pith. sign in

arxiv: 2509.13267 · v2 · submitted 2025-09-16 · 📊 stat.ME · stat.ML

Learning discrete Bayesian networks with hierarchical Dirichlet shrinkage

Alexander Dombowsky , David B. Dunson This is my paper

Pith reviewed 2026-05-18 15:52 UTC · model grok-4.3

classification 📊 stat.ME stat.ML
keywords discrete Bayesian networkshierarchical priorsDirichlet shrinkagestructure learningMetropolis-adjusted LangevinGibbs samplingsparse categorical data
0
0 comments X

The pith

A hierarchical prior on conditional probabilities shrinks discrete Bayesian networks to low-dimensional latent parameters.

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

The paper proposes a hierarchical model that places a prior on the conditional probability tables of each node given its parents in a discrete Bayesian network. This induces posterior shrinkage toward simpler, lower-dimensional latent representations rather than estimating full high-dimensional tables directly. Sampling from the resulting posterior is achieved by embedding a Metropolis-adjusted Langevin algorithm inside a Gibbs sampler, after verifying that the relevant full conditional is log-concave under mild conditions. Structure-learning procedures are then constructed that respect the directed acyclic graph constraint while using the hierarchical prior. The approach is tested on sparse count data, graph recovery tasks, and a breast cancer prognostic network.

Core claim

The central claim is that a hierarchical Dirichlet model for node-parent conditional probabilities in discrete Bayesian networks induces a posteriori shrinkage to low-dimensional latent parameters. Posterior samples of these latent variables are generated via the Metropolis-adjusted Langevin algorithm within a Gibbs sampler. The full conditional distribution is shown to be log-concave under mild conditions, which supports efficient sampling. Structure-learning algorithms are developed that incorporate the hierarchical prior while preserving the DAG property.

What carries the argument

The hierarchical Dirichlet shrinkage prior placed directly on the conditional probability tables of each node given its parents, which concentrates posterior mass on a lower-dimensional latent representation.

If this is right

  • Improved parameter estimation when cell counts are sparse.
  • More reliable recovery of network structure in simulated settings.
  • Principled selection among competing DAGs.
  • Practical application to prognostic networks in categorical medical data.

Where Pith is reading between the lines

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

  • The shrinkage mechanism could be combined with other structure priors to further regularize very high-dimensional networks.
  • The same hierarchical construction might apply to learning in other discrete graphical models that suffer from parameter proliferation.
  • Testing the method on longitudinal or time-stamped categorical data would check whether the latent-parameter reduction remains effective outside static networks.

Load-bearing premise

The full conditional distribution is log-concave under mild conditions, allowing the Metropolis-adjusted Langevin step to sample efficiently inside the Gibbs sampler.

What would settle it

Simulations in which the posterior mass does not concentrate on the low-dimensional latent parameters, or in which the structure-learning algorithms recover the true DAG no better than standard non-hierarchical methods, would falsify the claimed benefit.

Figures

Figures reproduced from arXiv: 2509.13267 by Alexander Dombowsky, David B. Dunson.

Figure 1
Figure 1. Figure 1: An example DAG for variables x1, x2, x3, x4, and x5. categories. This fact motivates the shrinkage of conditional probabilities toward node-specific latent prior means, bypassing the need to specify high-dimensional hyperparameters. Therefore, there are two layers to the model: (i) high-dimensional conditional probabilities and (ii) low-dimensional latent prior means. After marginalizing the first layer, w… view at source ↗
Figure 2
Figure 2. Figure 2: Fitted values Pr xpxj “ 1 | xj´1, nq for MLEs πpj|j´1 p1q “ p3{4, 1{3, 0{5, 3{7, 0{3, 3{3q and αj varying in t0.01, 1, 2, 5, 10, 100u, with comparison to the true values of Prpxj “ 1 | xj´1q in the leftmost column. For instance, [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The HiDDeN MAP estimate for the Markov blanket of lung cancer in the LUCAS data, [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Traceplot for the log-posterior log fptj , zj | nq for the lung cancer variable in the LUCAS dataset. lung cancer (all of which are presence/absence). Algorithm 2 is run to select a parent set from 2 p´1 possibilities for each of the six variables, with the number of iterations equal to 10, 000, 200 of which are discarded as burn-in, and the stepsizes chosen according to the acceptance probabilities of the… view at source ↗
Figure 5
Figure 5. Figure 5: Median probability model for the METABRIC variables after fitting HiDDeN (top), group [PITH_FULL_IMAGE:figures/full_fig_p022_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Posterior edge probabilities for the METABRIC dataset. The [PITH_FULL_IMAGE:figures/full_fig_p024_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Traceplots for the log-posterior for TBS (a), CHT (b), and DFC (c). Burn-in iterations have [PITH_FULL_IMAGE:figures/full_fig_p036_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Two possible DAGs, G1 and G2, respectively, for p “ 3 variables. both cases, we simulate data according to Prpx1q “ 1{k1, Prpx2q “ 1{k2, and Prpx3 “ 1 | x2q „ Unifp0, 1q; if G1 is the true DAG; Prpx3 “ 1 | x1, x2q „ Unifp0, 1q; if G2 is the true DAG. For each true DAG and replication, we compute Pr xpGtrue | nq via HiDDeN, as well as the BIC, AIC, and the BDE score for G1 and G2. The HiDDeN MCMC sampler wi… view at source ↗
Figure 9
Figure 9. Figure 9: Network structure for a subset of variables in the ALARM dataset ( [PITH_FULL_IMAGE:figures/full_fig_p039_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: The values of b Pr xpx new j | xPapjq , n, Gq estimated from n “ 200 observations in the ALARM network. tFalse, Trueu is coded as t0, 1u and tLow, Normal, Highu is coded as t1, 2, 3u. 40 [PITH_FULL_IMAGE:figures/full_fig_p040_10.png] view at source ↗
read the original abstract

A discrete Bayesian network is a directed acyclic graph (DAG) consisting of categorical variables. Two popular approaches for DBN modeling include classification and nonparametric methods. However, both methods often require a large number of parameters, such as high-order interactions in the former and cell probabilities in the latter. In this article, we propose a hierarchical model for node-parent conditional probabilities, inducing shrinkage to low-dimensional latent parameters aposteriori. We generate samples from the posterior distribution of these latent variables using the Metropolis-adjusted Langevin algorithm within a Gibbs sampler. Moreover, we verify that the full conditional distribution is log-concave under mild conditions, facilitating efficient sampling. We then detail several algorithms for structure learning that incorporate our hierarchical prior and preserve the DAG property. Through simulations, we evaluate the performance of our method for sparse counts, discovering graph structure, and selecting between competing DAGs. We conclude with an application to uncovering prognostic network structure from a breast cancer dataset.

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 hierarchical Dirichlet model for the conditional probability tables of discrete Bayesian networks that induces posterior shrinkage toward low-dimensional latent parameters. Posterior inference on the latents is performed via a Metropolis-adjusted Langevin algorithm (MALA) embedded in a Gibbs sampler, with the claim that the relevant full conditional is log-concave under mild conditions. The prior is incorporated into several structure-learning algorithms that preserve the DAG property. Performance is assessed via simulations on sparse counts, graph recovery, and model selection, followed by an application to prognostic network structure in a breast cancer dataset.

Significance. If the log-concavity result and resulting sampler efficiency hold for general parent sets, the approach supplies a practical Bayesian shrinkage mechanism for high-dimensional discrete BN parameters, potentially improving inference under sparse data relative to saturated or nonparametric alternatives while retaining interpretability through the latent-parameter hierarchy.

major comments (2)
  1. [Abstract / Sampling Method] Abstract and sampling section: the assertion that 'the full conditional distribution is log-concave under mild conditions' is load-bearing for the efficiency of the MALA-within-Gibbs sampler. No explicit statement of the mild conditions, nor a derivation or Hessian analysis for arbitrary parent cardinalities and sparse count regimes, is supplied; without this the claimed mixing guarantees and downstream structure-learning reliability cannot be verified.
  2. [Hierarchical Model] Section on hierarchical model: the low-dimensional latent parameters to which the node-parent CPTs shrink are introduced without a precise mapping from parent-set cardinality to latent dimension. This leaves open whether the shrinkage remains effective (and the log-concavity claim intact) when parent sets grow or when observed counts are extremely sparse, both of which are central to the simulation experiments.
minor comments (2)
  1. [Model Specification] Notation for the hierarchical Dirichlet layers and the latent-parameter dimension should be introduced with an explicit equation or diagram early in the model section to avoid ambiguity when parent sets differ across nodes.
  2. [Simulations] Simulation tables would benefit from reporting effective sample sizes or autocorrelation times for the MALA chains to substantiate the efficiency claim beyond visual trace plots.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive review of our manuscript. We address each major comment below in detail and indicate where revisions will be made to improve clarity and rigor.

read point-by-point responses

  1. Referee: [Abstract / Sampling Method] Abstract and sampling section: the assertion that 'the full conditional distribution is log-concave under mild conditions' is load-bearing for the efficiency of the MALA-within-Gibbs sampler. No explicit statement of the mild conditions, nor a derivation or Hessian analysis for arbitrary parent cardinalities and sparse count regimes, is supplied; without this the claimed mixing guarantees and downstream structure-learning reliability cannot be verified.

    Authors: We agree that an explicit statement of the mild conditions and supporting derivation would strengthen the presentation. The conditions are that all observed counts are strictly positive and that the latent Dirichlet parameters lie in the interior of the probability simplex. In the revised manuscript we will add a precise statement of these conditions in the sampling section and include a full Hessian analysis of the log full-conditional density in a new appendix. The analysis shows that the Hessian remains negative definite for any finite parent cardinality provided the positivity conditions hold, which covers the sparse-count regimes examined in our simulations. We will also note that the MALA step-size tuning used in the experiments already reflects the curvature under these conditions. revision: yes

  2. Referee: [Hierarchical Model] Section on hierarchical model: the low-dimensional latent parameters to which the node-parent CPTs shrink are introduced without a precise mapping from parent-set cardinality to latent dimension. This leaves open whether the shrinkage remains effective (and the log-concavity claim intact) when parent sets grow or when observed counts are extremely sparse, both of which are central to the simulation experiments.

    Authors: The latent dimension d is a fixed hyperparameter chosen independently of parent-set cardinality (typically d=2 or 3 in our experiments) so that the shrinkage strength increases with the size of the CPT. We will revise the hierarchical-model section to state this mapping explicitly: for a node with c categories and parent configuration of size m, the CPT is of dimension c by (product of parent cardinalities), yet the latent vector remains d-dimensional. We will add a short paragraph discussing why the log-concavity result is unaffected by parent-set growth under the stated positivity conditions, and we will include a brief additional simulation with larger parent sets to confirm that posterior shrinkage remains effective even when counts are extremely sparse. revision: yes

Circularity Check

0 steps flagged

No significant circularity: hierarchical prior and MALA-Gibbs sampler rely on standard Bayesian modeling and MCMC techniques

full rationale

The paper proposes a hierarchical Dirichlet model for conditional probability tables that shrinks toward low-dimensional latent parameters, then samples the posterior via MALA embedded in a Gibbs sampler while verifying log-concavity of the full conditional under mild conditions. These steps are presented as direct applications of existing Bayesian hierarchical modeling and Langevin dynamics; no equation reduces a claimed prediction or uniqueness result to a fitted parameter or prior self-definition by construction. The structure-learning algorithms are described as preserving the DAG property via standard topological constraints. No load-bearing self-citation chain or ansatz smuggling is evident in the provided derivation outline. The central claims therefore remain independent of the inputs they are meant to explain.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The model introduces latent low-dimensional parameters and relies on a log-concavity assumption for sampling; hyperparameters of the Dirichlet hierarchy are likely present but unspecified in the abstract.

free parameters (1)
  • Dirichlet concentration hyperparameters
    These control the strength of shrinkage toward the latent structure and are part of the hierarchical prior.
axioms (1)
  • domain assumption Full conditional distributions are log-concave under mild conditions
    Invoked to justify efficient use of Metropolis-adjusted Langevin algorithm within the Gibbs sampler.
invented entities (1)
  • low-dimensional latent parameters no independent evidence
    purpose: To induce posterior shrinkage on the conditional probability tables
    These are the target of the hierarchical prior; no independent evidence outside the model is provided in the abstract.

pith-pipeline@v0.9.0 · 5686 in / 1347 out tokens · 42399 ms · 2026-05-18T15:52:12.065672+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

58 extracted references · 58 canonical work pages · 1 internal anchor

  1. [1]

    Agresti, A. (2002). Categorical Data Analysis . John Wiley & Sons, Incorporated

    work page 2002

  2. [2]

    Alam, M. H., J. Peltonen, J. Nummenmaa, and K. J \"a rvelin (2019). Tree-structured hierarchical D irichlet process. In Distributed Computing and Artificial Intelligence, Special Sessions, 15th International Conference , pp.\ 291--299. Springer International Publishing

    work page 2019

  3. [3]

    Atchad \'e , Y. F. (2006). An adaptive version for the M etropolis adjusted L angevin algorithm with a truncated drift. Methodology and Computing in Applied Probability\/ 8\/ (2), 235--254

    work page 2006

  4. [4]

    Corani, and M

    Azzimonti, L., G. Corani, and M. Scutari (2022). A B ayesian hierarchical score for structure learning from related data sets. International Journal of Approximate Reasoning\/ 142 , 248--265

    work page 2022

  5. [5]

    Corani, and M

    Azzimonti, L., G. Corani, and M. Zaffalon (2017). Hierarchical multinomial- D irichlet model for the estimation of conditional probability tables. In 2017 IEEE International Conference on Data Mining (ICDM) , pp.\ 739--744

    work page 2017

  6. [6]

    Corani, and M

    Azzimonti, L., G. Corani, and M. Zaffalon (2019). Hierarchical estimation of parameters in B ayesian networks. Computational Statistics & Data Analysis\/ 137 , 67--91

    work page 2019

  7. [7]

    Barbieri, M. M. and J. O. Berger (2004). Optimal predictive model selection . The Annals of Statistics\/ 32\/ (3), 870 -- 897

    work page 2004

  8. [8]

    Beinlich, I. A., H. J. Suermondt, R. M. Chavez, and G. F. Cooper (1989). The ALARM monitoring system: A case study with two probabilistic inference techniques for belief networks. In J. Hunter, J. Cookson, and J. Wyatt (Eds.), AIME 89 , Berlin, Heidelberg, pp.\ 247--256. Springer Berlin Heidelberg

    work page 1989

  9. [9]

    Berger, J. O. (1985). Statistical Decision Theory and Bayesian Analysis . Springer Science & Business Media

    work page 1985

  10. [10]

    Bernardo, J. M. and A. F. Smith (1994). Bayesian Theory . John Wiley & Sons

    work page 1994

  11. [11]

    Surjanovic, S

    Biron-Lattes, M., N. Surjanovic, S. Syed, T. Campbell, and A. Bouchard-Cote (2024, 02--04 May). autoMALA : Locally adaptive M etropolis-adjusted L angevin algorithm. In S. Dasgupta, S. Mandt, and Y. Li (Eds.), Proceedings of The 27th International Conference on Artificial Intelligence and Statistics , Volume 238 of Proceedings of Machine Learning Research...

    work page 2024

  12. [12]

    Bishop, C. M. (2006). Pattern Recognition and Machine Learning . Springer

    work page 2006

  13. [13]

    Blei, D. M., A. Y. Ng, and M. I. Jordan (2003). Latent D irichlet allocation. Journal of Machine Learning Research\/ 3 , 993--–1022

    work page 2003

  14. [14]

    Castelletti, F. and S. Peluso (2021). Equivalence class selection of categorical graphical models. Computational Statistics & Data Analysis\/ 164 , 107304

    work page 2021

  15. [15]

    Catal\'an Cerezo, D. (2023). Parametric learning of probabilistic graphical models from multi-sourced data. Master's thesis, Universitat de Barcelona

    work page 2023

  16. [16]

    Catalano, M. and C. Del Sole (2025). Hierarchical random measures without tables. arXiv preprint arXiv:2505.02653\/

    work page arXiv 2025

  17. [17]

    Chakrabarti, A., Y. Ni, E. R. A. Morris, M. L. Salinas, R. S. Chapkin, and B. K. Mallick (2024). Graphical D irichlet process for clustering non-exchangeable grouped data. Journal of Machine Learning Research\/ 25\/ (323), 1--56

    work page 2024

  18. [18]

    Chen, S. X. and J. S. Liu (1997). Statistical applications of the P oisson-binomial and conditional B ernoulli distributions. Statistica Sinica\/ 7\/ (4), 875--892

    work page 1997

  19. [19]

    Chen, Y. and X. Ye (2011). Projection onto a simplex. arXiv preprint arXiv:1101.6081\/

    work page internal anchor Pith review Pith/arXiv arXiv 2011

  20. [20]

    Das, S., Y. Niu, Y. Ni, B. K. Mallick, and D. Pati (2024). Blocked G ibbs sampler for hierarchical D irichlet processes. Journal of Computational and Graphical Statistics\/ In Press

    work page 2024

  21. [21]

    Dawid, A. P. and S. L. Lauritzen (1993). Hyper markov laws in the statistical analysis of decomposable graphical models. The Annals of Statistics\/ 21\/ (3), 1272--1317

    work page 1993

  22. [22]

    Dwivedi, R., Y. Chen, M. J. Wainwright, and B. Yu (2019). Log-concave sampling: M etropolis- H astings algorithms are fast. Journal of Machine Learning Research\/ 20\/ (183), 1--42

    work page 2019

  23. [23]

    Friedman, N. (1998). The B ayesian structural EM algorithm. In Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence , UAI'98, pp.\ 129–--138. Morgan Kaufmann Publishers Inc

    work page 1998

  24. [24]

    Gelman, A., J. B. Carlin, H. S. Stern, D. B. Dunson, A. Vehtari, and D. B. Rubin (2013). Bayesian Data Analysis . Chapman and Hall/CRC

    work page 2013

  25. [25]

    Gabry, I

    Goodrich, B., J. Gabry, I. Ali, and S. Brilleman (2024). rstanarm: Bayesian applied regression modeling via Stan . R package version 2.32.1

    work page 2024

  26. [26]

    Gu, Y. and D. B. Dunson (2023). Bayesian pyramids: identifiable multilayer discrete latent structure models for discrete data. Journal of the Royal Statistical Society Series B: Statistical Methodology\/ 85\/ (2), 399--426

    work page 2023

  27. [27]

    Boffetta, C

    Hashim, D., P. Boffetta, C. La Vecchia, M. Rota, P. Bertuccio, M. Malvezzi, and E. Negri (2016). The global decrease in cancer mortality: trends and disparities. Annals of Oncology\/ 27\/ (5), 926--933

    work page 2016

  28. [28]

    Hausser, J. and K. Strimmer (2009). Entropy inference and the J ames- S tein estimator, with application to nonlinear gene association networks. Journal of Machine Learning Research\/ 10 , 1469–--1484

    work page 2009

  29. [29]

    Geiger, and D

    Heckerman, D., D. Geiger, and D. M. Chickering (1995). Learning B ayesian networks: The combination of knowledge and statistical data. Machine Learning\/ 20\/ (3), 197--243

    work page 1995

  30. [30]

    Hoffman, M. D., A. Gelman, et al. (2014). The N o- U - T urn sampler: adaptively setting path lengths in H amiltonian M onte C arlo. Journal of Machine Learning Research\/ 15\/ (1), 1593--1623

    work page 2014

  31. [31]

    Kass, R. E. and A. E. Raftery (1995). Bayes factors. Journal of the American Statistical Association\/ 90\/ (430), 773--795

    work page 1995

  32. [32]

    Kitson, N. K., A. C. Constantinou, Z. Guo, Y. Liu, and K. Chobtham (2023). A survey of B ayesian network structure learning. Artificial Intelligence Review\/ 56\/ (8), 8721--8814

    work page 2023

  33. [33]

    Kong, L., G. Chen, B. Huang, E. Xing, Y. Chi, and K. Zhang (2024). Learning discrete concepts in latent hierarchical models. In A. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang (Eds.), Advances in Neural Information Processing Systems , Volume 37, pp.\ 36938--36975. Curran Associates, Inc

    work page 2024

  34. [34]

    Lewis, A

    Kratzer, G., F. Lewis, A. Comin, M. Pittavino, and R. Furrer (2023). Additive B ayesian network modeling with the R package abn. Journal of Statistical Software\/ 105\/ (8), 1–41

    work page 2023

  35. [35]

    Caron, S

    Liang, X., A. Caron, S. Livingstone, and J. Griffin (2023). Structure learning with adaptive random neighborhood informed MCMC . Advances in Neural Information Processing Systems\/ 36 , 40760--40772

    work page 2023

  36. [36]

    Wang, and Y

    Lin, Z., Y. Wang, and Y. Hong (2022). The P oisson multinomial distribution and its applications in voting theory, ecological inference, and machine learning. arXiv preprint arXiv:2201.04237\/

    work page arXiv 2022

  37. [37]

    Lindley, D. V. (1964). The B ayesian analysis of contingency tables. The Annals of Mathematical Statistics\/ 35\/ (4), 1622--1643

    work page 1964

  38. [38]

    Lucas, P. J., L. C. van der Gaag , and A. Abu-Hanna (2004). Bayesian networks in biomedicine and health-care. Artificial Intelligence in Medicine\/ 30\/ (3), 201--214. Bayesian Networks in Biomedicine and Health-Care

    work page 2004

  39. [39]

    Marshall, T. and G. Roberts (2012). An adaptive approach to L angevin MCMC . Statistics and Computing\/ 22 , 1041--1057

    work page 2012

  40. [40]

    Meinshausen, N. and P. B \"u hlmann (2006). High-dimensional graphs and variable selection with the Lasso . The Annals of Statistics\/ 34\/ (3), 1436--1462

    work page 2006

  41. [41]

    Nolan, E., G. J. Lindeman, and J. E. Visvader (2023). Deciphering breast cancer: from biology to the clinic. Cell\/ 186\/ (8), 1708--1728

    work page 2023

  42. [42]

    Pearl, J. (1985). Bayesian networks: A model of self-activated memory for evidential reasoning. In Proceedings of the 7th conference of the Cognitive Science Society, University of California, Irvine, CA, USA , pp.\ 15--17

    work page 1985

  43. [43]

    Pearl, J. (1988). Probabilistic reasoning in intelligent systems: networks of plausible inference . Morgan Kaufmann

    work page 1988

  44. [44]

    Pearl, J. (2009). Causality: Models, Reasoning, and Inference . Cambridge University Press

    work page 2009

  45. [45]

    Pereira, B., S.-F. Chin, O. M. Rueda, H.-K. M. Vollan, E. Provenzano, H. A. Bardwell, M. Pugh, L. Jones, R. Russell, S.-J. Sammut, et al. (2016). The somatic mutation profiles of 2,433 breast cancers refine their genomic and transcriptomic landscapes. Nature Communications\/ 7\/ (1), 11479

    work page 2016

  46. [46]

    Perotte, A., F. Wood, N. Elhadad, and N. Bartlett (2011). Hierarchically supervised latent D irichlet allocation. In J. Shawe-Taylor, R. Zemel, P. Bartlett, F. Pereira, and K. Weinberger (Eds.), Advances in Neural Information Processing Systems , Volume 24. Curran Associates, Inc

    work page 2011

  47. [47]

    Buntine, G

    Petitjean, F., W. Buntine, G. I. Webb, and N. Zaidi (2018). Accurate parameter estimation for B ayesian network classifiers using hierarchical D irichlet processes. Machine Learning\/ 107\/ (8), 1303--1331

    work page 2018

  48. [48]

    Rijmen, F. (2008). Bayesian networks with a logistic regression model for the conditional probabilities. International Journal of Approximate Reasoning\/ 48\/ (2), 659--666

    work page 2008

  49. [49]

    Roberts, G. O. and J. S. Rosenthal (1998). Optimal scaling of discrete approximations to L angevin diffusions. Journal of the Royal Statistical Society: Series B (Statistical Methodology)\/ 60\/ (1), 255--268

    work page 1998

  50. [50]

    Ronning, G. (1989). Maximum likelihood estimation of D irichlet distributions. Journal of statistical computation and simulation\/ 34\/ (4), 215--221

    work page 1989

  51. [51]

    Scutari, M. (2010). Learning B ayesian networks with the bnlearn R package. Journal of Statistical Software\/ 35\/ (3), 1--22

    work page 2010

  52. [52]

    and J.-B

    Scutari, M. and J.-B. Denis (2021). Bayesian networks: with examples in R . Chapman and Hall/CRC

    work page 2021

  53. [53]

    Ferlay, R

    Sung, H., J. Ferlay, R. L. Siegel, M. Laversanne, I. Soerjomataram, A. Jemal, and F. Bray (2021). Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA: a Cancer Journal for Clinicians\/ 71\/ (3), 209--249

    work page 2021

  54. [54]

    Jordan, M

    Teh, Y., M. Jordan, M. Beal, and D. Blei (2006). Hierarchical D irichlet processes. Journal of the American Statistical Association\/ 101\/ (476), 1566--1581

    work page 2006

  55. [55]

    Trayes, K. P. and S. E. Cokenakes (2021). Breast cancer treatment. American Family Physician\/ 104\/ (2), 171--178

    work page 2021

  56. [56]

    Wood, S. N. (2011). Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. Journal of the Royal Statistical Society: Series B\/ 73\/ (1), 3--36

    work page 2011

  57. [57]

    Petitjean, and W

    Zhang, H., F. Petitjean, and W. Buntine (2020). Bayesian network classifiers using ensembles and smoothing. Knowledge and Information Systems\/ 62 , 3457--3480

    work page 2020

  58. [58]

    Zhang, J., Y. Song, C. Zhang, and S. Liu (2010). Evolutionary hierarchical D irichlet processes for multiple correlated time-varying corpora. In Proceedings of the 16th ACM SIGKDD International Conference on Knowledge Discovery and Data mining , pp.\ 1079--1088

    work page 2010