Dealing with label switching in mixture models

By Matthew Stephens. 2000

1,189 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/e3df93eb425d83dd0342aa59a1fa9045d111637e

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Statistics and Probability

Citations in the corpus

Singularity, misspecification and the convergence rate of EM (2018, background): “The main difficulty for analyzing the MLE in such settings can be attributed to the associated geometry of the parameters and the inevitable label switching between the mixtures [23, 25].”
A new statistical method for haplotype reconstruction from population data. (2001, methodology): “We do this by specifying a loss function , which gives the loss for reporting the estimates when the true haplotypes are H, SSD SSD ˆ ˆ L(H ,Q; H) (H ,Q) and attempting to minimize the posterior expected loss by means of methods analogous to those in Stephens (2000). The particular loss function we ”
Bayesian Methods for Hidden Markov Models (2002, methodology): “Stephens (1997, 2000b) proposed a method for identifying mixture components when prior information about h is unavailable.”
pong: fast analysis and visualization of latent clusters in population genetic data (2015, background): “This is known as label switching: for a fixed value of K and identical genetic input, column ~q j in the Q matrix produced by one run may not correspond to column ~q j in the Q matrix produced by another run (Jakobsson and Rosenberg, 2007; Jasra et al., 2005; Stephens, 2000).”
Markov Chain Monte Carlo Methods and the Label Switching Problem in Bayesian Mixture Modeling (2005, methodology): “As noted by Richardson and Green (1997) and Stephens (2000a), mixing within k is often improved by using variable dimension samplers.”
A simple new approach to variable selection in regression, with application to genetic fine mapping (2018, background): “For example, the posterior mean will not be a sensible estimate for b1, : : : , bL because it averages over the L! modes (Stephens, 2000).”
Consistency of the MLE under mixture models (2016, background): “ switching. Quotient topology as suggested by Redner (1981) can be used to avoid this problem. Label switching leads to technical diﬃculties for Bayesian analysis; see Fru¨hwirth-Schnatter (2011) and Stephens (2000). (KW2) Continuity: The component parameter space Θ is a closed set. For all x and an”
Testing hypotheses via a mixture estimation model (2014, background): “Another computational challenge that plagues Bayesian estimation for most mixture models is “label switching” (Celeux et al., 2000; Stephens, 2000; Jasra et al., 2005).”
Deviance information criteria for missing data models (2006, background): “…the MCMC sample means of the simulated values of θ, but, as mentioned in Section 3.1 and further discussed in Celeux et al. (2000), these estimators are not meaningful if no identifiability constraint is imposed on the model (see also Stephens 2000), and even then they often lead to negative pD’s.”
Fast Bayesian Inference in Dirichlet Process Mixture Models (2011, methodology): “In addition, as noted by Stephens (2000), there are in general very many choices of γ having identical or close to identical marginal likelihoods.”
Estimating Marginal Likelihoods for Mixture and Markov Switching Models Using Bridge Sampling Techniques (2004, neutral): “The constraint is included to achieve formal identification of the model which is identified only up to permutations of the labels of the states I, (see Celeux et al. 2000; Stephens 2000; Friihwirth-Schnatter 2001b).”
A nonparametric view of network models and Newman–Girvan and other modularities (2009, background): “However, although the surface is multimodal, in the examples we have considered and generally when the signal is strong, optimization from several starting points using a label switching algorithm (19) works well.”
Mixture of Regression Models With Varying Mixing Proportions: A Semiparametric Approach (2012, background): “Note that in the M-step, the nonparametric functions are estimated simultaneously at a set of grid points; thus, the classification probabilities in the E-Step can be estimated globally to avoid the label switch problem (see, e.g., Celeux, Hurn, and Robert 2000; Stephens 2000; Yao and Lindsay 2009).”
DESCRIBING DISABILITY THROUGH INDIVIDUAL-LEVEL MIXTURE MODELS FOR MULTIVARIATE BINARY DATA. (2007, background): “Note that label switching could present a problem for calculating posterior means and the model selection criteria based on them [Stephens (2000)].”
The Gibbs and splitmerge sampler for population mixture analysis from genetic data with incomplete baselines (2006, background): “(2000) view the poor mixing in a positive light because the label-switching problem of mixture posterior distributions (Diebolt and Robert 1994; Richardson and Green 1997; Stephens 2000) is averted.”
A Multilevel Mixture IRT Model With an Application to DIF (2010, background): “As noted earlier, label switching can be observed, when distinct jumps occur in the traces of a parameter and when the density for the parameter has multiple modes (Stephens, 2000).”
Rapid changes in growth, condition, size and age of small pelagic fish in the Mediterranean (2014, neutral): “For our data, this improved the results significantly as label switching (when within or between 175 chains labels switch between peaks) was avoided, allowing the model to converge (Stephens 2000; Jasra et al. 2005).”