Choosing starting values for the EM algorithm for getting the highest likelihood in multivariate Gaussian mixture models

By Christophe Biernacki, Gilles Celeux, Gérard Govaert. Computational Statistics & Data Analysis, 2003

685 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Statistics and Probability

View Paper: https://www.semanticscholar.org/paper/42d87434ad937a3489d0379278b2d47693391258

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

A robust EM clustering algorithm for Gaussian mixture models (2012, background): “Although some authors have considered the initial problems for the EM algorithm [14,15] and some have considered estimation of the number of components [15,16], there has been less consideration about robustness to initial values associated with the number of components for the EM algorithm.”
Regression Density Estimation With Variational Methods and Stochastic Approximation (2012, methodology): “This strategy of “short runs” to identify a promising solution to follow to full convergence is similar to the one recommended for initialization of the EM algorithm for maximum likelihood estimation of mixture models by Biernacki, Celeux, and Govaert (2003).”
Finite Mixtures of Generalized Linear Regression Models (2008, methodology): “hierarchical clustering, k -means, spectral clustering Use short runs of EM, CEM or SEM with different initializations (Biernacki et al., 2003). Use different subsets of the complete data: sampling (Wehrens et al., 2004) and incremental method (Fraley et al., 2005). ⇒ Scharl et al. (2010) recommend ”
FlexMix Version 2: Finite Mixtures with Concomitant Variables and Varying and Constant Parameters (2008, methodology): “Diﬀerent initialization strategies for the EM algorithm have been proposed, as its convergence to the optimal solution depends on the initialization (Biernacki, Celeux, and Govaert 2003; Karlis and Xekalaki 2003).”
Finite mixture models and model-based clustering (2010, methodology): “This is especially important for initialization strategies involving starting the EM algorithm at random points (emEM [17] and RndEM [78]).”
Editorial: recent developments in mixture models (2003, methodology): “Contributions are given here by McLachlan et al., 2003(SI) as well as Biernacki et al., 2003(SI).”
Local solutions in the estimation of growth mixture models. (2006, background): “Future work may help to increase the efficiency and the probability of locating the maximum likelihood solution for a given model by implementing more intelligent strategies for selecting start values ( Biernacki, Celeux, & Govaert, 2003 ).”
A simple nonparametric approach to estimating the distribution of random coefficients in structural models (2016, neutral): “Our estimator for models with homogeneous parameters is defined as (similarly to (7))”
Semiparametric Mixture Models for Multivariate Count Data, with Application (2004, positive): “Moreover, we implemented a modified EM algorithm which adopts as starting values for model parameters the values estimated by a CEM (Classification EM) algorithm as suggested by Biernacki et al. (2003). It is simple, performs well in a lot of situations regardless of the particular form for the mixi”