Professional Documents
Culture Documents
Bayseian Estimate
Bayseian Estimate
Pattern
Classification
Chapter 3:
Maximum-Likelihood and Bayesian
Parameter Estimation (part 2)
Bayesian Estimation (BE)
Bayesian Parameter Estimation: Gaussian Case
Bayesian Parameter Estimation: General
Estimation
Problems of Dimensionality
Computational Complexity
Component Analysis and Discriminants
Hidden Markov Models
2
Example on MLE
But we know
p ( xk | ) ~ N ( , ) and p ( ) ~ N ( 0 , 0 )
2 2
P ( | D) ~ N ( n , )2
n
1 n 1 1 n
0
p ( | D) exp 2 2 2 2
2 0
2
xk 2
0
k 1
1 n 1 n n 0
and 2 2 ˆ n 2
2
n 2
02 n 0
Pattern Classification, Chapter 3
8
4
9
n 02 2
n ˆ
2 n
0
n0 0 n 0
2 2 2
0
2 2
and n
2
n 02 2
From these equations we see as n increases:
the variance decreases monotonically
the estimate of p(|D) becomes more peaked
We know 2
and how to compute n and σ n2
(Desired class-conditional density P(x | Dj, j))
Therefore: using P(x | Dj, j) together with P(j)
And using Bayes formula, we obtain the
Bayesian classification rule:
Max P ( j | x , D Max P ( x | j , D j ) P ( j )
j j
Pattern Classification, Chapter 3
11
4
12
Problems of Dimensionality
Problems involving 50 or 100 features are
common (usually binary valued)
Note: microarray data might entail ~20000
real-valued features
Classification accuracy dependant on
dimensionality
amount of training data
discrete vs continuous
where : r 2 ( 1 2 ) t 1 ( 1 2 )
r 2 is the squared Mahalanobi s distance
lim P ( error ) 0
r
diag(12 , 22 ,..., 2d )
2
i d i2
2
r i1
i 1 i
Do we remember what conditional independence is?
Example for binary features:
Let pi= Pr[xi=1|1] then P(x|1) is the product of the pi
Most useful features are the ones for which the difference between
the means is large relative to the standard deviation
Doesn’t require independence
“bigoh” notation
f(x) = O(h(x)) “big oh of h(x)”
If: (c 0 , x 0 ) 2 ; f (x ) c 0 h(x )
(An upper bound on f(x) grows no worse than h(x) for
sufficiently large x!)
f(x) = 2+3x+4x2
g(x) = x2
f(x) = O(x2)
Pattern Classification, Chapter 3
20
7
21
Total = O(d2n)
Total for c classes = O(cd2n) O(d2n)
Overfitting
Dimensionality of model vs size of training data
Issue: not enough data to support the model
Possible solutions:
Reduce model dimensionality
Make (possibly incorrect) assumptions to better estimate
Overfitting
Estimate better
use data pooled from all classes
normalization issues
use pseudo-Bayesian form 0 + (1-)n
“doctor” by thresholding entries
reduces chance correlations
assume statistical independence
zero all off-diagonal elements
Pattern Classification, Chapter 3
24
25
Shrinkage
Σ( ) (1 ) Σ I for 0 1
Pattern Classification, Chapter 3
25
Component Analysis and Discriminants 26
Markov Chains
Applications:
speech recognition, gesture recognition,
parts of speech tagging and DNA sequencing,
= (aij, T)
P(T | ) = a14 . a42 . a22 . a21 . a14 .
P((1) = i)