Pith. sign in

REVIEW 2 major objections 5 minor 40 references

Recovering Imbalanced Clusters via Gradient-Based Projection Pursuit

T0 review · 2 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read Gradient projection pursuit provably recovers imbalanced cluster directions with ~(d^2 p^2) samples, and imbalance reduces cost.

desk verdict The sample-complexity story is plausible and the appendix is substantial, but Theorems 1 and 2 overclaim success probability due to an initialization gap that needs a simple fix. read the letter →

arxiv 2502.02668 v2 pith:MFMK5OSG submitted 2025-02-04 cs.LG

Martin Eppert , Satyaki Mukherjee , Debarghya Ghoshdastidar This is my paper · ORCID
classification cs.LG MSC 62H1262F12
keywords projectionpursuitplantedvectorimbalancedclustersgradientascentsamplecomplexitylow-degreepolynomialsRiemannianBernoulli-Rademacher
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 proves a sample-complexity guarantee for projection pursuit in a planted-vector model: a gradient-based algorithm recovers the hidden direction of an imbalanced two-cluster distribution with $\widetilde{\Theta}(d^2 p^2)$ samples, where $d$ is dimension and $p$ is the probability of the rare cluster. The central claim is that imbalance helps—as $p$ shrinks, the sample cost drops, so recovering a rare cluster is provably easier than recovering a balanced one. The same proof machinery gives a Bernoulli–Rademacher planted-vector guarantee with $\widetilde{\Theta}(d^3 p^4)$ samples, and the paper derives low-degree lower bounds close to both rates. The upshot is that projection pursuit is not only cheap to run, at $\mathcal{O}(nd)$ per epoch, but also statistically analyzable, coming within a $\sqrt{dp}$ factor of a conjectured computational-hardness benchmark. On FashionMNIST and the Human Activity Recognition dataset, the ReLU² projection separates one class from the rest even with only a few hundred samples.

What carries the argument

The load-bearing object is the planted-vector model $x\sim N(\nu u^\ast, I_d-u^\ast u^{\ast\top})$ with $\nu\sim B(p)$, two clusters at $\mu_1=\sqrt{(1-p)/p}$ and $\mu_2=-\sqrt{p/(1-p)}$; unit variance means PCA-type whitening cannot see the signal. The argument runs on the Riemannian gradient of the ReLU² projection index, projected orthogonal to the current direction and renormalized; Lemma 4 shows the expected gradient along $u^\ast$ is $\Theta(\langle u,u^\ast\rangle^2/\sqrt{p})$ in the early regime and $\Theta(\langle u,u^\ast\rangle)$ near convergence, so a large first learning rate $\eta_1=\Omega(\sqrt{dp})$ makes geometric progress and a small second rate $\eta_2=\Theta(1)$ polishes the estimate. Fresh mini-batches make gradient estimates independent across the $s=\Theta(\log d)$ steps, and normalized-sample initialization yields overlap $\Theta(1/\sqrt{pd})$ with probability $p$, beating random initialization for small $p$.

What would settle it

Run only the initialization stage on synthetic planted data: for each $p$ and $d$, draw $\lceil C/p\rceil$ normalized samples and record the frequency with which the best overlap reaches $\Theta(1/\sqrt{pd})$; if that frequency converges to $1-e^{-C}$ instead of to $1$, Theorem 1's probability claim is false as stated.

Watch

Extended reading notes

Core claim

In the paper's own terms, Theorem 1 states that for any $\beta>0$ and $p\in(1/\sqrt{d},1/2)$, there exist parameters such that Algorithm 1, run with $\varphi(x)=\max\{0,x\}^2$, outputs $\hat{u}$ with $\langle\hat{u},u^\ast\rangle\ge 1-\beta$ with probability $1-o(1)$, using $n=\widetilde{\Theta}(d^2p^2)$ samples from the imbalanced-clusters planted-vector model $\nu\sim B(p)$. The proof is organized by Lemma 1, which reduces convergence to three preconditions: at least one normalized-sample initialization must start with overlap $\Theta(1/\sqrt{pd})$; the Riemannian gradient estimate must expand that overlap multiplicatively while $\langle u,u^\ast\rangle$ is below a threshold; and the projection index $\psi=\varphi$ must separate converged from unconverged directions by a sample-efficient threshold. A separate low-degree calculation (Theorem 3) shows that no low-degree polynomial test can solve the hypothesis-testing problem for $n=\widetilde{\Theta}(d^{1.5}p)$ samples, so the algorithm's guarantee sits near the conjectured computational limit.

Load-bearing premise

The proof leans on having at least one of the random starting directions begin close to the true direction with near-certain probability, but the bound it proves gives each starting direction only a constant chance of being close, so the stated number of starts does not deliver the claimed near-certainty.

Editorial extensions

If this is right

  • Imbalanced clusters are provably easier: in the planted model, the sample-complexity bound $\widetilde{\Theta}(d^2p^2)$ decreases as the rare cluster becomes rarer.
  • The general Lemma 1 template applies to other projection indices and distributions; the paper instantiates it for kurtosis on Bernoulli–Rademacher vectors at $\widetilde{\Theta}(d^3p^4)$ samples.
  • Under the Low-Degree Conjecture, no polynomial-time test can beat $\widetilde{\Theta}(d^{1.5}p)$ samples in the imbalanced setting, so the gradient algorithm is within a $\sqrt{dp}$ factor of the conjectured computational limit.
  • On FashionMNIST and Human Activity Recognition, the ReLU² projection yields directions with high information gain about class labels even when only a few hundred samples are available.

Reading between the lines

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

  • An implicit consequence is that the ReLU² projection is effectively locating a rare, heavy-tailed group, so the algorithm should transfer to outlier detection; this is testable by contaminating standard data with a rare cluster and checking whether the sample-complexity slope tracks $p$.
  • The $\sqrt{dp}$ gap to the low-degree lower bound is left open, but the paper's moment-based analysis suggests a spectral method built on higher-order moments, exploiting the distribution's skewness, might close it.
  • Because fresh mini-batches are needed only for the analysis, the low-dimensional experiments with resampling suggest a practical streaming variant could reuse a single pass over data; that is a natural testable extension.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 5 minor

Summary. The paper studies gradient-based projection pursuit in the planted-vector model. It proposes a two-stage Riemannian gradient-ascent algorithm (Algorithm 1) using normalized-sample initializations and fresh mini-batches, and analyzes its sample complexity for two signal distributions: imbalanced two-cluster signals with the ReLU^2 projection index, and Bernoulli-Rademacher sparse signals with a kurtosis index. The central claims are that the imbalanced-cluster direction is recoverable with ~Θ(d^2 p^2) samples with probability 1-o(1), and that the Bernoulli-Rademacher direction is recoverable with ~Θ(d^3 p^4) samples with probability 1-o(1). The paper also derives a low-degree lower bound of the form n = ~O(d^{1.5} p) for the imbalanced setting, and reports synthetic and real-data experiments on FashionMNIST and the Human Activity Recognition dataset.

Significance. If the theorem statements were fully proved, the main contribution would be a provable demonstration that, in planted-vector projection pursuit, imbalanced clusters become statistically easier as the minority probability p shrinks, complementing existing spectral and low-degree results. The low-degree calculation extends the Mao-Wein framework to a non-symmetric signal distribution, and the experiments, including the information-gain evaluation on real data, support the practical relevance of the method. The paper is also careful to label the low-degree conjecture as a conjecture and to state limitations in the appendix. However, the probability claims in Theorems 1 and 2 currently outrun the proofs: two union-bound steps support only constant success probability, so the significance is conditional on a repair.

major comments (2)
  1. [§2.4, Lemma 2 and Theorem 1; also Theorem 2] Precondition 1 of Lemma 1 requires that, with probability 1-o(1), at least one of the ninit normalized-sample initializations satisfies <u0,u*> ≥ Θ(1/sqrt(pd)). Lemma 2 establishes this event only with per-initialization probability Θ(p). For ninit = Ω(1/p), as stated in Theorem 1, or ninit = Θ(1/p), as stated in Theorem 2, the probability that at least one initialization succeeds is 1 - (1 - Θ(p))^{ninit} = 1 - exp(-Θ(1)), a constant strictly between 0 and 1, not 1-o(1). Since every subsequent gradient-concentration argument in Appendices B.1 and C.1 starts from this good initialization, the claimed success probability in Theorems 1 and 2 is not proved as written. The natural repair is to state ninit = ω(1/p), for example ninit = ceil((log d)/p), which makes the initialization failure probability exp(-Θ(ninit p)) -> 0 and remains compatible with the stated total sample budgets in both theorems.
  2. [§2.3, Lemma 1 and Appendix A; applications in Appendices B.1 and C.1] Precondition 2 of Lemma 1 is stated per gradient step with failure probability 1 - O(1/s), and the proof applies a union bound over s = Ω(log d) steps. This yields a total failure probability of O(1), not o(1): (1 - C/s)^s tends to e^{-C}. The applications in the proofs of Theorems 1 and 2 also state concentration bounds 'with probability at least 1 - Θ(1/s)' and then iterate over s = Θ(log d) steps, so the same constant-failure issue arises there. To obtain the 1-o(1) conclusion of Lemma 1, the per-step failure probability must be o(1/s), for example 1/s^{1+ε}, which would require restating Precondition 2 or choosing the concentration parameters and sample size accordingly. As written, the high-probability claim in Theorems 1 and 2 is again only a constant success probability.
minor comments (5)
  1. [Theorem 1 statement] The parameter range is misstated: the theorem says p ∈ (1/2, 1/sqrt(d)), while the surrounding text and the proof use p ∈ (1/sqrt(d), 1/2).
  2. [Remark 1] The interval p ∈ (1/sqrt(p), 1/2) is a typo; it should be p ∈ (1/sqrt(d), 1/2).
  3. [Algorithm 1 pseudocode] The pseudocode is ambiguous: line 5 passes a single vector u to the gradient-ascent subroutine, but Algorithm 2 loops over j = 1,...,ninit and needs a collection of initializations; lines 5-6 also overwrite ^u before line 7 selects among the set {^u_j}. Please clarify that the two-stage ascent is applied to each of the ninit initializations and that the returned set is produced before selection.
  4. [Section 5, conclusion] The conclusion describes the low-degree bound as 'n = ~Θ(d^{1.5} p)', but Theorem 3 only demonstrates failure of low-degree tests for n = O(d^{1.5} p); this is a lower bound of Ω(d^{1.5} p), not a matching Θ statement, since no corresponding upper bound is proved.
  5. [Figure 1 caption] In the caption of Figure 1b, the parameter choices are listed as p ∈ {d^{-0.5}, d^{-0.5}, 0.3}; the repeated d^{-0.5} appears to be a typo, likely d^{-0.4}, and the red-line slopes (3, 1.8, 1) should be checked against those parameter values.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the sample-complexity theorems are derived from the planted model by direct moment computations and are benchmarked against external results (Mao-Wein, Dudeja-Hsu, Spielman et al., Qu et al.); the flagged initialization-probability gap is a proof error, not an input-output reduction.

full rationale

Walking the derivation chain, I find no step that reduces to its own input by construction. Theorems 1 and 2 are proven by direct computation from the planted-vector model: Lemma 4 (resp. Lemma 7) computes E[<g_u(X), u*>] explicitly for phi(x)=max{0,x}^2 (resp. x^4), Lemma 3 (resp. Lemma 8) supplies Bernstein-type concentration whose algebra yields the stated n = Theta~(d^2 p^2) (resp. Theta~(d^3 p^4)) from the model parameters, and Lemmas 6/10 derive the selection test from the same expectations rather than assuming it. No theorem constant is fitted to data: eta_1, eta_2, s, and ninit are chosen analytically in the proofs, and the red phase-transition lines in Fig. 1 are post hoc illustrations of the proved exponents, not fitted inputs that are then relabeled as predictions. The lower bound (Theorem 3) is an extension of the external Mao-Wein [24] machinery, rests on the explicitly labeled external Low Degree Conjecture, and is compared honestly as a gap (d^1.5 p vs d^2 p^2) rather than being used to tune the algorithm. The reference list contains no self-citations, so the self-citation patterns (kinds 3-5) do not apply; the imbalanced-clusters analysis is not a rename of a known result since the lower bound there is new relative to the cited work. Two flagged passages are genuine limitations but are not circularity: (i) Appendix B.1's claim that Lemma 2 'fulfills Precondition 1' is quantitatively wrong as stated, since Lemma 2 yields a good initialization with probability only Theta(p) per draw and ninit = Omega(1/p) therefore gives constant, not 1-o(1), success probability; this is a probability-calculus gap (repairable with ninit = omega(1/p)) that breaks the theorem's probability claim but does not identify any quantity with its own input; (ii) the Low Degree Conjecture and Remark 2's unproved hierarchical counterexample are acknowledged open points, not derivation-by-definition. The paper is self-contained against external benchmarks, so the honest finding is no circularity (score 0).

Assumptions & free parameters 0 free parameters · 3 assumptions · 0 invented entities

The core theoretical claims rest on the planted vector model with unit variance, the explicitly conjectural low-degree framework for the lower bound, and standard high-dimensional concentration. No fitted parameters enter the central sample-complexity bounds; hyperparameters in the theorems are chosen to satisfy the proofs, and the experimental choices η1=√(dp), η2=0.5 follow the theorem prescriptions.

assumptions (3)
  • domain assumption Low-Degree Conjecture: if the degree-D likelihood ratio norm stays bounded, then no polynomial-time algorithm solves the hypothesis testing problem.
    Invoked in Section 3 (Conjecture paragraph) to turn Theorem 3's bounded low-degree likelihood ratio into a computational lower bound for all polynomial-time estimators. The paper explicitly labels this as a conjecture, so the lower bound is conditional.
  • domain assumption Planted vector model: X_i ~ N(ν_i u*, I_d − u* u*⊤) with ν_i ~ F having unit variance.
    Definition 1 fixes the data model for all theoretical claims. Real-data experiments (FashionMNIST, HAR) do not satisfy this model, so the theoretical guarantees do not transfer to them.
  • standard math High-dimensional Gaussian concentration and Bernstein-type inequalities as used in Lemmas 3, 5, 8, 9, and 17.
    The proofs rely on standard sub-Gaussian and sub-exponential concentration bounds, union bounds over epsilon-nets, and operator norm estimates for Gaussian matrices; these are unproved background results.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Recovering Imbalanced Clusters via Gradient-Based Projection Pursuit." pith.science (2026). https://pith.science/paper/MFMK5OSG

@misc{pith2026250202668,
  author       = {Pith},
  title        = {Pith review of: Recovering Imbalanced Clusters via Gradient-Based Projection Pursuit},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MFMK5OSG}},
  note         = {Machine review of arXiv:2502.02668}
}
read the original abstract

Projection Pursuit is a classic exploratory technique for finding interesting projections of a dataset. We propose a method for recovering projections containing either Imbalanced Clusters or a Bernoulli-Rademacher distribution using a gradient-based technique to optimize the projection index. As sample complexity is a major limiting factor in Projection Pursuit, we analyze our algorithm's sample complexity within a Planted Vector setting where we can observe that Imbalanced Clusters can be recovered more easily than balanced ones. Additionally, we give a generalized result that works for a variety of data distributions and projection indices. We compare these results to computational lower bounds in the Low-Degree-Polynomial Framework. Finally, we experimentally evaluate our method's applicability to real-world data using FashionMNIST and the Human Activity Recognition Dataset, where our algorithm outperforms others when only a few samples are available.

Figures

Figures reproduced from arXiv: 2502.02668 by the authors.

Figure 1
Figure 1. Phase transitions in Projection Pursuit using gradient-based algorithms. The horizontal and vertical axes correspond to log2 d and log2 n. Each pixel shows the average value of the absolute inner product between the predicted and signal directions, where white corresponds to 1 and black to 0. 4.2. Comparison to Other Methods Here, we will apply Algorithm 1 to the planted vector problem with other projection indices.… view at source ↗
Figure 2
Figure 2. Methods used in the empirical comparison. On the left: Methods using the gradient-based algorithm. On the right: Spectral methods. ReLU2 and Kurtosis correspond to the projection indices we study in Theorem 1 and Theorem 2. Additionally, 9 [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Comparison of different methods in the planted vector setting. We plot the average inner product between the signal direction and the recovered direction by each algorithm over 30 datasets. 4.3. Experiments with Real Data We compare the algorithms by measuring their performance on two different datasets: FashionMNIST [37] and the Human Activity Recognition Dataset [16]. The FashionMNIST dataset contains 28 × 28 gray… view at source ↗
Figures from the paper (3 more)
Figure 5
Figure 5. Figure 5: A comparison of projection pursuit approaches on the Human Activity Recognition using Smartphones Benchmark. We plot the achieved information gain using projections produced by different projection indices. The box plots are generated with the 30 candidate projections …
Figure 4
Figure 4. Figure 4: A comparison of projection pursuit approaches on fashionMNIST. We plot the achieved information gain using projections produced by different projection indices. The box plots are generated with the 30 candidate projections generated by the algorithms. For demonstration…
Figure 6
Figure 6. Figure 6: Histograms of projections recovered by Algorithm 1 on FashionMNIST using n = 600 samples. Classes are plotted using different colors. cluster positioned towards one extreme. 5. Conclusion We consider the performance of gradient-based algorithms for projection pursuit i…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 35 canonical work pages

  1. [1]

    N. Alon, M. Krivelevich, B. Sudakov, Finding a large hidd en clique in a random graph, Random Structures & Algorithms 1 3 (1998) 457–466

    work page 1998

  2. [2]

    J. M. Arevalillo, H. Navarro, Skewness-kurtosis model- based projection pursuit with application to summarizing g ene expression data, Math- ematics 9 (2021)

    work page 2021

  3. [3]

    Y . Bai, Q. Jiang, J. Sun, Subgradient descent learns orth ogonal dictionaries, openreview.org, in: 7th Internation al Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2 019, pp. 1–37

    work page 2019

  4. [4]

    D. P . Bertsekas, Incremental least squares methods and t he extended Kalman filter, SIAM Journal on Optimization 6 (19 96) 807–822

    work page

  5. [5]

    Cabana, R

    E. Cabana, R. E. Lillo, H. Laniado, Multivariate outlier detection based on a robust mahalanobis distance with shrinkage estimators, Statistical Papers 62 (2021) 1583–1609

    work page 2021

  6. [6]

    Davis, M

    D. Davis, M. Diaz, K. Wang, Clustering a mixture of gaussi ans with unknown covariance, arXiv preprint arXiv:2110.01 602 (2021). 12

    work page 2021

  7. [7]

    Diakonikolas, S

    I. Diakonikolas, S. Karmalkar, S. Pang, A. Potechin, Sum -of-squares lower bounds for non-gaussian component analy sis, IEEE, in: 2024 IEEE 65th Annual Symposium on Foundations of Computer Scien ce (FOCS), pp. 949–958

    work page 2024

  8. [8]

    Dudeja, D

    R. Dudeja, D. Hsu, Statistical-Computational Trade-o ffs in Tensor PCA and Related Problems via Communication Complexity, arXiv e-prints (2022) arXiv:2204.07526

    arXiv 2022

Show all 40 references
  1. [9]

    Friedman, J

    J. Friedman, J. Tukey, A projection pursuit algorithm fo r exploratory data analysis, IEEE Transactions on Computer s C-23 (1974) 881–890

    1974

  2. [10]

    J. H. Friedman, Exploratory projection pursuit, Journ al of the American Statistical Association 82 (1987) 249–26 6

    1987

  3. [11]

    S. B. Hopkins, T. Schramm, J. Shi, D. Steurer, Speeding u p sum-of-squares for tensor decomposition and planted spar se vectors, CoRR abs/1512.02337 (2015)

    2015 arXiv

  4. [12]

    S. B. K. Hopkins, Statistical inference and the sum of sq uares method, Ph.D. thesis, Cornell University, 2018

    2018

  5. [13]

    A. Hyv¨ arinen, New approximations of di fferential entropy for independent component analysis and pr ojection pursuit, Advances in Neural Information Processing Systems 10 (NIPS 1997), volume 10, M IT Press, 1997, pp. 273–279

    1997

  6. [14]

    Hyvarinen, Fast and robust fixed-point algorithms fo r independent component analysis, IEEE Transactions on Neu ral Networks 10 (1999) 626–634

    A. Hyvarinen, Fast and robust fixed-point algorithms fo r independent component analysis, IEEE Transactions on Neu ral Networks 10 (1999) 626–634

    1999

  7. [15]

    Hyv¨ arinen, J

    A. Hyv¨ arinen, J. Karhunen, E. Oja, Ica by maximization of nongaussianity, John Wiley & Sons, Ltd, 2001, pp. 165–202

    2001

  8. [16]

    Jorge, A

    R.-O. Jorge, A. Davide, G. Alessandro, O. Luca, P . Xavie r, Human Activity Recognition Using Smartphones, UCI Machi ne Learning Repos- itory, 2013. DOI: https://doi.org/10.24432/C54S4K

    2013 doi

  9. [17]

    Kearns, E fficient noise-tolerant learning from statistical queries, J

    M. Kearns, E fficient noise-tolerant learning from statistical queries, J . ACM 45 (1998) 983–1006

    1998

  10. [18]

    Kollo, Multivariate skewness and kurtosis measures with an application in ica, Journal of Multivariate Analysi s 99 (2008) 2328–2338

    T. Kollo, Multivariate skewness and kurtosis measures with an application in ica, Journal of Multivariate Analysi s 99 (2008) 2328–2338

    2008

  11. [19]

    Kunisky, A

    D. Kunisky, A. S. Wein, A. S. Bandeira, Notes on Computat ional Hardness of Hypothesis Testing: Predictions using th e Low-Degree Likelihood Ratio, arXiv e-prints (2019) arXiv:1907.11636

    2019 arXiv

  12. [20]

    Loperfido, Singular value decomposition of the third multivariate moment, Linear Algebra and its Applications 4 73 (2015) 202–216

    N. Loperfido, Singular value decomposition of the third multivariate moment, Linear Algebra and its Applications 4 73 (2015) 202–216. Special issue on Statistics

    2015

  13. [21]

    Loperfido, Kurtosis maximization for outlier detect ion in garch models: Maf 2018, 2018, pp

    N. Loperfido, Kurtosis maximization for outlier detect ion in garch models: Maf 2018, 2018, pp. 455–459

    2018

  14. [22]

    Loperfido, Skewness-based projection pursuit: A com putational approach, Computational Statistics and Data An alysis 120 (2018) 42–57

    N. Loperfido, Skewness-based projection pursuit: A com putational approach, Computational Statistics and Data An alysis 120 (2018) 42–57

    2018

  15. [23]

    Loperfido, Kurtosis-based projection pursuit for ou tlier detection in financial time series, The European Journ al of Finance 26 (2020) 142–164

    N. Loperfido, Kurtosis-based projection pursuit for ou tlier detection in financial time series, The European Journ al of Finance 26 (2020) 142–164

    2020

  16. [24]

    C. Mao, A. S. Wein, Optimal Spectral Recovery of a Plante d V ector in a Subspace, arXiv e-prints (2021) arXiv:2105.15081

    2021 arXiv

  17. [25]

    R. Meka, A. Potechin, A. Wigderson, Sum-of-squares low er bounds for planted clique, CoRR abs /1503.06447 (2015)

    2015 arXiv

  18. [26]

    Neyman, E

    J. Neyman, E. S. Pearson, K. Pearson, Ix. on the problem o f the most e fficient tests of statistical hypotheses, Philosophical Tran sactions of the Royal Society of London. Series A, Containing Papers of a Mathematical or Physical Character 231 (1933) 289–337

    1933

  19. [27]

    Pe˜ na, F

    D. Pe˜ na, F. J. Prieto, The kurtosis coe fficient and the linear discriminant function, Statistics & Pr obability Letters 49 (2000) 257–261

    2000

  20. [28]

    Pe˜ na, F

    D. Pe˜ na, F. J. Prieto, Cluster identification using pro jections, Journal of the American Statistical Association 96 (2001) 1433–1445

    2001

  21. [29]

    Pe˜ na, F

    D. Pe˜ na, F. J. Prieto, Multivariate outlier detection and robust covariance matrix estimation, Technometrics 43 (2001) 286–310

    2001

  22. [30]

    Pe˜ na, F

    D. Pe˜ na, F. J. Prieto, Combining random and specific dir ections for outlier detection and robust estimation in high -dimensional multivariate data, Journal of Computational and Graphical Statistics 16 (2007) 228–254

    2007

  23. [31]

    Pe˜ na, F

    D. Pe˜ na, F. J. Prieto, J. Viladomat, Eigenvectors of a kurtosis matrix as interesting directions to reveal cluster structure, Journal of Multivariate Analysis 101 (2010) 1995–2007

    2010

  24. [32]

    Pokojovy, J

    M. Pokojovy, J. M. Jobe, A robust deterministic a ffine-equivariant algorithm for multivariate location and sc atter, Computational Statistics & Data Analysis 172 (2022) 107475

    2022

  25. [33]

    P¨ a¨ aj¨ arvi, J

    P . P¨ a¨ aj¨ arvi, J. LeBlanc, Skewness maximization forimpulsive sources in blind deconvolution, Report - Helsink i University of Technology, Signal Processing Laboratory 46 (2004) 304–307

    2004

  26. [34]

    Q. Qu, J. Sun, J. Wright, Finding a sparse vector in a subs pace: Linear sparsity using alternating directions, Advan ces in Neural Information Processing Systems 4 (2014)

    2014

  27. [35]

    D. A. Spielman, H. Wang, J. Wright, Exact recovery of spa rsely-used dictionaries, PMLR, in: Proceedings of the 25th Annual Conference on Learning Theory, volume 23 of Proceedings of Machine Learni ng Research, Edinburgh, Scotland, pp. 37.1–37.18

  28. [36]

    V ershynin, High-Dimensional Probability, 2020

    R. V ershynin, High-Dimensional Probability, 2020

    2020

  29. [37]

    H. Xiao, K. Rasul, R. V ollgraf, Fashion-mnist: a novel i mage dataset for benchmarking machine learning algorithms , CoRR abs /1708.07747 (2017)

    2017 arXiv

  30. [38]

    Zadik, M

    I. Zadik, M. J. Song, A. S. Wein, J. Bruna, Lattice-based methods surpass sum-of-squares in clustering, PMLR, in: P . -L. Loh, M. Raginsky (Eds.), Proceedings of Thirty Fifth Conference on Learning Theory, volume 178 of Proceedings of Machine Learning Resea rch, pp. 1247– 1248

  31. [39]

    Zarzoso, P

    V . Zarzoso, P . Comon, Comparative speed analysis of fas tica, in: M. E. Davies, C. J. James, S. A. Abdallah, M. D. Plumb ley (Eds.), Independent Component Analysis and Signal Separation, vol ume 4666 of Lecture Notes in Computer Science, Springer, Ber lin, Heidelberg, 2007, ...

    2007

  32. [40]

    Y . Zhai, Z. Y ang, Z. Liao, J. Wright, Y . Ma, Complete dict ionary learning via l4-norm maximization over the orthogon al group, J. Mach. Learn. Res. 21 (2020). Appendix A. Proof of Lemma 1 Proof. Start by choosing an initialization j∈ [ninit] for which ⟨ u0, j, u∗ ⟩ ≥ a whi...

    2020

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.