28
Characterizing the Function Space for Bayesian Kernel Models Natesh S. Pillai, Qiang Wu, Feng Liang Sayan Mukherjee and Robert L. Wolpert JMLR 2007 Presented by: Mingyuan Zhou Duke University January 20, 2012

Presented by: Mingyuan Zhou Duke University January 20, 2012

  • Upload
    adora

  • View
    31

  • Download
    0

Embed Size (px)

DESCRIPTION

Characterizing the Function Space for Bayesian Kernel Models Natesh S. Pillai, Qiang Wu, Feng Liang Sayan Mukherjee and Robert L. Wolpert JMLR 2007. Presented by: Mingyuan Zhou Duke University January 20, 2012. Outline. Reproducing kernel Hilbert space (RKHS) Bayesian kernel model - PowerPoint PPT Presentation

Citation preview

Page 1: Presented by: Mingyuan Zhou Duke University January 20, 2012

Characterizing the Function Space for Bayesian Kernel Models

Natesh S. Pillai, Qiang Wu, Feng LiangSayan Mukherjee and Robert L. Wolpert

JMLR 2007

Presented by: Mingyuan ZhouDuke UniversityJanuary 20, 2012

Page 2: Presented by: Mingyuan Zhou Duke University January 20, 2012

Outline

• Reproducing kernel Hilbert space (RKHS)• Bayesian kernel model

– Gaussian processes– Levy processes

• Gamma process• Dirichlet process• Stable process

– Computational and modeling considerations• Posterior inference• Discussion

Page 3: Presented by: Mingyuan Zhou Duke University January 20, 2012

RKHS

In functional analysis (a branch of mathematics), a reproducing kernel Hilbert space is a Hilbert space of functions in which pointwise evaluation is a continuous linear functional. Equivalently, they are spaces that can be defined by reproducing kernels.

http://en.wikipedia.org/wiki/Reproducing_kernel_Hilbert_space

Page 4: Presented by: Mingyuan Zhou Duke University January 20, 2012

A finite kernel based solution

The direct adoption of the finite representation is not a fully Bayesian model since it depends on the (arbitrary) training data sample size . In addition, this prior distribution is supported on a finite-dimensional subspace of the RKHS. Our coherent fully Bayesian approach requires the specification of a prior distribution over the entire space H.

Page 5: Presented by: Mingyuan Zhou Duke University January 20, 2012

Mercer kernel

Page 6: Presented by: Mingyuan Zhou Duke University January 20, 2012

Bayesian kernel model

Page 7: Presented by: Mingyuan Zhou Duke University January 20, 2012
Page 8: Presented by: Mingyuan Zhou Duke University January 20, 2012

Properties of the RKHS

Page 9: Presented by: Mingyuan Zhou Duke University January 20, 2012

Properties of the RKHS

Page 10: Presented by: Mingyuan Zhou Duke University January 20, 2012

Bayesian kernel models and integral operators

Page 11: Presented by: Mingyuan Zhou Duke University January 20, 2012
Page 12: Presented by: Mingyuan Zhou Duke University January 20, 2012

Two concrete examples

Page 13: Presented by: Mingyuan Zhou Duke University January 20, 2012

Two concrete examples

Page 14: Presented by: Mingyuan Zhou Duke University January 20, 2012

Bayesian kernel models

Page 15: Presented by: Mingyuan Zhou Duke University January 20, 2012

Gaussian processes

Page 16: Presented by: Mingyuan Zhou Duke University January 20, 2012

Levy processes

Page 17: Presented by: Mingyuan Zhou Duke University January 20, 2012

Levy processes

Page 18: Presented by: Mingyuan Zhou Duke University January 20, 2012

Poisson random fields

Page 19: Presented by: Mingyuan Zhou Duke University January 20, 2012

Poisson random fields

Page 20: Presented by: Mingyuan Zhou Duke University January 20, 2012

Dirichlet Process

Page 21: Presented by: Mingyuan Zhou Duke University January 20, 2012

Symmetric alpha-stable processes

Page 22: Presented by: Mingyuan Zhou Duke University January 20, 2012

Symmetric alpha-stable processes

Page 23: Presented by: Mingyuan Zhou Duke University January 20, 2012

Computational and modeling considerations

• Finite approximation for Gaussian processes

• Discretization for pure jump processes

Page 24: Presented by: Mingyuan Zhou Duke University January 20, 2012

Posterior inference

• Levy process model

– Transition probability proposal– The MCMC algorithm

Page 25: Presented by: Mingyuan Zhou Duke University January 20, 2012
Page 26: Presented by: Mingyuan Zhou Duke University January 20, 2012

Classification of gene expression data

Page 27: Presented by: Mingyuan Zhou Duke University January 20, 2012

Classification of gene expression data

Page 28: Presented by: Mingyuan Zhou Duke University January 20, 2012

Discussion• This paper formulates a coherent Bayesian perspective for

regression using a RHKS model.• The paper stated an equivalence under certain conditions of

the function class G and the RKHS induced by the kernel. This implies: – (a) a theoretical foundation for the use of Gaussian processes, Dirichlet

processes, and other jump processes for non-parametric Bayesian kernel models.

– (b) an equivalence between regularization approaches and the Bayesian kernel approach.

– (c) an illustration of why placing a prior on the distribution is natural approach in Bayesian non-parametric modelling.

• A better understanding of this interface may lead to a better understanding of the following research problems:– Posterior consistency– Priors on function spaces– Comparison of process priors for modeling– Numerical stability and robust estimation