Pith. sign in

REVIEW 2 major objections 5 minor 51 references

A closed-form matrix-valued regularization parameter for the graphical lasso matches cross-validation accuracy and support recovery while cutting runtime by orders of magnitude.

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 · grok-4.5

2026-07-11 00:55 UTC pith:DIXZ6VRG

load-bearing objection Clean closed-form matrix Λ that fixes reselection probability a priori, hits the √(log p/n) rate, and matches CV accuracy at far lower cost. the 2 major comments →

arxiv 2607.07735 v1 pith:DIXZ6VRG submitted 2026-07-07 stat.ML cs.LG

The Regularization Parameter: Sparse Precision Matrix Estimation

Aryan Eftekhari , Daniel Sergio Vega , Ernst-Jan Camiel Wit , Olaf Schenk This is my paper
classification stat.ML cs.LG MSC 62H1062H1265F22
keywords sparse precision matrixinverse covarianceGaussian graphical modelsregularization parametersampling distributionhigh-dimensional statistics
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.

Estimating a sparse precision matrix from high-dimensional data requires choosing a regularization parameter that balances sparsity against estimation error; the usual route is costly cross-validation. This paper derives a closed-form matrix-valued parameter directly from the sampling distribution of the first-order optimality conditions of the ℓ1-regularized Gaussian maximum-likelihood estimator. By fixing in advance the probability that each nonzero entry continues to satisfy its optimality condition under resampling, the method removes any need for repeated solves or candidate grids. The resulting parameter automatically attains the asymptotic rate that theory already links to consistency and exact support recovery. On synthetic Gaussian and heavy-tailed data, as well as gene-expression and brain-connectivity networks, the estimator matches or exceeds the accuracy of cross-validated baselines while running several orders of magnitude faster.

Core claim

The paper shows that the single matrix formula Λ_ij = erf^{-1}((p-c)/p) √(2 V_ij / k) produces a regularization parameter whose entries scale as √(log p / n). Under standard conditions this scaling is already known to guarantee that the graphical-lasso estimator is consistent and sparsistent; the same formula, obtained by prescribing a constant reselection probability for nonzero entries and a vanishing false-selection probability for zero entries, yields empirical accuracy and support recovery comparable to cross-validation at a fraction of the computational cost.

What carries the argument

The reselection probability: the survival function of the folded-normal distribution of the resampled gradient evaluated at the point where |G_ij| equals Λ_ij. Setting this probability equal to 1/2 + γ and enforcing dimensional invariance by letting γ shrink as c/p produces the closed-form matrix Λ that is used for every subsequent solve.

Load-bearing premise

The variance that appears inside the closed-form formula is derived under the assumption that the observations are exactly Gaussian; for other distributions the same expression is retained and only the effective sample size is reduced.

What would settle it

On a large non-Gaussian data set, compute the empirical frequency with which the nonzero entries of the estimator continue to satisfy the first-order condition under repeated resampling; if that frequency systematically departs from the prescribed value 1/2 + γ_c, the claim that the formula correctly controls reselection probability is false.

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

2 major / 5 minor

Summary. The paper proposes a closed-form, matrix-valued regularization parameter for the ℓ₁-regularized Gaussian MLE (graphical lasso). Starting from the first-order optimality conditions, it models the sample-mean gradient under resampling as approximately Gaussian, obtains the folded-normal survival function (selection probability), and inverts the reselection probability for nonzero entries to produce Λ_ij = erf^{-1}((p-c)/p) √(2 V_ij / k) (Eq. 3.14). The construction is shown to recover the asymptotic rate √(log p / n) that, under the conditions of Ravikumar et al. (2011), yields consistency and sparsistency. Empirically, the resulting aquic implementation matches or exceeds cross-validated glasso/QUIC, CLIME, and TIGER on Gaussian and Student-t synthetic graphs as well as gene-microarray and fMRI data, while reducing runtime by orders of magnitude.

Significance. Selecting the regularization parameter remains a practical bottleneck for high-dimensional precision-matrix estimation; a closed-form, algorithm-agnostic alternative that eliminates cross-validation while preserving the known consistency rate is therefore of clear interest. The derivation is clean, the asymptotic argument is standard, and the empirical package (code, multi-distribution synthetic suites, two real applications, standard-deviation bands) is unusually thorough for the area. The free parameter c is a transparent sparsity proxy rather than a hidden tuning knob. Even with the acknowledged Gaussian limitation on V_ij, the work supplies a usable, reproducible tool that can be adopted immediately.

major comments (2)
  1. Section 3.1 / Eq. (3.3): the closed-form variance V_ij = S̄_ii S̄_jj + S̄_ij^{2} is exact only under Gaussianity (Isserlis–Wick). For non-Gaussian data the same formula is retained and only k is reduced. While the asymptotic rate (3.20) does not depend on this expression, the finite-sample claim that each nonzero entry attains a prescribed reselection probability ½+γ is theoretically uncontrolled outside the Gaussian case. A short remark or a non-Gaussian variance estimator (e.g., sample fourth-moment plug-in) would strengthen the probabilistic interpretation without altering the main construction.
  2. Section 3.3 / Eq. (3.15): the guideline s ≤ 3c is presented as a “crude upper bound” that motivates the choice of c. The bound relies on the heuristic s ≤ s_∞ together with the asymptotic relation z̄_∞ = s_∞/2 + c. Because s_∞ is unknown a priori, the practical recommendation “choose c of order tens” remains informal. A brief sensitivity study or a data-driven default for c (for example via a single pilot estimate of average degree) would make the method more turn-key.
minor comments (5)
  1. Figure 1 caption and surrounding text: the distinction between “reselection probability” and “minimum false-selection probability” is clear once the reader reaches Eqs. (3.5)–(3.8), but the figure itself would benefit from an explicit legend entry for the solid black versus dotted green curves.
  2. Section 4.1.1: the non-Gaussian experiments fix c = 30 while varying k; a short sentence explaining why c is held constant (rather than jointly tuned) would help the reader interpret the trade-off.
  3. Table 1 / Table 2: “Desirable values … shown in bold” is helpful, yet the bolding criterion is not stated (highest F1? lowest NLL?). A one-line note would remove ambiguity.
  4. Typographical: “reseletion” appears once in the abstract; “reselection” is used thereafter. Standardize.
  5. References: the arXiv identifier in the header (2607.07735) is future-dated; confirm the final citation key before publication.

Circularity Check

0 steps flagged

No significant circularity: the closed-form Λ is an algebraic inversion that enforces a prescribed reselection probability, and consistency follows from an external theorem.

full rationale

The derivation chain begins from the first-order optimality conditions (3.1)–(3.2) of the ℓ₁-regularized Gaussian MLE. The selection/reselection probability F is the survival function of the folded-normal sampling distribution of the gradient (3.3)–(3.5). Setting F(Λ; Λ) = 1/2 + γ and solving for Λ (3.6)–(3.7) is simply an algebraic rearrangement that forces the desired probability by construction; the paper states this explicitly (“By defining Λ(γ)ij this way, we will fix the reselection probability at a desired value”). The subsequent dimensional-invariance step that replaces γ by γc so that the false-selection probability equals c/p (3.13)–(3.14) is likewise definitional. The asymptotic claim Λ ∝ √(log p / n) follows from the known expansion of erf⁻¹ (Blair et al.) together with k ∝ n (3.17)–(3.20) and is then fed into the external consistency/sparsistency theorems of Ravikumar et al. (2011). No quantity is fitted to data and later reported as a prediction, no uniqueness theorem is imported from the authors’ own prior work, and the self-citations that appear (SQUIC algorithmic papers) are used only for implementation, not for the theoretical argument. The construction is therefore self-contained and non-circular.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 0 invented entities

The central claim rests on three free parameters (c, k, and the implicit Gaussian variance formula), standard high-dimensional statistics assumptions, and the external consistency theorem of Ravikumar et al. No new physical entities are postulated.

free parameters (3)
  • c (target false-selection count per row) = 1, 5 or 30
    Chosen by hand from the discrete set {1,5,30} on the basis of domain heuristics (e.g., “tens of regulators per gene”); controls both sparsity and the asymptotic false-selection rate ϕ(γc)=c/p.
  • k (resample size) = n/2
    Fixed to n/2 (or occasionally n/4, n/6) without data-driven selection; enters the scale of every Λij.
  • Vij variance formula
    Closed-form expression Vij = S̄ii S̄jj + S̄ij^{2} is exact only under Gaussianity; retained for non-Gaussian data as an unadjusted free modeling choice.
axioms (3)
  • domain assumption Sample-mean gradient entries are asymptotically normal with the Isserlis–Wick variance (Eq. 3.3).
    Invoked to obtain the folded-normal survival function that is inverted for Λ; exact only for Gaussian Z.
  • standard math Ravikumar et al. (2011) Theorems 1–2: Λij ≍ √(log p / n) implies consistency and sparsistency under irrepresentability and eigenvalue conditions.
    Used verbatim in Section 3.4 to transfer the derived rate into the desired statistical guarantees.
  • ad hoc to paper The estimator remains sparse when the average selection per row z̄(γ) is bounded independently of p (Eq. 3.11–3.13).
    Motivates the dimension-correction that replaces fixed γ by γc; not derived from first principles but imposed to keep s∞ finite.

pith-pipeline@v1.1.0-grok45 · 20643 in / 2788 out tokens · 37363 ms · 2026-07-11T00:55:18.587561+00:00 · methodology

0 comments
read the original abstract

Sparse precision matrix estimation provides an interpretable and computationally efficient framework for modeling conditional dependencies in high-dimensional, low-sample-size data. A recurring challenge is appropriately selecting the regularization parameter that controls estimator sparsity and strikes a balance between underfitting and overfitting. We propose a closed-form, matrix-valued regularization parameter derived from the sampling distribution of the first-order optimality conditions of the $\ell_1$-regularized Gaussian maximum-likelihood estimator. By prescribing the probability that each nonzero entry of the estimator satisfies its optimality condition under resampling, we eliminate the need for cross-validation. The resulting regularization parameter is shown to attain asymptotic scaling properties that, under standard conditions, provide consistency and sparsistency of the estimator. On synthetic Gaussian and non-Gaussian datasets, as well as real-world gene microarray and neuroimaging applications, the proposed approach achieves estimation accuracy comparable to cross-validation, delivers superior support recovery, and reduces runtime by several orders of magnitude.

Figures

Figures reproduced from arXiv: 2607.07735 by Aryan Eftekhari, Daniel Sergio Vega, Ernst-Jan Camiel Wit, Olaf Schenk.

Figure 1
Figure 1. Figure 1: (Left Panel): The probability density functions of the normal and folded-normal random variables X and |X|, respectively, with the highlighted area indicating their difference. (Right Panel): The selection probability as a function of the regularization parameter Λij for varying gradient magnitudes |Gij |; see (3.4). The boundary case for an entry with Θˆ ij ̸= 0 and |Gij | = Λij (solid black line) corresp… view at source ↗
Figure 2
Figure 2. Figure 2: (Left Panel): Empirical reselection probability versus γ, defined as the probability that nonzero entries of Θˆ satisfy the optimality conditions in (3.1) under resampling; see (3.6) and Section 3.1. The theoretical reselection probability of 1 2 + γ is indicated by horizontal reference lines. (Right Panel): Off-diagonal nonzeros per row in Θˆ using Λ(γ) from (3.7). Results are shown as a function of γ ∈ (… view at source ↗
Figure 3
Figure 3. Figure 3: (Left Panel): Number of off-diagonal (OD) nonzeros per row, s, in Θˆ for varying p. We compare fixed γ with γc from (3.14) for different c. (Right Panel): Number of off-diagonal (OD) nonzeros per row s versus c, using Λ from (3.14), for varying sample sizes n (light gray lines) with p = 128. All test use k = n/2 and n = p/2. Substituting γc into (3.7) yields the final regularization parameter Λ (γc) ij ; f… view at source ↗
Figure 4
Figure 4. Figure 4: Results for p = 128 and n = p/2, where samples are drawn from a zero-mean Student’s t-distribution with varying degrees of freedom ν. The methods aquic uses different values of k with a fixed value of c = 30. Note cv-glasso and cv-quic produce similar results. All results are averaged over 10 trials, with shaded bands indicating one standard deviation. 4.1.3. Fixed Sample Ratio Tests. In [PITH_FULL_IMAGE:… view at source ↗
Figure 5
Figure 5. Figure 5: Results for fixed dimension p = 128 across varying sample-to-dimension ratios p/n, with samples drawn from a zero-mean Gaussian distribution. The methods aquic uses different values of c with a fixed value of k = n/2. Note cv-glasso and cv-quic produce similar results, except for the runtime. All results are averaged over 10 trials, with shaded bands indicating one standard deviation [PITH_FULL_IMAGE:figu… view at source ↗
Figure 6
Figure 6. Figure 6: Results for fixed sample ratio test of n = p/2 across varying dimension p, with samples drawn from a zero-mean Gaussian distribution. The methods aquic uses different values of c with a fixed value of k = n/2. Note cv-glasso and cv-quic produce similar results, except for the runtime. We report metrics only for methods with a runtime under 103 seconds. All results are averaged over 10 trials, with shaded b… view at source ↗
Figure 7
Figure 7. Figure 7: (Left) The adjacency matrix A of the weighted conditional-dependency network, reordered to highlight the split between the two detected communities. (Right) A spatial visualization of the left and right hemispheres, showing the bipartition into cortical (red) and subcortical (black) parcels. Node size decreases with proximity to the longitudinal plane. Both visualizations are based on the estimate produced… 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

51 extracted references · 51 canonical work pages

  1. [1]

    A. P. Dempster,Covariance selection, Biometrics, 28 (1972), pp. 157–175, https://doi.org/10. 2307/2528966

    work page 1972

  2. [2]

    Meinshausen and P

    N. Meinshausen and P. B ¨uhlmann,High-dimensional graphs and variable selection with the Lasso, The Annals of Statistics, 34 (2006), pp. 1436 – 1462, https://doi.org/10.1214/ 009053606000000281

    work page 2006

  3. [3]

    Yuan and Y

    M. Yuan and Y. Lin,Model selection and estimation in the gaussian graphical model, Biometrika, 94 (2007), pp. 19–35, https://doi.org/10.1093/biomet/asm018

    work page doi:10.1093/biomet/asm018 2007

  4. [4]

    Friedman, T

    J. Friedman, T. Hastie, and R. Tibshirani,Sparse inverse covariance estimation with the graphical lasso, Biostatistics, 9 (2007), pp. 432–441, https://doi.org/10.1093/biostatistics/ kxm045

    work page doi:10.1093/biostatistics/ 2007

  5. [5]

    Banerjee, L

    O. Banerjee, L. E. Ghaoui, and A. d’Aspremont,Model selection through sparse maxi- mum likelihood estimation for multivariate gaussian or binary data, Journal of Machine Learning Research, 9 (2008), pp. 2261–2286, https://jmlr.org/papers/volume9/banerjee08a/ banerjee08a.pdf

    work page 2008

  6. [6]

    Yuan,High dimensional inverse covariance matrix estimation via linear programming, The Journal of Machine Learning Research, 11 (2010), pp

    M. Yuan,High dimensional inverse covariance matrix estimation via linear programming, The Journal of Machine Learning Research, 11 (2010), pp. 2261–2286, http://jmlr.org/papers/ v11/yuan10b.html

    work page 2010

  7. [7]

    Fan and R

    J. Fan and R. Li,Variable selection via nonconcave penalized likelihood and its oracle properties, Journal of the American statistical Association, 96 (2001), pp. 1348–1360, https://doi.org/ 10.1198/016214501753382273

    work page doi:10.1198/016214501753382273 2001

  8. [8]

    T. Cai, W. Liu, and X. Luo,A constrained ℓ1 minimization approach to sparse precision matrix estimation, Journal of the American Statistical Association, 106 (2011), pp. 594–607, https://doi.org/10.1198/jasa.2011.tm10155. 16EFTEKHARI ET AL

    work page doi:10.1198/jasa.2011.tm10155 2011

  9. [9]

    J. Fan, Y. Feng, and Y. Wu,Network exploration via the adaptive lasso and scad penalties, Annals of Applied Statistics, 3 (2009), pp. 521–541, https://doi.org/10.1214/08-AOAS215

    work page doi:10.1214/08-aoas215 2009

  10. [10]

    Shalom, E

    G. Shalom, E. Treister, and I. Yavneh,pista: Preconditioned iterative soft thresholding algorithm for graphical lasso, SIAM Journal on Scientific Computing, 46 (2024), pp. S445– S466, https://doi.org/10.1137/22M1496128

    work page doi:10.1137/22m1496128 2024

  11. [11]

    Hsieh, I

    C.-j. Hsieh, I. Dhillon, P. Ravikumar, and M. Sustik,Sparse inverse covariance matrix estimation using quadratic approximation, in Advances in Neural Information Processing Systems, J. Shawe-Taylor, R. Zemel, P. Bartlett, F. Pereira, and K. Q. Weinberger, eds., vol. 24, Curran Associates, Inc., 2011, https://proceedings.neurips.cc/paper/2011/file/ 2ba8698...

    work page 2011

  12. [12]

    Hsieh, M

    C.-J. Hsieh, M. A. Sustik, I. S. Dhillon, P. K. Ravikumar, and R. Poldrack,Big & quic: Sparse inverse covariance estimation for a million variables, in Advances in Neural Information Processing Systems, C. J. C. Burges, L. Bottou, M. Welling, Z. Ghahramani, and K. Q. Weinberger, eds., vol. 26, Curran Associates, Inc., 2013, https://proceedings. neurip...

    work page 2013

  13. [13]

    Bollh¨ofer, A

    M. Bollh¨ofer, A. Eftekhari, S. Scheidegger, and O. Schenk,Large-scale Sparse Inverse Covariance Matrix Estimation, SIAM Journal on Scientific Computing, 41 (2019), pp. A380– A401, https://doi.org/10.1137/17M1147615

    work page doi:10.1137/17m1147615 2019

  14. [14]

    Eftekhari, M

    A. Eftekhari, M. Bollh¨ofer, and O. Schenk,Distributed Memory Sparse Inverse Covariance Matrix Estimation on High-Performance Computing Architectures, in ACM/IEEE Inter- national Conference on High Performance Computing, Networking, Storage and Analysis (SC18), 2018, http://doi.org/10.1109/SC.2018.00023

    work page doi:10.1109/sc.2018.00023 2018

  15. [15]

    Eftekhari, D

    A. Eftekhari, D. Pasadakis, M. Bollh ¨ofer, S. Scheidegger, and O. Schenk,Block- enhanced precision matrix estimation for large-scale datasets, Journal of computational science, 53, article 101389 (2021), https://doi.org/10.1016/j.jocs.2021.101389

    work page doi:10.1016/j.jocs.2021.101389 2021

  16. [16]

    A. Eftekhari,Scalable algorithms for high-dimensional graphical lasso and function approxi- mation, PhD thesis, Universit` a della Svizzera italiana, Lugano, 2021, https://susi.usi.ch/ usi/documents/319150

    work page 2021

  17. [17]

    Vujaˇci´c, A

    I. Vujaˇci´c, A. Abbruzzo, and E. Wit,A computationally fast alternative to cross-validation in penalized gaussian graphical models, Journal of Statistical Computation and Simulation, 85 (2015), pp. 3628–3640, https://doi.org/10.1080/00949655.2014.992020

    work page doi:10.1080/00949655.2014.992020 2015

  18. [18]

    Wilson, M

    A. Wilson, M. Kasy, and L. Mackey,Approximate cross-validation: Guarantees for model assessment and selection, in International conference on artificial intelligence and statistics, PMLR, 2020, pp. 4530–4540, https://proceedings.mlr.press/v108/wilson20a/wilson20a.pdf

    work page 2020

  19. [19]

    Tran and G

    C. Tran and G. Yu,A completely tuning-free and robust approach to sparse precision matrix estimation, in International Conference on Machine Learning, vol. 162, PMLR, 2022, pp. 21733–21750, https://proceedings.mlr.press/v162/tran22b.html

    work page 2022

  20. [20]

    L. Wang, B. Peng, J. Bradic, R. Li, and Y. Wu,A tuning-free robust and efficient approach to high-dimensional regression, Journal of the American Statistical Association, 115 (2020), pp. 1700–1714, https://doi.org/10.1080/01621459.2020.1840989

    work page doi:10.1080/01621459.2020.1840989 2020

  21. [21]

    H. Liu, K. Roeder, and L. Wasserman,Stability approach to regularization selec- tion (stars) for high dimensional graphical models, Advances in neural informa- tion processing systems, 23 (2010), https://papers.nips.cc/paper files/paper/2010/hash/ 301ad0e3bd5cb1627a2044908a42fdc2-Abstract.html

    work page 2010

  22. [22]

    Akaike,Information theory and an extension of the maximum likelihood principle, in Selected papers of hirotugu akaike, Springer, 1998, pp

    H. Akaike,Information theory and an extension of the maximum likelihood principle, in Selected papers of hirotugu akaike, Springer, 1998, pp. 199–213, https://doi.org/10.1007/ 978-1-4612-1694-0 15

    work page 1998

  23. [23]

    Schwarz,Estimating the dimension of a model, The annals of statistics, 6 (1978), pp

    G. Schwarz,Estimating the dimension of a model, The annals of statistics, 6 (1978), pp. 461–464, https://doi.org/10.1214/aos/1176344136

    work page doi:10.1214/aos/1176344136 1978

  24. [24]

    Chen and Z

    J. Chen and Z. Chen,Extended bayesian information criteria for model selection with large model spaces, Biometrika, 95 (2008), pp. 759–771, https://doi.org/10.1093/biomet/asn034

    work page doi:10.1093/biomet/asn034 2008

  25. [25]

    Liu and L

    H. Liu and L. Wang,TIGER: A tuning-insensitive approach for optimally estimating Gaussian graphical models, Electronic Journal of Statistics, 11 (2017), pp. 241 – 294, https://doi.org/ 10.1214/16-EJS1195

    work page doi:10.1214/16-ejs1195 2017

  26. [26]

    Zou,The adaptive lasso and its oracle properties, Journal of the American statistical association, 101 (2006), pp

    H. Zou,The adaptive lasso and its oracle properties, Journal of the American statistical association, 101 (2006), pp. 1418–1429, https://doi.org/10.1198/016214506000000735

    work page doi:10.1198/016214506000000735 2006

  27. [27]

    Ravikumar, M

    P. Ravikumar, M. J. Wainwright, G. Raskutti, and B. Yu,High-dimensional covariance estimation by minimizing L1-penalized log-determinant divergence, Electronic Journal of Statistics, 5 (2011), pp. 935 – 980, https://doi.org/10.1214/11-EJS631

    work page doi:10.1214/11-ejs631 2011

  28. [28]

    A. J. Rothman, P. J. Bickel, E. Levina, and J. Zhu,Sparse permutation invariant covariance estimation, Electronic Journal of Statistics, 2 (2008), pp. 494 – 515, https://doi.org/10. 1214/08-EJS176. SPARSE PRECISION MATRIX ESTIMATION17

    work page 2008

  29. [29]

    Boyd and L

    S. Boyd and L. Vandenberghe,Convex optimization, Cambridge university press, 2004. ISBN 0-521-83378-7

    work page 2004

  30. [30]

    Oztoprak, J

    F. Oztoprak, J. Nocedal, S. Rennie, and P. A. Olsen,Newton-like methods for sparse inverse covariance estimation, Advances in neural information pro- cessing systems, 25 (2012), https://proceedings.neurips.cc/paper files/paper/2012/file/ b3967a0e938dc2a6340e258630febd5a-Paper.pdf

    work page 2012

  31. [31]

    Gnedenko and A

    B. Gnedenko and A. Kolmogorov,Limit Distributions for Sums of Independent Random Variables, Creative Media Partners, LLC, 2021

    work page 2021

  32. [32]

    Isserlis,On a formula for the product-moment coefficient of any order of a normal frequency distribution in any number of variables, Biometrika, 12 (1918), pp

    L. Isserlis,On a formula for the product-moment coefficient of any order of a normal frequency distribution in any number of variables, Biometrika, 12 (1918), pp. 134–139, https://doi.org/10.2307/2331932

    work page doi:10.2307/2331932 1918

  33. [33]

    Wick,The evaluation of the collision matrix, Physical review, 80 (1950), p

    G.-C. Wick,The evaluation of the collision matrix, Physical review, 80 (1950), p. 268, https: //doi.org/10.1103/PhysRev.80.268

    work page doi:10.1103/physrev.80.268 1950

  34. [34]

    F. C. Leone, L. S. Nelson, and R. Nottingham,The folded normal distribution, Technometrics, 3 (1961), pp. 543–550, https://doi.org/https://doi.org/10.2307/1266560

    work page doi:10.2307/1266560 1961

  35. [35]

    Blair, C

    J. Blair, C. Edwards, and J. H. Johnson,Rational chebyshev approximations for the inverse of the error function, Mathematics of Computation, 30 (1976), pp. 827–830, https: //doi.org/10.1090/S0025-5718-1976-0421040-7

    work page doi:10.1090/s0025-5718-1976-0421040-7 1976

  36. [36]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay,Scikit-learn: Machine learning in Python, Journal of Machine Learning Research, 12 (2011), pp. 2825–2830, http://jmlr.org/papers/v12/pedregosa11a.html

    work page 2011

  37. [37]

    X. Li, T. Zhao, X. Yuan, and H. Liu,The flare package for high dimensional linear regression and precision matrix estimation in r, Journal of Machine Learning Research, 16 (2015), pp. 553–557, http://jmlr.org/papers/v16/li15a.html

    work page 2015

  38. [38]

    T. Zhao, H. Liu, K. Roeder, J. Lafferty, and L. Wasserman,The huge package for high- dimensional undirected graph estimation in r, Journal of Machine Learning Research, 13 (2012), pp. 1059–1062, https://jmlr.org/papers/volume13/zhao12a/zhao12a.pdf

    work page 2012

  39. [39]

    B. E. Stranger, A. C. Nica, M. S. Forrest, A. Dimas, C. P. Bird, C. Beazley, C. E. Ingle, M. Dunning, P. Flicek, D. Koller, S. Montgomery, S. Tavar ´e, P. Deloukas, and E. T. Dermitzakis,Population genomics of human gene expression, Nature Genetics, 39 (2007), pp. 1217–1224, https://doi.org/10.1038/ng2142

    work page doi:10.1038/ng2142 2007

  40. [40]

    Mohammadi and E

    A. Mohammadi and E. C. Wit,Bayesian Structure Learning in Sparse Gaussian Graphical Models, Bayesian Analysis, 10 (2015), pp. 109–138, https://doi.org/10.1214/14-BA889. Publisher: International Society for Bayesian Analysis

    work page doi:10.1214/14-ba889 2015

  41. [41]

    Bhadra and B

    A. Bhadra and B. K. Mallick,Joint High-Dimensional Bayesian Variable and Covariance Selection with an Application to eQTL Analysis, Biometrics, 69 (2013), pp. 447–457, https://doi.org/10.1111/biom.12021

    work page doi:10.1111/biom.12021 2013

  42. [42]

    M. E. J. Newman,Networks: An Introduction, Oxford University Press, 2011, https://doi.org/ 10.1093/acprof:oso/9780199206650.001.0001

    work page doi:10.1093/acprof:oso/9780199206650.001.0001 2011

  43. [43]

    Russo, H

    R. Russo, H. J. Herrmann, and L. de Arcangelis,Brain modularity controls the critical behavior of spontaneous activity, Scientific Reports, 4 (2014), p. 4312, https://doi.org/10. 1038/srep04312

    work page 2014

  44. [44]

    M. F. Glasser, T. S. Coalson, E. C. Robinson, C. D. Hacker, J. Harwell, E. Yacoub, K. Ugurbil, J. Andersson, C. F. Beckmann, M. Jenkinson, S. M. Smith, and D. C. Van Essen,A multi-modal parcellation of human cerebral cortex, Nature, 536 (2016), pp. 171–178, https://doi.org/10.1038/nature18933

    work page doi:10.1038/nature18933 2016

  45. [45]

    R. L. Buckner, F. M. Krienen, A. Castellanos, J. C. Diaz, and B. T. T. Yeo,The organization of the human cerebellum estimated by intrinsic functional connectivity, Journal of Neurophysiology, 106 (2011), pp. 2322–2345, https://doi.org/10.1152/jn.00339.2011

    work page doi:10.1152/jn.00339.2011 2011

  46. [46]

    E. Y. Choi, B. T. T. Yeo, and R. L. Buckner,The organization of the human striatum estimated by intrinsic functional connectivity, Journal of Neurophysiology, 108 (2012), pp. 2242–2263, https://doi.org/10.1152/jn.00270.2012

    work page doi:10.1152/jn.00270.2012 2012

  47. [47]

    J. L. Ji, M. Spronk, K. Kulkarni, G. Repov ˇs, A. Anticevic, and M. W. Cole,Mapping the human brain’s cortical-subcortical functional network organization, NeuroImage, 185 (2019), pp. 35–57, https://doi.org/10.1016/j.neuroimage.2018.10.006

    work page doi:10.1016/j.neuroimage.2018.10.006 2019

  48. [48]

    Liang, J

    Z. Liang, J. King, and N. Zhang,Uncovering intrinsic connectional architecture of functional networks in awake rat brain, Journal of Neuroscience, 31 (2011), pp. 3776–3783, https: //doi.org/10.1523/jneurosci.4557-10.2011

    work page doi:10.1523/jneurosci.4557-10.2011 2011

  49. [49]

    Karypis and V

    G. Karypis and V. Kumar,METIS: A Software Package for Partitioning Unstructured Graphs, Partitioning Meshes, and Computing Fill-Reducing Orderings of Sparse Matrices, 18EFTEKHARI ET AL. University of Minnesota, September 1998, https://hdl.handle.net/11299/215346. Retrieved from the University Digital Conservancy

    work page 1998

  50. [50]

    Clauset, M

    A. Clauset, M. E. J. Newman, and C. Moore,Finding community structure in very large networks, Phys. Rev. E, 70, article 066111 (2004), https://doi.org/10.1103/PhysRevE.70. 066111

    work page doi:10.1103/physreve.70 2004

  51. [51]

    Husson, J

    F. Husson, J. Josse, and G. Saporta,Jan de leeuw and the french school of data analysis, Journal of Statistical Software, 73 (2016), pp. 1–18, https://doi.org/10.18637/jss.v073.i06

    work page doi:10.18637/jss.v073.i06 2016