pith. sign in

arxiv: 2109.11142 · v4 · submitted 2021-09-23 · 📊 stat.ME · math.ST· stat.TH

Sparse PCA: A New Scalable Estimator Based On Integer Programming

Kayhan Behdin , Rahul Mazumder This is my paper

Pith reviewed 2026-05-24 13:51 UTC · model grok-4.3

classification 📊 stat.ME math.STstat.TH
keywords sparse PCAmixed integer programmingspiked covariance modelsupport recoveryestimation error boundsscalable optimizationGaussian noise model
0
0 comments X

The pith

A new mixed integer program for sparse PCA uses the spiked covariance model to scale to 20,000 features with statistical guarantees on error and support recovery.

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

The paper formulates a new estimator for sparse principal component analysis as a mixed integer program. It derives the formulation from the spiked covariance model and properties of the multivariate Gaussian distribution, then proves guarantees for estimation error and support recovery. The guarantees continue to hold under departures from the exact model and when the optimization problem is solved only approximately. A custom solver for the program runs on instances with up to 20,000 features in minutes and produces results with competitive statistical properties on both synthetic and real data.

Core claim

The central claim is that a mixed integer program derived from the spiked covariance model yields a sparse PCA estimator with provable bounds on estimation error and support recovery; the program can be solved at scale with a custom algorithm that outperforms earlier exact MIP approaches while the statistical guarantees remain valid for approximate solutions and modest model misspecification.

What carries the argument

A mixed integer program that encodes the selection of sparse principal components by exploiting the structure of the spiked covariance model and the multivariate Gaussian likelihood.

If this is right

  • The estimator recovers the true support with high probability under the spiked model.
  • Statistical guarantees continue to apply when the MIP is solved only to moderate accuracy.
  • The same guarantees hold when the data deviate mildly from the exact spiked model.
  • Problems with twenty thousand features become computationally feasible on standard hardware.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same model-based MIP construction may be adaptable to other sparse estimation tasks that admit a low-rank plus noise decomposition.
  • If warm-start or branching heuristics can be further improved, the method could reach dimensions beyond twenty thousand features.
  • Because real data rarely match the spiked model exactly, empirical checks on non-Gaussian or multi-spike covariances would clarify practical robustness.

Load-bearing premise

The data are generated from (or are close to) a spiked covariance model with multivariate Gaussian noise.

What would settle it

A simulation in which data are drawn from a covariance matrix lacking a single dominant spike and the estimator recovers an incorrect support set with high probability.

Figures

Figures reproduced from arXiv: 2109.11142 by Kayhan Behdin, Rahul Mazumder.

Figure 1
Figure 1. Figure 1: Estimation and support recovery error as available from different methods on synthetic data with p = 10000 and different values of n (along the x-axis), as discussed in the text (Section 5.1.2). The left panels show results with s = 5 and right panels s = 10. The top panels compare estimation performance and bottom ones compare support recovery. Our proposed approach results in high-quality estimation perf… view at source ↗
Figure 2
Figure 2. Figure 2: Numerical results for the synthetic dataset with p = 10000, n = 7500 in Section 5.1.3. The left panel shows the estimation performance and the right panel shows the support recovery performance. s = 5 s = 10 |sin ∠(ˆu, u ∗ )| 0.025 0.050 0.075 0.100 5000 10000 15000 20000 n method CovThresh SPCA−SLS SPCA−SLSR SPCAvSLR 0.0 0.2 0.4 0.6 5000 10000 15000 20000 n method CovThresh SPCA−SLS SPCA−SLSR SPCAvSLR [P… view at source ↗
Figure 3
Figure 3. Figure 3: Numerical results for the synthetic dataset with p = 20000 in Section 5.1.4. The left panel shows the statistical performance for s = 5. The right panel shows the statistical performance for s = 10. λ chosen as in Section 5.1.1), TrunPow, SPCAvSLR and CovTresh for this data matrix and two values of s ∈ {4, 5}. Parameter selection is done similar to the synthetic data experiments in Section 5.1. We use the … view at source ↗
Figure 4
Figure 4. Figure 4: Comparison of estimated PCs by different algorithms for the real dataset in Section 5.2. The left panel shows the results for s = 4 and the right panel shows the results for s = 5. In both cases, SPCA-SLS reaches the optimality gap of less than 12% and SPCA-SLSR reaches the optimality gap of less than 10% after 10 minutes. to a denser support. 6 Conclusion In this paper, we consider a discrete optimization… view at source ↗
read the original abstract

We consider the Sparse Principal Component Analysis (SPCA) problem under the well-known spiked covariance model. Recent work has shown that the SPCA problem can be reformulated as a Mixed Integer Program (MIP) and can be solved to global optimality, leading to estimators that are known to enjoy optimal statistical properties. However, prior MIP algorithms for SPCA appear to be limited in terms of scalability to up to a thousand features or so. In this paper, we propose a new estimator for SPCA which can be formulated as a MIP. Different from earlier work, we make use of the underlying spiked covariance model and properties of the multivariate Gaussian distribution to arrive at our estimator. We establish statistical guarantees for our proposed estimator in terms of estimation error and support recovery. We derive guarantees under departures from the spiked covariance model, and for approximate solutions to the optimization problem. We propose a custom algorithm to solve the MIP, which scales better than off-the-shelf solvers, and demonstrate that our approach can be much more computationally attractive compared to earlier exact MIP-based approaches for the SPCA problem. Our numerical experiments on synthetic and real datasets show that our algorithms can address problems with up to 20,000 features in minutes; and generally result in favorable statistical properties compared to existing popular approaches for SPCA.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a new mixed-integer programming (MIP) formulation for sparse principal component analysis (SPCA) under the spiked covariance model. It derives the MIP objective using properties of the multivariate Gaussian distribution, establishes statistical guarantees for estimation error and support recovery both under the exact model and under departures from it (as well as for approximate solutions), introduces a custom solver claimed to scale to 20,000 features, and reports favorable performance versus prior MIP and other SPCA methods on synthetic and real data.

Significance. If the statistical guarantees are correctly established and the custom solver delivers the claimed scalability without sacrificing the exactness properties, the work would supply a practically usable, theoretically supported exact method for high-dimensional SPCA that improves on the scalability limits of earlier MIP formulations (previously limited to roughly 1,000 features).

major comments (2)
  1. [Estimator derivation and guarantees under departures] The section deriving the estimator and the subsequent section on statistical guarantees under departures: the MIP objective is obtained by explicit use of the multivariate Gaussian likelihood under the spiked covariance model; the departure analysis is described as controlling bias terms, but it is not shown whether this preserves the exact equivalence between the original optimization problem and the MIP formulation. If the MIP form is tied to Gaussianity, the support-recovery guarantee may fail for distributions with infinite fourth moments even when the covariance remains spiked.
  2. [Guarantees for approximate solutions] The section on approximate solutions: the claim that statistical guarantees extend to approximate MIP solutions is load-bearing for the practical utility of the custom solver, yet the error bounds relating the approximation gap to the estimation and support-recovery errors are not stated explicitly enough to verify whether they remain useful when the solver is terminated early on large instances.
minor comments (2)
  1. [Abstract and experimental results] The abstract states that the approach 'can address problems with up to 20,000 features in minutes' but does not specify the hardware, MIP solver tolerances, or exact timing metrics used in the experiments; this detail belongs in the experimental section for reproducibility.
  2. [Notation and model setup] Notation for the spiked model parameters (e.g., spike strength, noise variance) should be introduced once and used consistently when stating the estimation-error bounds.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive feedback on our manuscript. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

  1. Referee: [Estimator derivation and guarantees under departures] The section deriving the estimator and the subsequent section on statistical guarantees under departures: the MIP objective is obtained by explicit use of the multivariate Gaussian likelihood under the spiked covariance model; the departure analysis is described as controlling bias terms, but it is not shown whether this preserves the exact equivalence between the original optimization problem and the MIP formulation. If the MIP form is tied to Gaussianity, the support-recovery guarantee may fail for distributions with infinite fourth moments even when the covariance remains spiked.

    Authors: The MIP formulation is derived from the exact Gaussian likelihood under the spiked model, so the equivalence to the original problem holds under that model. For departures, we bound the difference between the Gaussian objective and the true (non-Gaussian) objective by a bias term whose size is controlled under finite fourth-moment assumptions; the MIP is then solved exactly on the Gaussian objective, and the resulting estimator is shown to be close to the true sparse vector via this bias control. The support-recovery guarantee under departures therefore requires the same moment conditions used for the bias bound. We will revise the manuscript to state these moment conditions explicitly and to clarify that the equivalence is with respect to the Gaussian objective (not the departed distribution). revision: partial

  2. Referee: [Guarantees for approximate solutions] The section on approximate solutions: the claim that statistical guarantees extend to approximate MIP solutions is load-bearing for the practical utility of the custom solver, yet the error bounds relating the approximation gap to the estimation and support-recovery errors are not stated explicitly enough to verify whether they remain useful when the solver is terminated early on large instances.

    Authors: We agree that the dependence of the statistical error on the MIP approximation gap should be stated more explicitly. In the current draft we show that an additive optimality gap of size ε increases the estimation error by a term linear in ε (scaled by problem constants) and that support recovery continues to hold provided ε is smaller than a gap proportional to the minimal signal strength. We will expand this section with the precise constants and a short discussion of how large an ε remains tolerable for early termination on instances with 10,000–20,000 features. revision: yes

Circularity Check

0 steps flagged

No circularity: MIP estimator and guarantees derived from external spiked covariance model

full rationale

The paper formulates its MIP estimator by invoking the standard spiked covariance model plus multivariate Gaussian properties (abstract), which are external to the paper and not self-defined or fitted from the target data. Guarantees for estimation error and support recovery are stated under the model and explicit departures, without reducing to a fitted parameter or self-citation chain. No equations or steps in the provided text exhibit self-definitional equivalence, fitted-input-as-prediction, or load-bearing self-citation that collapses the derivation to its inputs by construction. The central claim remains independent of the paper's own outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review; the central claim rests on the spiked covariance model and Gaussian assumptions used to derive both the MIP and the guarantees. No free parameters or invented entities are mentioned.

axioms (1)
  • domain assumption Data follows (or is close to) the spiked covariance model with multivariate Gaussian distribution
    Invoked to arrive at the estimator and to establish statistical guarantees (abstract).

pith-pipeline@v0.9.0 · 5762 in / 1163 out tokens · 31600 ms · 2026-05-24T13:51:38.601553+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    u rk, Alper Atamt \

    M Selim Akt \"u rk, Alper Atamt \"u rk, and Sinan G \"u rel. A strong conic quadratic reformulation for machine-job assignment with controllable processing times. Operations Research Letters, 37 0 (3): 0 187--191, 2009

    work page 2009

  2. [2]

    Amini and Martin J

    Arash A. Amini and Martin J. Wainwright. High-dimensional analysis of semidefinite relaxations for sparse principal components. Ann. Statist., 37 0 (5B): 0 2877--2921, 10 2009. doi:10.1214/08-AOS664. URL https://doi.org/10.1214/08-AOS664

    work page doi:10.1214/08-aos664 2009

  3. [3]

    Archetypal analysis for sparse nonnegative matrix factorization: Robustness under misspecification

    Kayhan Behdin and Rahul Mazumder. Archetypal analysis for sparse nonnegative matrix factorization: Robustness under misspecification. arXiv preprint arXiv:2104.03527, 2021

    work page arXiv 2021

  4. [4]

    Certifiably optimal sparse principal component analysis

    Lauren Berk and Dimitris Bertsimas. Certifiably optimal sparse principal component analysis. Mathematical Programming Computation, 11 0 (3): 0 381--420, 2019

    work page 2019

  5. [5]

    Optimal detection of sparse principal components in high dimension

    Quentin Berthet and Philippe Rigollet. Optimal detection of sparse principal components in high dimension. Ann. Statist., 41 0 (4): 0 1780--1815, 08 2013. doi:10.1214/13-AOS1127. URL https://doi.org/10.1214/13-AOS1127

    work page doi:10.1214/13-aos1127 2013

  6. [6]

    Nonlinear programming

    Dimitri P Bertsekas. Nonlinear programming. Journal of the Operational Research Society, 48 0 (3): 0 334--334, 1997

    work page 1997

  7. [7]

    Sparse high-dimensional regression: Exact scalable algorithms and phase transitions

    Dimitris Bertsimas and Bart Van Parys. Sparse high-dimensional regression: Exact scalable algorithms and phase transitions. The Annals of Statistics, 48 0 (1): 0 300--323, 2020

    work page 2020

  8. [8]

    Best subset selection via a modern optimization lens

    Dimitris Bertsimas, Angela King, and Rahul Mazumder. Best subset selection via a modern optimization lens. The annals of statistics, 44 0 (2): 0 813--852, 2016

    work page 2016

  9. [9]

    Solving large-scale sparse pca to certifiable (near) optimality

    Dimitris Bertsimas, Ryan Cory-Wright, and Jean Pauphilet. Solving large-scale sparse pca to certifiable (near) optimality. arXiv preprint arXiv:2005.05195, 2020

    work page arXiv 2005

  10. [10]

    Convex optimization

    Stephen Boyd and Lieven Vandenberghe. Convex optimization. Cambridge university press, 2004

    work page 2004

  11. [11]

    Sparse pca from sparse linear regression

    Guy Bresler, Sung Min Park, and Madalina Persu. Sparse pca from sparse linear regression. In Proceedings of the 32nd International Conference on Neural Information Processing Systems, NIPS’18, page 10965–10975, Red Hook, NY, USA, 2018. Curran Associates Inc

    work page 2018

  12. [12]

    Tony Cai and Harrison H

    T. Tony Cai and Harrison H. Zhou. Optimal rates of convergence for sparse covariance matrix estimation. Ann. Statist., 40 0 (5): 0 2389--2420, 10 2012. doi:10.1214/12-AOS998. URL https://doi.org/10.1214/12-AOS998

    work page doi:10.1214/12-aos998 2012

  13. [13]

    A direct formulation for sparse pca using semidefinite programming

    Alexandre d'Aspremont, Laurent E Ghaoui, Michael I Jordan, and Gert R Lanckriet. A direct formulation for sparse pca using semidefinite programming. In Advances in neural information processing systems, pages 41--48, 2005

    work page 2005

  14. [14]

    Sparse pca via covariance thresholding

    Yash Deshpande and Andrea Montanari. Sparse pca via covariance thresholding. The Journal of Machine Learning Research, 17 0 (1): 0 4913--4953, 2016

    work page 2016

  15. [15]

    A convex integer programming approach for optimal sparse pca

    Santanu S Dey, Rahul Mazumder, and Guanyi Wang. A convex integer programming approach for optimal sparse pca. arXiv preprint arXiv:1810.09062, 2018

    work page arXiv 2018

  16. [16]

    An outer-approximation algorithm for a class of mixed-integer nonlinear programs

    Marco A Duran and Ignacio E Grossmann. An outer-approximation algorithm for a class of mixed-integer nonlinear programs. Mathematical programming, 36 0 (3): 0 307--339, 1986

    work page 1986

  17. [17]

    Optimal solutions for sparse principal component analysis

    Alexandre d’Aspremont, Francis Bach, and Laurent El Ghaoui. Optimal solutions for sparse principal component analysis. Journal of Machine Learning Research, 9 0 (Jul): 0 1269--1294, 2008

    work page 2008

  18. [18]

    When is best subset selection the" best"? arXiv preprint arXiv:2007.01478, 2020

    Jianqing Fan, Yongyi Guo, and Ziwei Zhu. When is best subset selection the" best"? arXiv preprint arXiv:2007.01478, 2020

    work page arXiv 2007

  19. [19]

    Frangioni and C

    A. Frangioni and C. Gentile. Perspective cuts for a class of convex 0–1 mixed integer programs. Math. Program., 106 0 (2): 0 225–236, May 2006. ISSN 0025-5610

    work page 2006

  20. [20]

    Sparse principal component analysis via axis-aligned random projections

    Milana Gataric, Tengyao Wang, and Richard J Samworth. Sparse principal component analysis via axis-aligned random projections. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 82 0 (2): 0 329--359, 2020

    work page 2020

  21. [21]

    Perspective reformulations of mixed integer nonlinear programs with indicator variables

    Oktay G \"u nl \"u k and Jeff Linderoth. Perspective reformulations of mixed integer nonlinear programs with indicator variables. Mathematical programming, 124 0 (1): 0 183--205, 2010

    work page 2010

  22. [22]

    Result analysis of the nips 2003 feature selection challenge

    Isabelle Guyon, Steve Gunn, Asa Ben-Hur, and Gideon Dror. Result analysis of the nips 2003 feature selection challenge. In Advances in Neural Information Processing Systems, volume 17. MIT Press, 2005

    work page 2003

  23. [23]

    Statistical learning with sparsity: the lasso and generalizations

    Trevor Hastie, Robert Tibshirani, and Martin Wainwright. Statistical learning with sparsity: the lasso and generalizations. Chapman and Hall/CRC, 2019

    work page 2019

  24. [24]

    Fast best subset selection: Coordinate descent and local combinatorial optimization algorithms

    Hussein Hazimeh and Rahul Mazumder. Fast best subset selection: Coordinate descent and local combinatorial optimization algorithms. Operations Research, 68 0 (5): 0 1517--1537, 2020

    work page 2020

  25. [25]

    Sparse regression at scale: Branch-and-bound rooted in first-order optimization

    Hussein Hazimeh, Rahul Mazumder, and Ali Saab. Sparse regression at scale: Branch-and-bound rooted in first-order optimization. arXiv preprint arXiv:2004.06152, 2020

    work page arXiv 2004

  26. [26]

    Analysis of a complex of statistical variables into principal components

    Harold Hotelling. Analysis of a complex of statistical variables into principal components. Journal of educational psychology, 24 0 (6): 0 417, 1933

    work page 1933

  27. [27]

    Johnstone

    Iain M. Johnstone. On the distribution of the largest eigenvalue in principal components analysis. Ann. Statist., 29 0 (2): 0 295--327, 04 2001. doi:10.1214/aos/1009210544. URL https://doi.org/10.1214/aos/1009210544

    work page doi:10.1214/aos/1009210544 2001

  28. [28]

    On consistency and sparsity for principal components analysis in high dimensions

    Iain M Johnstone and Arthur Yu Lu. On consistency and sparsity for principal components analysis in high dimensions. Journal of the American Statistical Association, 104 0 (486): 0 682--693, 2009

    work page 2009

  29. [29]

    A modified principal component technique based on the lasso

    Ian T Jolliffe, Nickolay T Trendafilov, and Mudassir Uddin. A modified principal component technique based on the lasso. Journal of computational and Graphical Statistics, 12 0 (3): 0 531--547, 2003

    work page 2003

  30. [30]

    Do semidefinite relaxations solve sparse pca up to the information limit? The Annals of Statistics, 43 0 (3): 0 1300--1322, 2015

    Robert Krauthgamer, Boaz Nadler, Dan Vilenchik, et al. Do semidefinite relaxations solve sparse pca up to the information limit? The Annals of Statistics, 43 0 (3): 0 1300--1322, 2015

    work page 2015

  31. [31]

    Jing Lei and Vincent Q. Vu. Sparsistency and agnostic inference in sparse PCA . The Annals of Statistics, 43 0 (1): 0 299 -- 322, 2015

    work page 2015

  32. [32]

    Exact and approximation algorithms for sparse pca

    Yongchun Li and Weijun Xie. Exact and approximation algorithms for sparse pca. arXiv preprint arXiv:2008.12438, 2020

    work page arXiv 2008

  33. [33]

    Conditional gradient algorithms for rank-one matrix approximations with a sparsity constraint

    Ronny Luss and Marc Teboulle. Conditional gradient algorithms for rank-one matrix approximations with a sparsity constraint. siam REVIEW, 55 0 (1): 0 65--98, 2013

    work page 2013

  34. [34]

    High-dimensional graphs and variable selection with the Lasso

    Nicolai Meinshausen and Peter Bühlmann. High-dimensional graphs and variable selection with the Lasso . The Annals of Statistics, 34 0 (3): 0 1436 -- 1462, 2006

    work page 2006

  35. [35]

    Spectral bounds for sparse pca: Exact and greedy algorithms

    Baback Moghaddam, Yair Weiss, and Shai Avidan. Spectral bounds for sparse pca: Exact and greedy algorithms. In Advances in neural information processing systems, pages 915--922, 2006

    work page 2006

  36. [36]

    Asymptotics of sample eigenstructure for a large dimensional spiked covariance model

    Debashis Paul. Asymptotics of sample eigenstructure for a large dimensional spiked covariance model. Statistica Sinica, 17 0 (4): 0 1617--1642, 2007

    work page 2007

  37. [37]

    Raskutti , M

    G. Raskutti , M. J. Wainwright , and B. Yu . Minimax rates of estimation for high-dimensional linear regression over _q -balls. IEEE Transactions on Information Theory, 57 0 (10): 0 6976--6994, 2011

    work page 2011

  38. [38]

    Alternating maximization: Unifying framework for 8 sparse pca formulations and efficient parallel codes

    Peter Richt \'a rik, Majid Jahani, Selin Damla Ahipa s ao g lu, and Martin Tak \'a c . Alternating maximization: Unifying framework for 8 sparse pca formulations and efficient parallel codes. Optimization and Engineering, pages 1--27, 2020

    work page 2020

  39. [39]

    High dimensional statistics

    Phillippe Rigollet and Jan-Christian H \"u tter. High dimensional statistics. Lecture notes for course 18S997, 2015

    work page 2015

  40. [40]

    Hanson-Wright inequality and sub-gaussian concentration

    Mark Rudelson and Roman Vershynin. Hanson-Wright inequality and sub-gaussian concentration . Electronic Communications in Probability, 18 0 (none): 0 1 -- 9, 2013

    work page 2013

  41. [41]

    Stewart and Ji-guang Sun

    G.W. Stewart and Ji-guang Sun. Matrix Perturbation Theory. Computer science and scientific computing. Academic Press, 1990

    work page 1990

  42. [42]

    High-dimensional probability: An introduction with applications in data science, volume 47

    Roman Vershynin. High-dimensional probability: An introduction with applications in data science, volume 47. Cambridge university press, 2018

    work page 2018

  43. [43]

    Samworth

    Tengyao Wang, Quentin Berthet, and Richard J. Samworth. Statistical and computational trade-offs in estimation of sparse principal components . The Annals of Statistics, 44 0 (5): 0 1896 -- 1930, 2016

    work page 1930

  44. [44]

    Perturbation bounds in connection with singular value decomposition

    Per- ke Wedin. Perturbation bounds in connection with singular value decomposition. BIT Numerical Mathematics, 12 0 (1): 0 99--111, 1972

    work page 1972

  45. [45]

    A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis

    Daniela M Witten, Robert Tibshirani, and Trevor Hastie. A penalized matrix decomposition, with applications to sparse principal components and canonical correlation analysis. Biostatistics, 10 0 (3): 0 515--534, 2009

    work page 2009

  46. [46]

    Truncated power method for sparse eigenvalue problems

    Xiao-Tong Yuan and Tong Zhang. Truncated power method for sparse eigenvalue problems. Journal of Machine Learning Research, 14 0 (Apr): 0 899--925, 2013

    work page 2013

  47. [47]

    The Schur complement and its applications, volume 4

    Fuzhen Zhang. The Schur complement and its applications, volume 4. Springer Science & Business Media, 2006

    work page 2006

  48. [48]

    Sparse principal component analysis

    Hui Zou, Trevor Hastie, and Robert Tibshirani. Sparse principal component analysis. Journal of computational and graphical statistics, 15 0 (2): 0 265--286, 2006

    work page 2006