(N_{1,p} + T) / N_{1,p}. Under (A3), this is a polynomial quantity in the problemparameters

    Authors: Continuing the verification: the LHS of (II) is of order sigma^2 * (N/N_{1k}) * (N_{1k} + T) * p_N^{-5}, which simplifies to sigma^2 * N * (1 + T/N_{1k}) * p_N^{-5}. Since p_N = max{N_{1,p}, T} and N_{1,p} >= N_{1k}, we have p_N >= N_{1k}, so the LHS is at most sigma^2 * N * (1 + T/N_{1k}) * N_{1k}^{-5}. Under (A2), N, T, N_{1k} are all polynomially related, so the factor N * N_{1k}^{-5} decays at least as fast as N_{1k}^{-4} (using N <= c * N_{1k} * (N/N_{1k}) and the dimensional regularity conditions). Meanwhile, the RHS term sigma^2 * (r + zeta_N) / rho_N * ||U_{2k}^T x||^2 is of order sigma^2 * r * log(p_N) * (N/N_{1,p}) * (r/N) = sigma^2 * r^2 * log(p_N) / N_{1,p}, which is polynomially bounded away from zero under (A3) and (A4). The ratio LHS/RHS is thus of order (N * N_{1k}^{-5}) / (r^2 * log(p_N) / N_{1,p}) ~ N * N_{1,p} / (N_{1k}^5 * r^2 * log(p_N)), which goes to zero as the dimensions grow, since N_{1k}^5 grows polynomially while the numerator grows only linearly. The same argument applies to the p_T terms by symmetry. Therefore, both inequalities (I) and (II) are satisfied for sufficiently large dimensions, and the constant c_0 can be chosen to absorb the finite-dimensional cases. In summary: the key observation is that the LHS of (4) always contains a factor of the form p_N^{-5} or p_N^{-10} (or the analogous T terms), which decays polynomially in the dimensions, while the RHS Upsilon_xy contains terms that are polynomial in the SNR and dimensional parameters under (A3) and (A4). The polynomial decay of the tail probabilities always dominates the polynomial terms in Upsilon_xy for sufficiently large dimensions, and the constant c_0 > 0 can be chosen small enough to handle the transition regime. We will add a remark after Theorem 1 summarizing this argument, revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper's central results are derived from first principles without circular reductions. Theorem 1's upper bound arises from a spectral analysis of pooled matrices, with the error decomposition (Lemma 14) and Haar compression argument (Lemma 6) providing genuine, non-tautological bounds. The phase transition in K is a consequence of the error structure: pooling terms carry 1/ρ_N and 1/ρ_T factors (which grow with K) while the irreducible layer-specific term σ²(N/N_{1k})∥U_{2k}^T x∥²∥V_{2k}^T y∥² does not depend on K. The minimax lower bounds (Theorems 2 and 3) use independent parametric submodel constructions with van Trees inequality, and their matching to the upper bound is up to acknowledged rank and logarithmic factors. Self-citations (e.g., to Agterberg 2026 for shared-subspace context, or to Yan and Wainwright 2024 for the K=1 special case) provide background or comparison points rather than load-bearing premises that would make the derivation circular. Corollary 4 reduces to Theorem 1 by an explicit change of variables, which is a valid mathematical reduction, not a circularity. No fitted parameters are renamed as predictions, and no result is defined in terms of its own conclusion.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities or postulated objects. The Tucker2 model, SVD, and spectral methods are standard. The clipped spectral inverse is a methodological innovation, not an invented entity. All assumptions are standard domain assumptions for low-rank tensor models.

free parameters (3)
  • tau (clipping threshold) = 0.01 (simulations), 10^-2 to 10^-4 (real data)
    Tuning parameter for clipped spectral inverse. Theory requires tau <= c_l*N1k/(2N); simulations show results are stable across tau values, acting as safeguard rather than tuning parameter.
  • r (rank) = 3 (real data), 5-6 (simulations)
    The Tucker2 rank is chosen by the user. Not fitted to data in the theoretical analysis but must be specified for the algorithm.
  • c_l, c_u (conditioning constants) = Not numerically specified
    Constants bounding the restricted Gram matrices in Assumption A1. Assumed to satisfy 0 < c_l <= c_u < infinity but not estimated from data.
axioms (5)
  • domain assumption Tucker2 decomposition: M = C x_1 U x_2 V x_3 I_K with U^T U = V^T V = I_r
    Section 2.1. Assumes shared unit and time factors across layers with slice-specific cores. This is the central structural assumption.
  • domain assumption Gaussian noise: E_ijt ~ N(0, sigma^2) i.i.d.
    Section 2.1, equation (2). Used throughout proofs for concentration and Haar measure arguments.
  • domain assumption Assumption A1: restricted Gram matrices well-conditioned with c_l > 0
    Section 2.1. Necessary for identifiability (Propositions 18-19) but restricts missingness patterns to those where observed blocks retain all r latent directions.
  • domain assumption Assumption A3: signal-to-noise ratio theta <= c_0
    Section 2.1. Requires noise small relative to signal strength. Becomes less stringent as K grows but still needed.
  • domain assumption Assumption A4: incoherence parameters nu_x, nu_y of constant order
    Section 2.1. Standard incoherence condition on query vectors relative to factor matrices.

pith-pipeline@v1.1.0-glm · 77674 in / 2718 out tokens · 366247 ms · 2026-07-08T09:27:15.242672+00:00 · methodology

0 comments
read the original abstract

We study the estimation of bilinear forms from noisy, partially observed tensor data. The signal follows a Tucker2 model, with shared unit and time factors across tensor layers and slice-specific cores. The missingness pattern is structured and motivated by staggered adoption designs, which are common in causal inference and related applications. We first analyse the four-block missingness pattern, the basic building block for general staggered adoption, and propose a spectral algorithm that pools information across layers and targets the functional directly, rather than completing the entire tensor. We prove a non-asymptotic mean squared error bound that exhibits a phase transition in the number of layers, showing when pooling improves estimation, and match it with a local minimax lower bound up to constants. We then extend the construction to general staggered adoption designs via an anchored four-block reduction, and derive analogous theoretical guarantees. Finally, we validate our theoretical findings through experiments on both simulated and real-world datasets.

Figures

Figures reproduced from arXiv: 2607.06330 by Alberto Bordino, Olga Klopp, Thomas B. Berrett.

Figure 1
Figure 1. Figure 1: Layer-specific four-block structure for K = 3. Each slice Y•,•,j contains N1j fully observed rows and N2j rows that are observed only in the first T1j columns. The total number of rows and columns are denoted by N and T, respectively. We use a, b, c to denote the observed blocks and d to denote the missing block. a Tucker2 decomposition (Kolda and Bader, 2009, Section 4) of rank (r, r, K), meaning that the… view at source ↗
Figure 2
Figure 2. Figure 2: Layer-specific staggered adoption for K = 3, target layer k = 1, and target block (a, b) = (5, 4). Panel 1 orders the target layer by its own rows, yielding a staircase missingness pattern; the orange dotted rectangle identifies the target rows S1,5 and columns Q1,4 used to form auxiliary matrices. Panels 2–3 show non-target layers under the same target-layer ordering, with general staggered missingness. T… view at source ↗
Figure 3
Figure 3. Figure 3: (a) Mean squared error of the five estimators as a function of [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Mean absolute estimation error over 100 random bilinear queries for each target block. Algorithm [PITH_FULL_IMAGE:figures/full_fig_p019_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Runtime (left) and accuracy (right) comparison for Algorithms [PITH_FULL_IMAGE:figures/full_fig_p019_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Estimates of Ψ(h) 0 (2) using the pooled tensor estimator of Algorithm 3 and its matrix counterpart. Confi￾dence intervals are computed as in [PITH_FULL_IMAGE:figures/full_fig_p022_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Outcome tensor from the merged OxCGRT–OWID panel. Rows are countries, columns are dates from [PITH_FULL_IMAGE:figures/full_fig_p023_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Estimates of Ψ(h) 0 (1) and ∆(h) (1) using the pooled tensor estimator of Algorithm 3 and its matrix counter￾part. Confidence intervals are computed as in Section 6.2 with B = 500. gives a significantly negative model-based estimate. Under the maintained low-rank counterfactual model and the assumed interpretation of policy timing, this is consistent with a reduction in new deaths associated with stay-at-h… view at source ↗
Figure 9
Figure 9. Figure 9: Castle Doctrine state-year-crime tensor used in the real-data simulation. Panels show logged crime rates for [PITH_FULL_IMAGE:figures/full_fig_p038_9.png] view at source ↗

discussion (0)

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

Reference graph

Works this paper leans on

47 extracted references · 47 canonical work pages · 4 internal anchors

  1. [1]

    Journal of Economic Literature , year =

    Abadie, Alberto , title =. Journal of Economic Literature , year =

    work page

  2. [2]

    Proceedings of the 36th Annual Conference on Learning Theory , year =

    Anish Agarwal and Munther Dahleh and Devavrat Shah and Dennis Shen , title =. Proceedings of the 36th Annual Conference on Learning Theory , year =

    work page

  3. [3]

    Agarwal, Anish and Choi, Jungjun and Yuan, Ming , year =. Robust. doi:10.48550/arXiv.2603.24833 , url =. 2603.24833 , archivePrefix =

    work page doi:10.48550/arxiv.2603.24833

  4. [4]

    Journal of Educational Psychology , volume=

    Estimating Causal Effects of Treatments in Randomized and Nonrandomized Studies , author=. Journal of Educational Psychology , volume=. 1974 , publisher=

    work page 1974

  5. [5]

    Operations Research , year =

    Anish Agarwal and Devavrat Shah and Dennis Shen , title =. Operations Research , year =

    work page

  6. [6]

    , journal =

    Athey, Susan and Imbens, Guido W. , journal =

    work page

  7. [7]

    Journal of the American Statistical Association , year =

    Susan Athey and Mohsen Bayati and Nikolay Doudchenko and Guido Imbens and Khashayar Khosravi , title =. Journal of the American Statistical Association , year =

    work page

  8. [8]

    2022 , eprint =

    Jonathan Auerbach and Martin Slawski and Shixue Zhang , title =. 2022 , eprint =

    work page 2022

  9. [9]

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

    Baharav, Tavor Z. and Nicol, Phillip B. and Irizarry, Rafael A. and Ma, Rong , year =. 2507.22170 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  10. [10]

    Journal of the American Statistical Association , year =

    Bai, Jushan and Ng, Serena , title =. Journal of the American Statistical Association , year =

    work page

  11. [11]

    Butts, Kyle and Gardner, John , year =. did2s:

    work page

  12. [12]

    Journal of Econometrics , year =

    Cahan, Ercument and Bai, Jushan and Ng, Serena , title =. Journal of Econometrics , year =

    work page

  13. [13]

    and Tao, Terence , journal =

    Cand\`es, Emmanuel J. and Tao, Terence , journal =. 2010 , volume =

    work page 2010

  14. [14]

    and Recht, Benjamin , year =

    Candès, Emmanuel J. and Recht, Benjamin , year =. Foundations of Computational Mathematics , publisher =

    work page

  15. [15]

    Foundations and Trends in Machine Learning , pages =

    Chen, Yuxin and Chi, Yuejie and Fan, Jianqing and Ma, Cong , title =. Foundations and Trends in Machine Learning , pages =. 2021 , publisher =

    work page 2021

  16. [16]

    Cheng, Cheng and Hoekstra, Mark , journal =

    work page

  17. [17]

    and Chen, Yuxin , title =

    Chi, Yuejie and Lu, Yue M. and Chen, Yuxin , title =. IEEE Transactions on Signal Processing , year =

    work page

  18. [18]

    Chikuse, Yasuko , title =

    work page

  19. [19]

    Journal of the American Statistical Association , year =

    Jungjun Choi and Ming Yuan , title =. Journal of the American Statistical Association , year =

    work page

  20. [20]

    Cunningham, Scott , year =

    work page

  21. [21]

    and Szarek, Stanislaw J

    Davidson, Kenneth R. and Szarek, Stanislaw J. , title =. Handbook of the Geometry of Banach Spaces , volume =. 2001 , pages =

    work page 2001

  22. [22]

    Zhang and Shu Yang , title =

    Chenyin Gao and Han Chen and Anru R. Zhang and Shu Yang , title =. 2025 , eprint =

    work page 2025

  23. [23]

    Gill and Boris Y

    Richard D. Gill and Boris Y. Levit , title =. Bernoulli , number =

    work page

  24. [24]

    Note on sampling without replacing from a finite collection of matrices

    Gross, David and Nesme, Vincent , year =. 1001.2738 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  25. [25]

    , title =

    Holland, Paul W. , title =. Journal of the American Statistical Association , volume =. 1986 , publisher =

    work page 1986

  26. [26]

    and Rubin, Donald B

    Imbens, Guido W. and Rubin, Donald B. , title =. 2015 , publisher =

    work page 2015

  27. [27]

    Keshavan, R. H. and Montanari, A. and Oh, S. , TITLE =. IEEE Transactions on Information Theory , FJOURNAL =. 2010 , NUMBER =

    work page 2010

  28. [28]

    , TITLE =

    Klopp, O. , TITLE =. Bernoulli , FJOURNAL =. 2014 , NUMBER =

    work page 2014

  29. [29]

    Kolda and Brett W

    Tamara G. Kolda and Brett W. Bader , title =. SIAM Review , volume =

    work page

  30. [30]

    , TITLE =

    Koltchinskii, Vladimir and Lounici, Karim and Tsybakov, Alexandre B. , TITLE =. The Annals of Statistics , FJOURNAL =. 2011 , NUMBER =

    work page 2011

  31. [31]

    2019 , eprint =

    Debmalya Mandal and David Parkes , title =. 2019 , eprint =

    work page 2019

  32. [32]

    Improved Guarantees for Heterogeneous Treatment-Effect Estimation via Matrix Completion

    Mehrotra, Anay and Tran, Phuc and Vu, Van H. and Zampetakis, Manolis , year =. 2605.30319 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  33. [33]

    Notices of the American Mathematical Society , year =

    Mezzadri, Francesco , title =. Notices of the American Mathematical Society , year =

    work page

  34. [34]

    Negahban and Martin J

    Sahand N. Negahban and Martin J. Wainwright , title =. Journal of Machine Learning Research , year =

    work page

  35. [35]

    Stewart, G. W. , title =. SIAM Journal on Numerical Analysis , year =

    work page

  36. [36]

    , journal =

    Tropp, Joel A. , journal =. 2012 , publisher =

    work page 2012

  37. [37]

    Tsybakov, Alexandre B , year =

    work page

  38. [38]

    Vershynin, Roman , publisher =

    work page

  39. [39]

    Wainwright, Martin J , year =

    work page

  40. [40]

    Mathematische Annalen , volume =

    Weyl, Hermann , title =. Mathematische Annalen , volume =

    work page

  41. [41]

    Wainwright , title =

    Eric Xia and Yuling Yan and Martin J. Wainwright , title =. 2024 , eprint =

    work page 2024

  42. [42]

    Wainwright , title =

    Yuling Yan and Martin J. Wainwright , title =. 2024 , eprint =

    work page 2024

  43. [43]

    Statistically and Computationally Optimal Estimation and Inference of Common Subspaces

    Agterberg, Joshua , year =. Statistically and. 2606.06483 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    Cortical

    Semedo, Jo. Cortical. Neuron , volume =

    work page

  45. [45]

    Representation

    Liu, Ziqi and Tian, Ye and Tang, Weijing , year =. Representation. 2602.11511 , archivePrefix =

    work page arXiv

  46. [46]

    IEEE Transactions on Information Theory , volume =

    Optimal Estimation of Shared Singular Subspaces Across Multiple Noisy Matrices , author =. IEEE Transactions on Information Theory , volume =

    work page

  47. [47]

    Inference for

    Arroyo, Jes. Inference for. Journal of Machine Learning Research , volume =

    work page