Kernel Methods Part 2

Kernel Methods
Part 2
Bing Han
June 26, 2008
Local Likelihood

Logistic Regression
Logistic Regression

After a simple calculation, we get

We denote the probabilities

Logistic regression models are usually fit by
maximum likelihood
Local Likelihood

The data has feature xi and classes {1,2,…,J}

The linear model is
Local Likelihood

Local logistic regression
N
P̂r(G  g i | X  x0 )   K  ( x0 , xi ) Pr(G  g i | X  xi )
i 1

The local log-likelihood for this J class model
Kernel Density Estimation
Kernel Density Estimation

We have a random sample x1, x2, …,xN, we
want to estimate probability density
A natural local estimate

Smooth Pazen estimate

Kernel Density Estimation

A popular choice is Gaussian Kernel

A natural generalization of the Gaussian
density estimate by the Gaussian product
kernel
Kernel Density Classification
fˆ j ( X )

Density estimates
Estimates of class priors ˆ j

By Bayes’ theorem

Kernel Density Classification
Naïve Bayes Classifier

Assume given a class G=j, the features Xk
are independent
Naïve Bayes Classifier

A generalized additive model
Similar to logistic regression
Radial Basis Functions


Functions can be represented as expansions
in basis functions
Radial basis functions treat kernel functions
as basis functions. This lead to model
Method of learning parameters

Optimize the sum-of squares with respect to
all the parameters:
Radial Basis Functions


Reduce the parameter set and assume a
constant value for  j   it will produce an
undesirable effect.
Renormalized radial basis functions
Radial Basis Functions
Mixture models

Gaussian mixture model for density
estimation

In general, mixture models can use any
component densities. The Gaussian mixture
model is the most popular.
Mixture models

If

If

Where
, Radial basis expansion
, kernel density estimate
Mixture models


The parameter are usually fit by maximum
likelihood, such as EM algorithm
The mixture model also provides an estimate
of the probability that observation i belong to
component m

Questions?