## Computational and Inferential Difficulties With Mixture Posterior Distributions (1999)

Venue: | Journal of the American Statistical Association |

Citations: | 111 - 12 self |

### BibTeX

@ARTICLE{Celeux99computationaland,

author = {Gilles Celeux and Merrilee Hurn and Christian P. Robert},

title = {Computational and Inferential Difficulties With Mixture Posterior Distributions},

journal = {Journal of the American Statistical Association},

year = {1999},

volume = {95},

pages = {957--970}

}

### Years of Citing Articles

### OpenURL

### Abstract

This paper deals with both exploration and interpretation problems related to posterior distributions for mixture models. The specification of mixture posterior distributions means that the presence of k! modes is known immediately. Standard Markov chain Monte Carlo techniques usually have difficulties with well-separated modes such as occur here; the Markov chain Monte Carlo sampler stays within a neighbourhood of a local mode and fails to visit other equally important modes. We show that exploration of these modes can be imposed on the Markov chain Monte Carlo sampler using tempered transitions based on Langevin algorithms. However, as the prior distribution does not distinguish between the different components, the posterior mixture distribution is symmetric and thus standard estimators such as posterior means cannot be used. Since this is also true for most non-symmetric priors, we propose alternatives for Bayesian inference for permutation invariant posteriors, including a cluster...

### Citations

903 | Monte Carlo Statistical Methods - Robert, Casella - 2004 |

440 | On Bayesian analysis of mixtures with an unknown number of components
- Richardson, Green
- 1997
(Show Context)
Citation Context ...e literature (for example, Titterington, Smith and Makov, 1985), and is usually solved by imposing an identifying constraint on the parameters, for instance p 1s: : :sp k (Mengersen and Robert, 1996, =-=Richardson and Green, 1997-=-, or Stephens, 1997). The identifiability constraint may even be the starting point for a non-informative modelling, as in Mengersen and Robert (1996) or Roeder and Wasserman (1997). The computational... |

324 | Marginal Likelihood from the Gibbs Output - Chib - 1995 |

119 | Efficient Metropolis jumping rules - Gelman, Roberts, et al. - 1996 |

114 | Practical bayesian density estimation using mixtures of normals - ROEDER, WASSERMANN - 1997 |

70 | Sampling from multimodal distributions using tempered transitions - Neal - 1996 |

53 |
Bayesian methods for mixtures of normal distributions
- STEPHENS
- 1997
(Show Context)
Citation Context ...erington, Smith and Makov, 1985), and is usually solved by imposing an identifying constraint on the parameters, for instance p 1s: : :sp k (Mengersen and Robert, 1996, Richardson and Green, 1997, or =-=Stephens, 1997-=-). The identifiability constraint may even be the starting point for a non-informative modelling, as in Mengersen and Robert (1996) or Roeder and Wasserman (1997). The computational difficulties resul... |

44 | The multiple-try method and local optimization in Metropolis sampling - Liu, Liang, et al. - 2000 |

41 | Exponential convergence of Langevin diffusions and their discrete approximations - Roberts, Tweedie - 1996 |

33 | Testing for mixtures: A Bayesian entropic approach (with discussion - MENGERSEN, ROBERT - 1996 |

24 | Estimation of finite mixture distributions by Bayesian sampling - Diebolt, Robert - 1994 |

17 | Bayesian inference for mixtures: The label-switching problem - Celeux - 1998 |

16 |
New loss functions in Bayesian imaging
- Rue
- 1995
(Show Context)
Citation Context ...meter labeling. It has recently been demonstrated however that apparently intractable estimators in Bayesian image analysis can be approximated using Markov chain Monte Carlo (Frigessi and Rue, 1997; =-=Rue, 1995-=-). We consider a similar approach here, suggesting two loss functions, the first when inference for the parameters is the issue, and the second when the predictive mixture distribution is of more inte... |

14 | Errors in binary images and a L" version of the Hausdorff metric. Nieuw Archief voor Wiskunde
- Baddeley
(Show Context)
Citation Context ...or is a way of measuring distance between two point configurations (of possibly different dimensions in a more general setting). We suggest the following, loosely based on the Baddeley \Delta metric (=-=Baddeley, 1992-=-) used as a loss function by Frigessi and Rue (1997) in a discrete pixel setting. The \Delta metric measures the distance between two binary images and is based on the difference for every pixel betwe... |

7 | Estimation des paramètres d’un mélange par échantillonnage bayésien. Notes aux Comptes–Rendus de l’Académie des Sciences I - Diebolt, Robert - 1990 |

6 |
Inference in mixture models
- Robert
- 1996
(Show Context)
Citation Context ...cognised in the literature (for example, Titterington, Smith and Makov, 1985), and is usually solved by imposing an identifying constraint on the parameters, for instance p 1s: : :sp k (Mengersen and =-=Robert, 1996-=-, Richardson and Green, 1997, or Stephens, 1997). The identifiability constraint may even be the starting point for a non-informative modelling, as in Mengersen and Robert (1996) or Roeder and Wasserm... |

5 | Geometric and subgeometric convergence of diffusions with given stationary distributions, and their discretizations - Stramer, Tweedie - 1997 |

4 |
Reparametrization issues in mixture estimation and their bearings on the Gibbs sampler
- Robert, Mengersen
- 1999
(Show Context)
Citation Context ...j ; �� 2 j ) and k X j=1 p j Exp( j ): However the approach suggested here extends directly to other types of mixtures (Poisson, Pareto, etc.), and also to other categories of latent variable mode=-=ls (Robert, 1998-=-) and non-identifiable problems. Similarly, we only consider the case of symmetric priors, but obviously, modest departures from symmetry would lead to similar behaviour when the prior does not counte... |

2 | Bayesian image classification using Baddeley’s delta loss - Rue, Frigessi - 1997 |

2 | Discussion of Richardson and Green' paper - Robert - 1997 |

2 | MCMC specificities of latent variable models
- Robert
- 1998
(Show Context)
Citation Context ...j ; �� 2 j ) and k X j=1 p j Exp( j ): However the approach suggested here extends directly to other types of mixtures (Poisson, Pareto, etc.), and also to other categories of latent variable mode=-=ls (Robert, 1998-=-) and non-identifiable problems. Similarly, we only consider the case of symmetric priors, but obviously, modest departures from symmetry would lead to similar behaviour when the prior does not counte... |

1 | Annealing Monte Carlo maximum likelihood with application to pedigree analysis - Geyer - 1995 |

1 | Markov chain Monte Carlo control spreadsheets for exponential mixture estimation - Gruet, Philippe - 1999 |

1 | MCMC diagnostics in action - Guihenneuc, Knight, et al. - 1999 |