Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Simple Norm Bounds for Polynomial Random Matrices via Decoupling

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Norm estimates for low-degree polynomial random matrices reduce to norms of deterministic derivative-block matrices.

desk verdict Useful decoupling-based recursion for polynomial random matrix norms, but Lemma 3.2 is stated as a first-moment inequality and doesn't justify the p-th moment step in the main theorem; the fix looks straightforward. read the letter →

arxiv 2412.07936 v1 pith:U62227QO submitted 2024-12-10 math.PR cs.CC

Madhur Tulsiani , June Wu This is my paper · ORCID
classification math.PRcs.CC MSC 60B2060E1515A45
keywords polynomialrandommatricesdecouplingSchattennormsmomentinequalitiesmatrixRosenthalinequalitygraphGaussianpolynomialsnormconcentration
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 that norm estimates for low-degree polynomial random matrices can be reduced to norm estimates of a constant number of deterministic matrices built from partial derivatives. For a permutation-symmetric multilinear polynomial random matrix $F$ of degree $d$ in independent mean-zero, unit-variance, bounded variables, it proves $\mathbb{E}\|F\|_{4t}^{4t} \le \sum_{a+b+c=d} (48dt)^{4dt} L^{4ct} \|F_{a,b,c}\|_{4t}^{4t}$, where the $F_{a,b,c}$ are deterministic derivative-block matrices. The argument combines decoupling, the matrix Rosenthal inequality, and a recursive linearization step, and it avoids the intricate trace-expansion combinatorics used in earlier applications. Because the bound is stated in terms of deterministic objects, a user only needs to understand the norms of those blocks; the paper shows this recovers known graph-matrix bounds and yields a quick bound for a tensor-network noise matrix.

What carries the argument

The engine is a two-step recursion. First, decoupling replaces the $d$ product variables in each monomial by independent copies, at the cost of a constant depending only on $d$; the paper uses an elementary partition proof, and obtains the improved constant $k^k$ when the index set carries graph structure. Then the decoupled matrix, viewed as a linear function of the last independent copy, is controlled by the matrix Rosenthal inequality, whose two covariance terms and one diagonal term produce exactly the blocks $F_{a+1,b,c}$, $F_{a,b+1,c}$, and $F_{a,b,c+1}$; repeating until all derivatives are taken leaves the deterministic derivative-block matrices $F_{a,b,c}$. Permutation symmetry is used to make the order of differentiation irrelevant to the Schatten norms that appear.

What would settle it

For a small fixed degree and moment, choose explicit coefficient matrices and bounded mean-zero unit-variance variables such as Rademacher random variables, enumerate all assignments exactly, and compare the true ratio $\mathbb{E}\|F\|_{4t}^{4t}/\mathbb{E}\|F^{\mathrm{dec}}\|_{4t}^{4t}$ with $(d^d)^{4t}$; if any instance exceeds the claimed constant, the decoupling lemma and hence the main theorem are false.

Watch

Extended reading notes

Core claim

The paper's central claim is that for a permutation-symmetric, homogeneous, multilinear polynomial random matrix $F(x)=\sum_{i\in T_n^d} A_i \prod_{j=1}^d x_{i_j}$, with independent mean-zero, unit-variance, bounded variables $|x_i|\le L$, the Schatten moment satisfies $$\mathbb{E}\|F\|_{4t}^{4t}\le \sum_{a+b+c=d}(48dt)^{4dt}$L^{{4ct}}$\|F_{a,b,c}\|_{4t}^{4t},$$ where each $F_{a,b,c}$ is a deterministic matrix assembled from $d$-th order partial derivatives of $F$. Thus a probabilistic norm bound becomes a check of a constant number of deterministic matrix norms, with no distribution-specific combinatorial expansion needed; moment growth and boundedness enter only through $L$.

Load-bearing premise

The load-bearing premise is that the decoupling step controls the full $4t$-th power with the constant $d^d$, i.e. $\mathbb{E}\|F\|_{4t}^{4t}\le (d^d)^{4t}\mathbb{E}\|F^{\mathrm{dec}}\|_{4t}^{4t}$; the paper's Lemma 3.2 states the constant as $d^d$ without the exponent, so if the correct power-form constant is larger, the main theorem's constants must change.

Editorial extensions

If this is right

  • For any multilinear polynomial random matrix of fixed degree over a product distribution with bounded, normalized variables, the expected $4t$-th Schatten power is controlled by finitely many deterministic derivative-block norms, making Markov-based high-probability norm bounds immediate.
  • In the dense graph-matrix setting, the derivative-block norms realize exactly the minimum vertex separator, and the dominant term reproduces the known $n^{(k-r)/2}$ scaling.
  • In the sparse setting, terms with $c>0$ contribute factors of $L^c$ and select a separator $S$ maximizing $L^{e(S)}n^{(k-|S|)/2}$, recovering the sparse separator bounds.
  • For Gaussian polynomial matrices, the same recursion fed by a Poincaré-type matrix inequality gives bounds in terms of expected matrices $\mathbb{E} P_{a,b}$, and the melon-graph example yields $\mathbb{E}\|M-\mathbb{E}M\|_{2t}^{2t}=O(n^{3t})$.
  • Non-homogeneous multilinear polynomials are handled by writing them as sums of homogeneous parts, at the cost of a factor $D^{4t}$ and a trace inequality.

Reading between the lines

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

  • Because only finitely many deterministic blocks appear, the bound is computable for concrete coefficient matrices, turning polynomial random-matrix norm estimates into a numerical linear algebra task.
  • The same derivative-block recursion could be applied to hypergraph or tensor-network matrices, with the melon-graph example suggesting the pattern.
  • The role of the $c$-type derivatives in sparse graph matrices points toward a weighted-separator interpretation that may transfer to other weighted random graph models.
  • The paper notes that boundedness can be replaced by moment growth; checking whether the $(48dt)^{4dt}$ constant remains stable for subgaussian variables is a natural stress test.
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

4 major / 5 minor

Summary. The paper proposes a decoupling-based recursive method for bounding Schatten-norm moments E‖F‖_{2t}^{2t} of random matrices whose entries are low-degree polynomials in independent variables. The main result (Theorem 1.1 / 4.5) bounds a degree-d multilinear permutation-symmetric polynomial matrix by the Schatten norms of O(d^3) deterministic derivative blocks F_{a,b,c}; Theorem 1.2 / 5.3 gives an analogous Gaussian bound; applications recover vertex-separator bounds for graph matrices and give a bound for the melon-graph noise matrix. The overall strategy is attractive and the applications are meaningful, but several stated inequalities are not the ones proved.

Significance. If the gaps are repaired, this is a useful unifying framework: it reduces a potentially complicated combinatorial moment computation to norms of deterministic derivative matrices, and it recovers known graph-matrix separator bounds and a tensor-network example with short proofs. The paper is honest about scope (multilinear case, bounded variables, Gaussian case) and cites relevant prior work, including [RT23] as a starting point. The central proof idea is standard, so the value is mostly expository and unifying rather than introducing a new technical engine; nevertheless the results are broadly useful and the paper is clearly written overall.

major comments (4)
  1. [Section 3, Lemma 3.2; proof of Theorem 4.5] Lemma 3.2 is stated and proved for E‖Σ B_i x_{i1}...x_{id}‖_{2t} ≤ d^d E‖dec‖_{2t}. The proof of Theorem 4.5 uses this as E‖F‖_{4t}^{4t} ≤ d^{4dt}E‖F^{(dec)}‖_{4t}^{4t}. Raising the lemma to the 4t-th power gives (E‖F‖_{4t})^{4t} ≤ d^{4dt}(E‖F^{(dec)}‖_{4t})^{4t}, whose left side is dominated by E‖F‖_{4t}^{4t}, so it does not imply the required moment bound. Lemma 3.3 and its use in Theorem 6.5 have the same problem. The partition proof can be rerun with F defined as the p-th moment, yielding the correct constant d^{dp} (respectively k^{kp} for the graph-structured case), so the gap is repairable, but as written the central decoupling step is unproved.
  2. [Theorem 5.3 and Claim 5.4] Claim 5.4 gives a per-step factor (2√2t)^{2t} and a branching into two centered terms plus expectation terms. Iterating the claim d times, as the proof of Theorem 5.3 instructs, accumulates a factor (2√2t)^{2td} (up to branching factors), whereas Theorem 5.3 states a total constant (2d√2t)^{2t}. For degree 2, the recursive proof yields at least a term of size (2√2t)^{4t}‖EP_{2,0}‖_{2t}^{2t}, while the theorem claims (4√2t)^{2t} as the constant for that term; for t ≥ 2 the former is larger, so the proof as written cannot establish the stated constant. The theorem or its proof needs to be corrected.
  3. [Theorem 4.2] The statement of Theorem 4.2 is E‖F−EF‖_{4t} ≤ 2(32t)^2 Σ L^c‖F_{a,b,c}‖_{4t}, but the proof defines E := E‖F‖_{4t}^{4t} and finishes with a bound on that 4t-th moment by the 4t-th powers of the derivative norms. As written the theorem and proof do not match; the statement appears to have lost the exponent 4t. This is a warm-up result, but it should be stated correctly.
  4. [Section 4, paragraph after Theorem 4.2; Claim 4.4] The reduction from derivatives of the decoupled polynomial to the matrices F_{a,b,c} of the original polynomial is justified only by a 2×2 example and the sentence 'This also applies to F(x) of higher degree.' Claim 4.4 works with derivatives of F(x(1),...,x(D)), while Theorem 4.5 states the bound in terms of F_{a,b,c} of F(x). The higher-degree case involves d! symmetrization and can change the multinomial constants; a formal statement and proof of the norm comparison are needed.
minor comments (5)
  1. [Proof of Theorem 4.5] There is a typo 'F((x)' where 'F(x)' is intended.
  2. [Proof of Theorem 4.2] The proof uses a strict inequality 'E < ...' where '≤' is intended.
  3. [Section 1] The word 'defintion' should be 'definition' in the discussion of sparse graph matrices.
  4. [Theorem 4.5 statement] The notation d = a + b + c overloads the degree d; renaming the degree or the split variables would improve readability.
  5. [Section 3] Since Lemma 3.2 and Lemma 3.3 are used exclusively through moment versions, the lemmas should be stated in moment form, with the correct exponent on the constant, so that the applications are directly supported.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central bounds are derived from external decoupling and matrix Rosenthal inequalities, with self-citations used only as context and benchmarks.

full rationale

The central derivation chain is self-contained and does not reduce to its own inputs. Lemma 3.2 is stated as a generalization of an external decoupling inequality from Rauhut [Rau10] and ultimately de la Pena-Gine [PG99]; Lemma 4.1 is the non-Hermitian matrix Rosenthal inequality derived from [MJC+12, Corollary 7.4]; and the Gaussian section builds on [HT20, Theorem 7.1]. None of these load-bearing inequalities is taken from the present authors' prior work. The only self-citations are [RT23] (co-authored by M. Tulsiani) and [JPR+21] (also involving Tulsiani). They appear as motivation (the introduction describes [RT23] as the prior Efron-Stein approach that stalls for non-Rademacher distributions), as standard graph-matrix definitions in Section 6, and as benchmarks for the recovered graph-matrix bounds. No theorem in the paper relies on a bound or uniqueness statement imported from these self-citations; the graph-matrix application derives the known bound from Theorem 4.5 and directly computes the norms of the deterministic derivative matrices. The constants in Theorems 4.5, 5.3, and 6.5 arise from external inequalities and combinatorial counting, not from fitting parameters to a target bound. The reader's flagged issue with Lemma 3.2 is that the stated lemma is an L1-norm decoupling bound while Theorems 4.2 and 4.5 use an unproved moment version with an extra exponent; that is a proof-gap or correctness concern, not circularity, because the needed moment inequality is not assumed as an input but would follow from the same external decoupling argument with a corrected exponent. No circular step was identified.

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

The proof relies on standard decoupling, matrix Rosenthal, and Poincaré inequalities, plus a structural permutation symmetry assumption. No fitted parameters, no new physical or mathematical entities, and no constants tuned to data.

assumptions (6)
  • standard math Decoupling inequality for moments of U-statistics (Lemma 3.1, attributed to de la Peña-Giné [PG99])
    Used to reduce the polynomial matrix to independent copies; assumed true for Schatten norms as cited.
  • standard math Matrix Rosenthal inequality for centered independent random matrices ([MJC+12] Corollary 7.4)
    Used in Lemma 4.1 and recursively in Claim 4.4 to lower the degree of the polynomial by differentiation.
  • standard math Huang-Tropp Poincaré inequality for Gaussian matrix-valued functions ([HT20] Theorem 7.1)
    Used in Lemma 5.1 for the Gaussian polynomial random matrix case.
  • standard math Trace inequality for sums of matrices ([SA13] Theorem 3.1)
    Used to split homogeneous parts and to control sums of means in the Gaussian recursion.
  • domain assumption Permutation symmetry can be assumed without loss for multilinear polynomials over commuting variables
    The paper symmetrizes coefficient matrices; this changes derivative matrices only up to constants, as discussed in Section 2 and Remark 2.5.
  • domain assumption Random variables are independent, mean zero, variance one, and bounded with |x_i| <= L
    Assumption of Theorems 1.1 and 4.5; the authors note it can be relaxed to subgaussian moment growth.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Simple Norm Bounds for Polynomial Random Matrices via Decoupling." pith.science (2026). https://pith.science/paper/U62227QO

@misc{pith2026241207936,
  author       = {Pith},
  title        = {Pith review of: Simple Norm Bounds for Polynomial Random Matrices via Decoupling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/U62227QO}},
  note         = {Machine review of arXiv:2412.07936}
}
read the original abstract

We present a new method for obtaining norm bounds for random matrices, where each entry is a low-degree polynomial in an underlying set of independent real-valued random variables. Such matrices arise in a variety of settings in the analysis of spectral and optimization algorithms, which require understanding the spectrum of a random matrix depending on data obtained as independent samples. Using ideas of decoupling and linearization from analysis, we show a simple way of expressing norm bounds for such matrices, in terms of matrices of lower-degree polynomials corresponding to derivatives. Iterating this method gives a simple bound with an elementary proof, which can recover many bounds previously required more involved techniques.

Figures

Figures reproduced from arXiv: 2412.07936 by the authors.

Figure 1
Figure 1. A shape τ As will be apparent from the analysis below, the norm bounds for such matrices behave different characterizations in the “dense” regime where p = Ω(1) and the “sparse” one where p = o(1). This is because the subgaussian norm L = ∆p ≈ p −1/2 is bounded in the first case, and growing in the second case. The bounds for the first case are stated in terms of the size the minimum vertex separator separating Uτ f… view at source ↗
Figure 2
Figure 2. F1,0,0, F0,1,0 and F2,3,0 The bound in Theorem 1.1 is then in terms of these new graph matrices where we delete all d edges, and split their endpoints between Uτ or Vτ to obtain U′ , V ′ ⊆ V(τ) = [k]. For an entry M[i ′ , j ′ ] of any such matrix, there is at most one ψ : V(τ) → [n] such that ψ(U′ ) = i ′ and ψ(V ′ ) = j ′ (assuming τ has no isolated 7 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. F0,0,1 As we saw above, increasing the intersection of U′ and V ′ decreases the norm, and so the matrices with c > 0 should have a smaller Schatten norm. However, they are now included in the bound with a multiplicative factor of L c . Thus, we simply look for a vertex separator S maximizing L e(S) · n (k−|S|)/2 where e(S) counts the (yellow) edges contained in S = U′ ∩ V ′ . This is precisely the “sparse vertex sep… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: The Melon Graph Theorem 6.7. (See also [Oue22] Theorem 7) Let T ∈ (Rn ) ⊗3 be a random noise tensor with i.i.d. standard Gaussian entries. For i, j ∈ [n], let Ti and Tj be the first-mode slices of T. The pure noise matrix M ∈ Rn×n associated to the melon graph has entr…

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Matrix Chaos Inequalities and Chaos of Combinatorial Type

    math.PR 2024-12 conditional novelty 7.0 of 10

    For polynomial random matrices, the expected norm is controlled by flattening norms of the coefficient tensor, with a simple mechanical rule for combinatorial-type chaoses.

Reference graph

Works this paper leans on

42 extracted references · 34 canonical work pages · cited by 1 Pith paper

  1. [1]

    Matrix poincar \'e inequalities and concentration

    Richard Aoun, Marwa Banna, and Pierre Youssef. Matrix poincar \'e inequalities and concentration. Advances in Mathematics , 371:107251, 2020

    work page 2020

  2. [2]

    Graph matrices: Norm bounds and applications

    Kwangjun Ahn, Dhruv Medarametla, and Aaron Potechin. Graph matrices: Norm bounds and applications. arXiv preprint https://arxiv.org/abs/1604.03423 , 2021

    arXiv 2021

  3. [3]

    Concentration inequalities for non-lipschitz functions with bounded derivatives of higher order

    Rados aw Adamczak and Pawe Wolff. Concentration inequalities for non-lipschitz functions with bounded derivatives of higher order. Probability Theory and Related Fields , 162(3):531--586, 2015

    work page 2015

  4. [4]

    Moment inequalities for functions of independent random variables

    St \'e phane Boucheron, Olivier Bousquet, G \'a bor Lugosi, and Pascal Massart. Moment inequalities for functions of independent random variables . The Annals of Probability , 33(2):514 -- 560, 2005

    work page 2005

  5. [5]

    Matrix concentration inequalities and free probability

    Afonso S Bandeira, March T Boedihardjo, and Ramon van Handel. Matrix concentration inequalities and free probability. Inventiones M athematicae , pages 1--69, 2023

    work page 2023

  6. [6]

    Higher order concentration of measure

    Sergey G Bobkov, Friedrich G \"o tze, and Holger Sambale. Higher order concentration of measure. Communications in Contemporary Mathematics , 21(03):1850043, 2019

    work page 2019

  7. [7]

    A nearly tight sum-of-squares lower bound for the planted clique problem

    Boaz Barak, Samuel Hopkins, Jonathan Kelner, Pravesh K Kothari, Ankur Moitra, and Aaron Potechin. A nearly tight sum-of-squares lower bound for the planted clique problem. SIAM Journal on Computing , 48(2):687--735, 2019

    work page 2019

  8. [8]

    Kothari, and Jeff Xu

    Mitali Bafna, Jun-Ting Hsieh, Pravesh K. Kothari, and Jeff Xu. Polynomial-time power-sum decomposition of polynomials. In 2022 IEEE 63rd Annual Symposium on Foundations of Computer Science (FOCS) , pages 956--967, 2022

    work page 2022

Show all 42 references
  1. [9]

    Resolving matrix S pencer conjecture up to poly-logarithmic rank

    Nikhil Bansal, Haotian Jiang, and Raghu Meka. Resolving matrix S pencer conjecture up to poly-logarithmic rank. In Proceedings of the 55th ACM Symposium on Theory of Computing , pages 1814--1819, 2023

    2023

  2. [10]

    Universality and sharp matrix concentration inequalities

    Tatiana Brailovskaya and Ramon van Handel. Universality and sharp matrix concentration inequalities. arXiv preprint arXiv:2201.05142 , 2022

    2022 arXiv

  3. [11]

    Fast algorithm for overcomplete order-3 tensor decomposition

    Jingqiu Ding, Tommaso d’Orsi, Chih-Hung Liu, David Steurer, and Stefan Tiegel. Fast algorithm for overcomplete order-3 tensor decomposition. In Conference on Learning Theory , pages 3741--3799. PMLR, 2022

    2022

  4. [12]

    Rong Ge, Qingqing Huang, and Sham M. Kakade. Learning mixtures of gaussians in high dimensions. In Proceedings of the Forty-Seventh Annual ACM Symposium on Theory of Computing , STOC '15, page 761–770, New York, NY, USA, 2015. Association for Computing Machinery

    2015

  5. [13]

    Random Tensors

    Răzvan Gheorghe Gurău. Random Tensors . Oxford University Press, 10 2016

    2016

  6. [14]

    Tensor principal component analysis via sum-of-square proofs

    Samuel B Hopkins, Jonathan Shi, and David Steurer. Tensor principal component analysis via sum-of-square proofs. In Conference on Learning Theory , pages 956--1006. PMLR, 2015

    2015

  7. [15]

    A robust spectral algorithm for overcomplete tensor decomposition

    Samuel B Hopkins, Tselil Schramm, and Jonathan Shi. A robust spectral algorithm for overcomplete tensor decomposition. In Conference on Learning Theory , pages 1683--1722. PMLR, 2019

    2019

  8. [16]

    Fast spectral algorithms from sum-of-squares proofs: tensor decomposition and planted sparse vectors

    Samuel B Hopkins, Tselil Schramm, Jonathan Shi, and David Steurer. Fast spectral algorithms from sum-of-squares proofs: tensor decomposition and planted sparse vectors. In Proceedings of the 48th ACM Symposium on Theory of Computing , pages 178--191, 2016

    2016

  9. [17]

    De Huang and Joel A. Tropp. From poincar \'e inequalities to nonlinear matrix concentration. Bernoulli , 2020

    2020

  10. [18]

    De Huang and Joel A. Tropp. Nonlinear matrix concentration via semigroup methods. Electronic Journal of Probability , 26:Art. No. 8, Jan 2021

    2021

  11. [19]

    Sum-of-squares lower bounds for sparse independent set

    Chris Jones, Aaron Potechin, Goutham Rajendran, Madhur Tulsiani, and Jeff Xu. Sum-of-squares lower bounds for sparse independent set. In Proceedings of the 62nd IEEE Symposium on Foundations of Computer Science , 2021

    2021

  12. [20]

    Exact nuclear norm, completion and decomposition for random overcomplete tensors via degree-4 sos

    Bohdan Kivva and Aaron Potechin. Exact nuclear norm, completion and decomposition for random overcomplete tensors via degree-4 sos. arXiv preprint arXiv:2011.09416 , 2020

    2011 arXiv

  13. [21]

    Concentration of multivariate polynomials and its applications

    Jeong Han Kim and Van H Vu. Concentration of multivariate polynomials and its applications. Combinatorica , 20(3):417--434, 2000

    2000

  14. [22]

    Decoupling Inequalities for Polynomial Chaos

    Stanislaw Kwapien. Decoupling Inequalities for Polynomial Chaos . The Annals of Probability , 15(3):1062 --1071, 1987

    1987

  15. [23]

    Estimates of moments and tails of gaussian chaoses

    Rafa Lata a. Estimates of moments and tails of gaussian chaoses. The Annals of Probability , 34(6):2315--2331, 2006

    2006

  16. [24]

    A note on quantum expanders, 2023

    Cécilia Lancien and Pierre Youssef. A note on quantum expanders, 2023. http://arxiv.org/abs/2302.07772 arXiv:2302.07772

    2023

  17. [25]

    Matrix concentration inequalities via the method of exchangeable pairs

    Lester Mackey, Michael Jordan, Richard Chen, Brendan Farrell, and Joel Tropp. Matrix concentration inequalities via the method of exchangeable pairs. The Annals of Probability , 42, 2012

    2012

  18. [26]

    McConnell and Murad S

    Terry R. McConnell and Murad S. Taqqu. Double integration with respect to symmetric stable processes. Technical report, Technical Report 618, Cornell Univ., 1984

    1984

  19. [27]

    Spectral methods from tensor networks

    Ankur Moitra and Alexander S Wein. Spectral methods from tensor networks. In Proceedings of the 51st ACM Symposium on Theory of Computing , pages 926--937, 2019

    2019

  20. [28]

    Rivasseau

    Mohamed Ouerfelli, Mohamed Tamaazousti, and V. Rivasseau. Random tensor theory for tensor decomposition. In AAAI Conference on Artificial Intelligence , 2022

    2022

  21. [29]

    New perspectives and tools for Tensor Principal Component Analysis and beyond

    Mohamed Ouerfelli. New perspectives and tools for Tensor Principal Component Analysis and beyond . PhD thesis, Université Paris-Saclay, 2022

    2022

  22. [30]

    Polynomial bounds for decoupling, with applications

    Ryan O'Donnell and Yu Zhao. Polynomial bounds for decoupling, with applications. In Proceedings of the 31st Conference on Computational Complexity , CCC '16, Dagstuhl, DEU, 2016. Schloss Dagstuhl--Leibniz-Zentrum fuer Informatik

    2016

  23. [31]

    Peña and Evarist Giné

    Víctor H. Peña and Evarist Giné. Decoupling: From dependence to independence . Springer-Verlag, 1999

    1999

  24. [32]

    Pena and S

    Victor H. Pena and S. J. Montgomery-Smith. Decoupling Inequalities for the Tail Probabilities of Multivariate U -Statistics . The Annals of Probability , 23(2):806 -- 816, 1995

    1995

  25. [33]

    Daniel Paulin, Lester Mackey, and Joel A. Tropp. Efron–Stein inequalities for random matrices . The Annals of Probability , 44(5):3431 -- 3473, 2016

    2016

  26. [34]

    Nonlinear R andom M atrices and A pplications to the S um of S quares H ierarchy

    Goutham Rajendran. Nonlinear R andom M atrices and A pplications to the S um of S quares H ierarchy . PhD thesis, University of Chicago, 2022

    2022

  27. [35]

    Compressive sensing and structured random matrices

    Holger Rauhut. Compressive sensing and structured random matrices. In Theoretical Foundations and Numerical Methods for Sparse Recovery , pages 1--92. De Gruyter, Berlin, New York, 2010

    2010

  28. [36]

    A statistical model for tensor pca

    Emile Richard and Andrea Montanari. A statistical model for tensor pca. In Neural Information Processing Systems , 2014

    2014

  29. [37]

    Concentration of polynomial random matrices via efron-stein inequalities

    Goutham Rajendran and Madhur Tulsiani. Concentration of polynomial random matrices via efron-stein inequalities. Proceedings of the 2023 Annual ACM-SIAM Symposium on Discrete Algorithms (SODA) , 2023

    2023

  30. [38]

    Trace inequalities for matrices

    Khalid Shebrawi and Hussien Albadawi. Trace inequalities for matrices. Bulletin of the Australian Mathematical Society , 87(1):139–148, 2013

    2013

  31. [39]

    Bernstein-like concentration and moment inequalities for polynomials of independent random variables: multilinear case

    Warren Schudy and Maxim Sviridenko. Bernstein-like concentration and moment inequalities for polynomials of independent random variables: multilinear case. arXiv preprint arXiv:1109.5193 , 2011

    2011 arXiv

  32. [40]

    Concentration and moment inequalities for polynomials of independent random variables

    Warren Schudy and Maxim Sviridenko. Concentration and moment inequalities for polynomials of independent random variables. In Proceedings of the twenty-third annual ACM-SIAM symposium on Discrete Algorithms , pages 437--446. SIAM, 2012

    2012

  33. [41]

    Joel A. Tropp. An introduction to matrix concentration inequalities. Foundations and Trends in Machine Learning , 8(1-2):1--230, 2015

    2015

  34. [42]

    High-Dimensional Probability: An Introduction with Applications in Data Science

    Roman Vershynin. High-Dimensional Probability: An Introduction with Applications in Data Science . Cambridge University Press, 2018

    2018

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.