pith. sign in

arxiv: 2309.09371 · v2 · submitted 2023-09-17 · 📊 stat.ME

Gibbs Sampling using Anti-correlation Gaussian Data Augmentation, with Applications to L1-ball-type Models

Yu Zheng , Leo L. Duan This is my paper

Pith reviewed 2026-05-24 07:09 UTC · model grok-4.3

classification 📊 stat.ME
keywords Gibbs samplingdata augmentationL1-ball priorsspike-and-slab priorsgeometric ergodicitylatent Gaussian modelsstructured sparsityblock sampling
0
0 comments X

The pith

Anti-correlation Gaussian data augmentation enables fast block Gibbs sampling for L1-ball-type models by canceling quadratic terms to make parameters conditionally independent.

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

The paper develops a data augmentation technique using anti-correlation Gaussians for L1-ball-type priors, which generalize spike-and-slab priors to allow exact zeros with positive probability under flexible structured sparsity. This augmentation cancels the quadratic exponent in the latent Gaussian distribution, enabling the parameters of interest to be updated jointly in a single block during Gibbs sampling. The result is an algorithm with very low computing cost per iteration, rapid mixing of the chains, and a geometric ergodicity guarantee in linear models. A sympathetic reader would care because it addresses the computational bottleneck in posterior inference for these flexible sparse models, outperforming general-purpose samplers like NUTS in speed and mixing. The method also extends directly to broader classes of latent Gaussian models including multivariate truncated Gaussians and latent Gaussian processes.

Core claim

By introducing the anti-correlation Gaussian as a latent variable that exactly cancels the quadratic term in the conditional distribution of the parameters, the method renders the parameters of interest conditionally independent given the latent variable, permitting a block update in the Gibbs sampler while preserving the correct marginal posterior.

What carries the argument

The anti-correlation Gaussian latent variable, which cancels the quadratic exponent term in the latent Gaussian distribution to induce conditional independence among parameters for block updates.

If this is right

  • Enables a block Gibbs sampler with very low computing cost per iteration compared to alternatives like the No-U-Turn sampler.
  • Produces rapid mixing of the Markov chains in practice.
  • Guarantees geometric ergodicity of the sampler in linear models.
  • Supports direct extensions to posterior estimation in general latent Gaussian models such as those with multivariate truncated Gaussians or latent Gaussian processes.

Where Pith is reading between the lines

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

  • The same cancellation approach could apply to other sparsity priors that rely on latent variable representations beyond the L1-ball construction.
  • It may improve scalability for high-dimensional problems where structured dependence among zero probabilities is modeled.
  • The technique might reduce reliance on gradient-based samplers in settings where the conditional independence structure can be exploited.

Load-bearing premise

The anti-correlation Gaussian can be constructed and sampled to exactly cancel the quadratic term while preserving the correct marginal posterior for the parameters of interest.

What would settle it

A numerical check on a simple linear regression model where the marginal posterior samples from the new blocked Gibbs sampler are compared to those from an exact method such as rejection sampling; systematic mismatch in the samples would falsify the claim that the marginal is preserved.

Figures

Figures reproduced from arXiv: 2309.09371 by Leo L. Duan, Yu Zheng.

Figure 1
Figure 1. Figure 1: Trace plots for the four algorithms: anti-correlation Gaussian (Anti-corr [PITH_FULL_IMAGE:figures/full_fig_p015_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Posterior density estimations of κ0 for varying dimensions p when (c, ρ) = (3, 0.5). 18 [PITH_FULL_IMAGE:figures/full_fig_p018_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Trace plots and autocorrelation function (ACF) plot. Each box on the [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Posterior estimates, point-wise standard deviations, and direct soft [PITH_FULL_IMAGE:figures/full_fig_p021_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Trace plots of the log-posterior densities during the burn-in stage. [PITH_FULL_IMAGE:figures/full_fig_p034_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Trace plots, data, ground truth, posterior estimates, point-wise vari [PITH_FULL_IMAGE:figures/full_fig_p035_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Traceplots and effective sample size per computing time (ESS/s) for [PITH_FULL_IMAGE:figures/full_fig_p036_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Estimates and ACF plots for anti-correlation Gaussian and SSVS. The [PITH_FULL_IMAGE:figures/full_fig_p038_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Under d = λp(XT X)/σ2 +ε with ε ∈ {10−6 , 10−4 , 10−2 , 1, 10, 10}, the mixings of anti-correlation blocked Gibbs sampler have almost no difference. 39 [PITH_FULL_IMAGE:figures/full_fig_p039_9.png] view at source ↗
read the original abstract

L1-ball-type priors are a recent generalization of the spike-and-slab priors. By transforming a continuous precursor distribution to the L1-ball boundary, it induces exact zeros with positive prior and posterior probabilities. With great flexibility in choosing the precursor and threshold distributions, we can easily specify models under structured sparsity, such as those with dependent probability for zeros and smoothness among the non-zeros. Motivated to significantly accelerate the posterior computation, we propose a new data augmentation that leads to a fast block Gibbs sampling algorithm. The latent variable, named ``anti-correlation Gaussian'', cancels out the quadratic exponent term in the latent Gaussian distribution, making the parameters of interest conditionally independent so that they can be updated in a block. Compared to existing algorithms such as the No-U-Turn sampler, the new blocked Gibbs sampler has a very low computing cost per iteration and shows rapid mixing of Markov chains. We establish the geometric ergodicity guarantee of the algorithm in linear models. Further, we show useful extensions of our algorithm for posterior estimation of general latent Gaussian models, such as those involving multivariate truncated Gaussian or latent Gaussian process.

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 an 'anti-correlation Gaussian' data augmentation for L1-ball-type priors (generalizations of spike-and-slab that induce exact zeros via transformation of a precursor distribution). This augmentation is claimed to cancel the quadratic term in the latent Gaussian, yielding a block Gibbs sampler in which the parameters of interest are conditionally independent, with very low per-iteration cost, rapid mixing, and a geometric ergodicity guarantee for linear models; extensions to multivariate truncated Gaussians and latent Gaussian processes are also presented.

Significance. If the augmentation construction is shown to recover the exact target marginal posterior, the blocked Gibbs sampler would provide a computationally attractive alternative to NUTS for structured-sparsity models, with explicit mixing-rate guarantees that are rare in this literature.

major comments (2)
  1. [Abstract (latent variable definition paragraph)] The central construction (anti-correlation Gaussian mean and covariance that exactly cancel the quadratic exponent while preserving the correct marginal on the parameters of interest) is asserted in the abstract but not derived explicitly; without the explicit joint specification and verification that the marginal recovers the L1-ball posterior for arbitrary precursor/threshold choices, the conditional-independence claim and all downstream ergodicity results rest on an unverified step.
  2. [Theoretical results (ergodicity section)] The geometric ergodicity guarantee for linear models is stated as established, yet the manuscript provides neither the key steps of the proof nor the conditions on the anti-correlation Gaussian under which the drift/minorization conditions hold; this is load-bearing for the algorithmic claim.
minor comments (2)
  1. Notation for the precursor distribution and threshold is introduced without a consolidated table of symbols, making it difficult to track the dependence structure across sections.
  2. Numerical comparisons with NUTS report mixing times but do not include effective sample size per CPU second or autocorrelation plots at multiple chain lengths, which would strengthen the 'rapid mixing' claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful and constructive report. The two major comments identify places where the manuscript would benefit from greater explicitness. We will revise accordingly and address each point below.

read point-by-point responses

  1. Referee: [Abstract (latent variable definition paragraph)] The central construction (anti-correlation Gaussian mean and covariance that exactly cancel the quadratic exponent while preserving the correct marginal on the parameters of interest) is asserted in the abstract but not derived explicitly; without the explicit joint specification and verification that the marginal recovers the L1-ball posterior for arbitrary precursor/threshold choices, the conditional-independence claim and all downstream ergodicity results rest on an unverified step.

    Authors: We agree that the abstract states the cancellation property without supplying the joint density or the marginalization argument. Section 2 of the manuscript defines the anti-correlation Gaussian and states that its mean and covariance are chosen to cancel the quadratic term arising from the L1-ball prior, but the explicit joint specification and the verification that the marginal on the parameters of interest recovers the target posterior for general precursor and threshold distributions are only sketched. In the revision we will insert a short dedicated paragraph (or subsection) that writes the joint density explicitly, shows the cancellation, and verifies the marginal for arbitrary choices of the precursor and threshold. This will make the conditional-independence claim and the subsequent ergodicity results rest on a fully documented step. revision: yes

  2. Referee: [Theoretical results (ergodicity section)] The geometric ergodicity guarantee for linear models is stated as established, yet the manuscript provides neither the key steps of the proof nor the conditions on the anti-correlation Gaussian under which the drift/minorization conditions hold; this is load-bearing for the algorithmic claim.

    Authors: Theorem 3.1 asserts geometric ergodicity of the blocked Gibbs sampler for linear models under the anti-correlation augmentation. The proof strategy (drift and minorization) is indicated, but the explicit drift function, the minorization constant, and the precise restrictions these impose on the parameters of the anti-correlation Gaussian are not written out. We will expand the proof section in the revision to include (i) the explicit form of the drift function, (ii) the conditions on the anti-correlation mean and covariance that guarantee a uniform minorization probability on a small set, and (iii) the main algebraic steps that verify the drift inequality. These additions will make the ergodicity claim fully self-contained. revision: yes

Circularity Check

0 steps flagged

No circularity: novel augmentation and ergodicity result are independently constructed

full rationale

The paper introduces the anti-correlation Gaussian as an explicit new latent construction whose joint specification is designed to cancel the quadratic term while recovering the target marginal. Geometric ergodicity is stated as a separately established guarantee for linear models. No equations, self-citations, or fitted parameters are shown to define the core sampler by construction; the derivation chain remains self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The paper rests on the domain assumption that flexible precursor and threshold distributions can encode structured sparsity, plus the invented latent variable whose sampling properties are asserted without external verification.

axioms (1)
  • domain assumption Flexible choice of precursor and threshold distributions induces structured sparsity with dependent zero probabilities and smoothness among non-zeros.
    Stated directly in the abstract as the modeling motivation.
invented entities (1)
  • anti-correlation Gaussian no independent evidence
    purpose: Cancels the quadratic exponent to render parameters conditionally independent for block sampling.
    New latent variable introduced to achieve the block update; no independent evidence supplied in abstract.

pith-pipeline@v0.9.0 · 5724 in / 1113 out tokens · 24662 ms · 2026-05-24T07:09:11.736533+00:00 · methodology

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

56 extracted references · 56 canonical work pages

  1. [1]

    Albert, J. H. and S. Chib (1993). Bayesian Analysis of Binary and Polychotomous Response Data . Journal of the American Statistical Association\/ 88\/ (422), 669--679

    work page 1993

  2. [2]

    Armagan, A., D. B. Dunson, and J. Lee (2013). Generalized Double Pareto Shrinkage . Statistica Sinica\/ 23\/ (1), 119

    work page 2013

  3. [3]

    Bai, R. and M. Ghosh (2019). On the Beta Prime Prior for Scale Parameters in High-Dimensional Bayesian Regression Models . Statistica Sinica\/

    work page 2019

  4. [4]

    Rodr \' guez, and N

    Betancourt, B., A. Rodr \' guez, and N. Boyd (2017). Bayesian Fused Lasso Regression for Dynamic Binary Networks . Journal of Computational and Graphical Statistics\/ 26\/ (4), 840--850

    work page 2017

  5. [5]

    Chakraborty, and B

    Bhattacharya, A., A. Chakraborty, and B. K. Mallick (2016). Fast Sampling With Gaussian Scale Mixture Priors in High-Dimensional Regression . Biometrika\/ 103\/ (4), 985--991

    work page 2016

  6. [6]

    Bhattacharya, A., D. Pati, N. S. Pillai, and D. B. Dunson (2015). Dirichlet--Laplace Priors for Optimal Shrinkage . Journal of the American Statistical Association\/ 110\/ (512), 1479--1490

    work page 2015

  7. [7]

    Bingham, E., J. P. Chen, M. Jankowiak, F. Obermeyer, N. Pradhan, T. Karaletsos, R. Singh, P. A. Szerlip, P. Horsfall, and N. D. Goodman (2019). Pyro: Deep Universal Probabilistic Programming . Journal of Machine Learning Research\/ 20 , 28:1--28:6

    work page 2019

  8. [8]

    Bottolo, L. and S. Richardson (2010). Evolutionary Stochastic Search for Bayesian Model Exploration . Bayesian Analysis\/ 5\/ (3), 583--618

    work page 2010

  9. [9]

    Gelman, M

    Carpenter, B., A. Gelman, M. D. Hoffman, D. Lee, B. Goodrich, M. Betancourt, M. A. Brubaker, J. Guo, P. Li, and A. Riddell (2017). Stan: A Probabilistic Programming Language . Journal of Statistical Software\/ 76

    work page 2017

  10. [10]

    Carvalho, C. M., N. G. Polson, and J. G. Scott (2010). The Horseshoe Estimator for Sparse Signals . Biometrika\/ 97\/ (2), 465--480

    work page 2010

  11. [11]

    Chandra, N. K., P. Mueller, and A. Sarkar (2021). Bayesian Scalable Precision Factor Analysis for Massive Sparse Gaussian Graphical Models . arXiv Preprint arXiv:2107.11316\/

    work page arXiv 2021

  12. [12]

    Chow, E. and Y. Saad (2014). Preconditioned Krylov Subspace Methods for Sampling Multivariate Gaussian Distributions . SIAM Journal on Scientific Computing\/ 36\/ (2), A588--A608

    work page 2014

  13. [13]

    Khare, and L

    Diaconis, P., K. Khare, and L. Saloff-Coste (2008). Gibbs Sampling, Exponential Families and Orthogonal Polynomials . Quality Engineering\/ 54 , 31--32

    work page 2008

  14. [14]

    Duane, S., A. D. Kennedy, B. J. Pendleton, and D. Roweth (1987). Hybrid Monte Carlo . Physics Letters B\/ 195\/ (2), 216--222

    work page 1987

  15. [15]

    Fahrmeir, L. and G. Tutz (2001). Multivariate Statistical Modelling Based on Generalized Linear Models . Springer Series in Statistics . Springer New York

    work page 2001

  16. [16]

    Gelman, A. (2014). Bayesian Data Analysis \/ (Third edition ed.). Chapman & Hall / CRC texts in statistical science. CRC Press

    work page 2014

  17. [17]

    George, E. I. and R. E. McCulloch (1995). Stochastic Search Variable Selection . Markov Chain Monte Carlo in Practice \/ 68 , 203--214

    work page 1995

  18. [18]

    Ghosh, J. and M. A. Clyde (2011). Rao--Blackwellization for Bayesian Variable Selection and Model Averaging in Linear and Binary Regression: A Novel Data Augmentation Approach . Journal of the American Statistical Association\/ 106\/ (495), 1041--1052

    work page 2011

  19. [19]

    Ghosh, J. and D. B. Dunson (2009). Default Prior Distributions and Efficient Posterior Computation in Bayesian Factor Analysis . Journal of Computational and Graphical Statistics\/ 18\/ (2), 306--320

    work page 2009

  20. [20]

    Girolami, M. and B. Calderhead (2011). Riemann Manifold Langevin and Hamiltonian Monte Carlo Methods . Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 73\/ (2), 123--214

    work page 2011

  21. [21]

    Griffin, M. and P. D. Hoff (2023). Structured Shrinkage Priors . Journal of Computational and Graphical Statistics\/ (in press), 1--22

    work page 2023

  22. [22]

    Dobra, and M

    Hans, C., A. Dobra, and M. West (2007). Shotgun Stochastic Search for “Large P” Regression . Journal of the American Statistical Association\/ 102\/ (478), 507--516

    work page 2007

  23. [23]

    Hoff, P. D. (2009). Simulation of the Matrix Bingham--Von Mises--Fisher Distribution, With Applications to Multivariate and Relational Data . Journal of Computational and Graphical Statistics\/ 18\/ (2), 438--456

    work page 2009

  24. [24]

    Hoff, P. D. (2017). Lasso, Fractional Norm and Structured Sparse Estimation Using a Hadamard Product Parametrization . Computational Statistics & Data Analysis\/ 115 , 186--198

    work page 2017

  25. [25]

    Hoffman, M. D. and A. Gelman (2014). The No-U-Turn Sampler: Adaptively Setting Path Lengths in Hamiltonian Monte Carlo . Journal of Machine Learning Research\/ 15\/ (1), 1593--1623

    work page 2014

  26. [26]

    Kang, J., B. J. Reich, and A.-M. Staicu (2018). Scalar-on-Image Regression via the Soft-Thresholded Gaussian Process . Biometrika\/ 105\/ (1), 165--184

    work page 2018

  27. [27]

    Taskar, et al

    Kulesza, A., B. Taskar, et al. (2012). Determinantal Point Processes for MacHine Learning . Foundations and Trends in Machine Learning\/ 5\/ (2--3), 123--286

    work page 2012

  28. [28]

    Kwon, D., M. T. Landi, M. Vannucci, H. J. Issaq, D. Prieto, and R. M. Pfeiffer (2011). An Efficient Stochastic Search for Bayesian Variable Selection With High-Dimensional Correlated Predictors . Computational Statistics & Data Analysis\/ 55\/ (10), 2807--2818

    work page 2011

  29. [29]

    Liu, J. S., W. H. Wong, and A. Kong (1994). Covariance Structure of the Gibbs Sampler With Applications to the Comparisons of Estimators and Augmentation Schemes . Biometrika\/ 81\/ (1), 27--40

    work page 1994

  30. [30]

    Livingstone, S. and G. Zanella (2022). The Barker Proposal: Combining Robustness and Efficiency in Gradient-Based McMc . Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 84\/ (2), 496--523

    work page 2022

  31. [31]

    Meyn, S. P. and R. L. Tweedie (1994). Computable Bounds for Geometric Convergence Rates of Markov Chains . The Annals of Applied Probability\/ 4\/ (4), 981--1011

    work page 1994

  32. [32]

    Mitchell, T. J. and J. J. Beauchamp (1988). Bayesian Variable Selection in Linear Regression . Journal of the American Statistical Association\/ 83\/ (404), 1023--1032

    work page 1988

  33. [33]

    Neal, R. M. (2003). Slice Sampling . The Annals of Statistics\/ 31\/ (3), 705--767

    work page 2003

  34. [34]

    Neal, R. M. (2011). MCMC Using Hamiltonian Dynamics . In S. Brooks, A. Gelman, G. Jones, and X.-L. Meng (Eds.), Handbook of Markov Chain Monte Carlo , Chapter 5. CRC Press

    work page 2011

  35. [35]

    Large N, Large P

    Nishimura, A. and M. A. Suchard (2022). Prior-Preconditioned Conjugate Gradient Method for Accelerated Gibbs Sampling in “Large N, Large P” Bayesian Sparse Regression . Journal of the American Statistical Association\/ , 1--14

    work page 2022

  36. [36]

    Piironen, J. and A. Vehtari (2017). Sparsity Information and Regularization in the Horseshoe and Other Shrinkage Priors . Electronic Journal of Statistics\/ 11\/ (2), 5018--5051

    work page 2017

  37. [37]

    Polson, N. G. and J. G. Scott (2010). Shrink Globally, Act Locally: Sparse Bayesian Regularization and Prediction . Bayesian Statistics\/ 9\/ (501-538), 105

    work page 2010

  38. [38]

    Polson, N. G., J. G. Scott, and J. Windle (2013). Bayesian Inference for Logistic Models Using P \'o lya--Gamma Latent Variables . Journal of the American Statistical Association\/ 108\/ (504), 1339--1349

    work page 2013

  39. [39]

    Qin, Q., J. P. Hobert, and K. Khare (2019). Estimating the Spectral Gap of A Trace-Class Markov Operator . Electronic Journal of Statistics\/ 13\/ (1), 1790 -- 1822

    work page 2019

  40. [40]

    Robert, C. P. (1995). Convergence Control Methods for Markov Chain Monte Carlo Algorithms . Statistical Science\/ 10\/ (3), 231--253

    work page 1995

  41. [41]

    Robert, C. P. and G. Casella (2004). Monte Carlo Statistical Methods \/ (2nd ed ed.). Springer texts in statistics. New York: Springer

    work page 2004

  42. [42]

    Roberts, G. O. and J. S. Rosenthal (2001). Markov Chains and De-Initializing Processes . Scandinavian Journal of Statistics\/ 28\/ (3), 489--504

    work page 2001

  43. [43]

    Ro c kov \'a , V. and E. I. George (2018). The Spike-and-Slab Lasso . Journal of the American Statistical Association\/ 113\/ (521), 431--444

    work page 2018

  44. [44]

    Rosenthal, J. S. (1994). Minorization Conditions and Convergence Rates for Markov Chain Monte Carlo . Journal of the American Statistical Association\/ 90 , 558--566

    work page 1994

  45. [45]

    Rossky, P. J., J. D. Doll, and H. L. Friedman (1978). Brownian Dynamics as Smart Monte Carlo Simulation . The Journal of Chemical Physics\/ 69\/ (10), 4628--4633

    work page 1978

  46. [46]

    Martino, and N

    Rue, H., S. Martino, and N. Chopin (2009). Approximate Bayesian Inference for Latent Gaussian Models by Using Integrated Nested Laplace Approximations . Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 71\/ (2), 319--392

    work page 2009

  47. [47]

    Scott, S. L. and H. R. Varian (2014). Predicting the Present With Bayesian Structural Time Series . International Journal of Mathematical Modelling and Numerical Optimisation\/ 5\/ (1-2), 4--23

    work page 2014

  48. [48]

    Tadesse, M. G. and M. Vannucci (2021). Handbook of Bayesian Variable Selection . CRC Press

    work page 2021

  49. [49]

    Tan, A., G. L. Jones, and J. P. Hobert (2013, January). On the Geometric Ergodicity of Two - Variable Gibbs Samplers . In Advances in Modern Statistical Theory and Applications : A Festschrift in honor of Morris L . Eaton , Volume 10, pp.\ 25--43. Institute of Mathematical Statistics

    work page 2013

  50. [50]

    Thomas, S. and W. Tu (2021). Learning Hamiltonian Monte Carlo in R . The American Statistician\/ 75\/ (4), 403--413

    work page 2021

  51. [51]

    Thompson, M. (2011). Slice Sampling With Multivariate Steps . University of Toronto Toronto, Canada

    work page 2011

  52. [52]

    Saunders, S

    Tibshirani, R., M. Saunders, S. Rosset, J. Zhu, and K. Knight (2005). Sparsity and Smoothness via the Fused Lasso . Journal of the Royal Statistical Society: Series B (Statistical Methodology)\/ 67\/ (1), 91--108

    work page 2005

  53. [53]

    Tikhonov, G., . H. Opedal, N. Abrego, A. Lehikoinen, M. M. de Jonge, J. Oksanen, and O. Ovaskainen (2020). Joint Species Distribution Modelling With the R-Package Hmsc . Methods in Ecology and Evolution\/ 11\/ (3), 442--447

    work page 2020

  54. [54]

    Wilhelm, S. and B. Manjunath, G. (2010). tmvtnorm: A Package for the Truncated Multivariate Normal Distribution . The R Journal\/ 2\/ (1), 25

    work page 2010

  55. [55]

    Xu, M. and L. L. Duan (2023). Bayesian Inference With the L1-Ball Prior: Solving Combinatorial Problems With Exact Zeros . Journal of the Royal Statistical Society: Series B (Statistical Methodology)\/ (in press)

    work page 2023

  56. [56]

    Yang, Y., M. J. Wainwright, and M. I. Jordan (2016). On the Computational Complexity of High-Dimensional Bayesian Variable Selection . The Annals of Statistics\/ 44\/ (6), 2497--2532

    work page 2016