Mixtures and Hierarchical
Mixtures of Experts
Background ----- Mixtures and hierarchical mixtures of
experts (ME and HME) are generalizations of the standard
mixture models to the regression context, which combine simple models
to form more powerful mixtures. Some pioneering works are done by M.
Jordan, R.
Jacobs, G. Hinton,
M. Tanner,
A. Zeevi,
and S. Waterhouse who provides a
bibliography and computing software.
See an online document for an overview of theoretical
developments of ME and HME:
Mixtures
and Hiearchical Mixtures of Experts: A Theoretical Overview. (Html)
Technical Reports
(A
sequence of papers on the theory of mixtures-of-experts: on
approximation; frequentists' and Bayesian consistency; asymptotic
normality; identifiability; VC dimension)
On the Approximation Rate of
Hierarchical Mixtures-of-Experts for Generalized Linear Models.
(with Tanner, M. A.) PostScript Neural
Computation 1999
Hierarchical Mixtures-of-Experts
for Exponential Family Regression Models: Approximation and Maximum
Likelihood Estimation. (with Tanner, M. A.) PostScript
Annals of Statistics 1999
On the Asymptotic Normality of
Hierarchical Mixtures-of-Experts for Generalized Linear Models.
(with Tanner, M. A.) PostScript IEEE
Trans. Information Theory 2000
On the Identifiability of
Mixtures-of-Experts. (with Tanner, M. A.) PostScript
Neural Networks 1999
The VC Dimension for Mixtures of
Binary Classifiers. PostScript Neural
Computation 2000
Hierarchical Mixtures-of-Experts for Generalized Linear
Models: Some Results on Denseness and consistency. (with Tanner,
M.A.; a short survey paper) PostScript
Proceedings of the Seventh International Workshop on Artificial
Intelligence and Statistics, 1999. Morgan Kaufmann.
On Consistency of Bayesian Inference with Mixtures of Logistic Regression. (with Ge, Y.)
Pdf (To appear in Neural Computation, 2006.)
Regression Analysis of Mean Lifetime:
Exploring Nonlinear Relationship with Heteroscedasticity.
(with Sun, Z.)
The International Journal of Biostatistics: Vol. 3 : Iss. 1, Article 5.
(2007).
Available at: http://www.bepress.com/ijb/vol3/iss1/5
(back to the referring page)
Return to main
NWU Statistics page
(last updated on 5/20/2005.)