Professional Documents
Culture Documents
SVM
SVM
2553
: .
:
505261
Tadee
Name
:Ms.Chalotorn
Title
Advisor
Keywords
: Dr.Ut Goenchanart
: statistical learning
ID
505261
Abstract
Support Vector Machines are supervised learning machines based on statistical learning
theory thatcan be used for pattern recognition and regression. Statistical learning theory can identify
ratherprecisely the factors that need to be taken into account to learn successfully certain simple
typesof algorithms , however , real-world applications usually need more complex models and
algorithms(such as neural networks), that makes them much harder to analyse theoretically.
SVMs can beseen as lying at the intersection of learning theory and practice. They construct
models that arecomplex enough (containing a large class of neural networks for instance) and yet that
are simpleenough to be analysed mathematically. This is because an SVM can be seen as a linear
algorithm in a high-dimensional space .
In this document , we will primarily concentrate on Support Vector Machines as used in
pattern recognition. In the first section we will introduce pattern recognition and hyperplane
classifiers,simple linear machines on which SVMs are based. We will then proceed to see how SVMs
are ableto go beyond the limitations of linear learning machines by introducing the kernel function,
whichpaves the way to find a nonlinear decision function. Finally, we sum it all up and mention
someareas in which Support Vector Machines have been applied and given excellent results.
CPE 489 (Seminar in Computer Engineering)
Support Vector Machines
.....
........
...
..
.
1.1 Introduction .....
1.2 ....
1.3 .....
1
1
1
1
2
2
3
6
8
9
9
2.7 ........
2.8 .
2.9 ..
2.10
2.11 Multiclass SVM
10
10
11
14
18
()
.......
20
3.1 ..
3.1.1 .........
3.1.2 ...
3.1.3 ...
3.1.4 Neural Network ..
20
21
21
21
22
.. ..
4.1 ......
23
23
24
2.1
2.2.1
2.2.2
2.10.1
..............................................................................
hyperplane margin ...........................................................................................
Hyperplane ..............................................................................................................
Linear PCA .................................................................................................................
3
4
5
13
1.1. Introduction
, ,
1.2.
1.3.
Machine
,
,
2
Support Vector Machines
2.1. (SVMs)
SVM
hyperplane
hyperplanes
hyperplane datapoints
SVM
K (x, y)
hyperplanes
hyperplanes
hyperplane x
hyperplane
nstant
con
K (x, y) Y x,
X corresppoinding x
x hyperrplane
c
convoluted
p - 1 hyperplaane hyperplannes
hyperplanee
hyperrplane
hyperpllane
hypperplane
2.1. 1
2.2. Forrmalizationn
D, n
,
1,1
C i 1
-1
p - h
hyperplane
I = 1 I = -- 1 hyyperplane
2.2.1
2 hyperpllane margin
m 2
SVM
M
1
2
http://en.wikipedia.org/wiki/Support__vector_machine
http://en.wikipedia.org/wiki/Support__vector_machine
hyperplanne
hyperplane
h
| ||
||
b
hypperplanes
hyperplanees
hypperplanes
hyperplanes
h
||
||
||W|||
2.22.23 hyperplane
http://wwww.cs.brownn.edu/coursess/archive/20006-2007/.../Buurges98.pdf
WX
for
:
1, , )
(
|| ||
2.3. Primal
||W|| W
||W|| || || ( w
b) (QP)
1
|| ||
2
(
1, , )
1
min
, ,
1
| |
2
hyperplanes
1 0
min
max
| |
1
cx
,
:
1
2.4. Dual
unconstrained hyperplane
,
|| ||
SVM
(
1, , )
0
0
2.5. hyperplanes
hyperplane hyperplanes
hyperplanes
hyperplane B = 0
2.6. Traansductive
Transducctive
SVMs
D
|
transduuctive
:
( W , b ,
1
| |
2
(
1
1
1,1
Transductivee Vladim
mir Vapnik 1998
2.7.
SVM
Ms .
regularrization Tikhhonov
2.8.
mislabeled
hyperplanee """"
M
Margin
hyperplane
m
misclassificat
tion
min
n ,
1
| |
2
)
1
(2)
min , , max ,
1
|| ||
2
C
Cortes Vapnik 2008 Kanellakis ACM
2.9.
hyperplane Vladimir Vapnik 1963
1992, Boser , Guyon Vapnik
( Aizerman et al) hyperplanes
margin - hyperplane
.
hyperplane
Gaussian , Hilbert
regularized
() :
(inhomogeneous) :
exp
1
|
> 0
Gaussian :
exp
tanh KX X
() > 0 c <0
Dot w
(ELM
Map2D)
22D PCA
P (PCA2DD) ""
ELMap2DD
PCA2D
PCCA K -
http://en.wikipedia.org/wiki/Principaal_component_analysis
2.10.
SVM
S
,
C, ) (
(
;
)
2.10.1
2.125 Vapnik
V
y = 1
y = -- 1
http://www..springerlink..com/index/kkm7krm468022r2114.pdf
(x) ()
2.10.2
hyperplane
2.12.1 SVM6
, Lagrangian
Lagrangian .
w :
.
QP
constrains (8) hyperplane
2.10.3 SVM
SVM reformulating
http://www.springerlink.com/index/km7krm46802r2114.pdf
y i -- (w T (x i) + B)
LS - SVM
2.12.2 LS - SVM7
http://www.springerlink.com/index/km7krm46802r2114.pdf
hypeerparamters regulaarization
LS - SVM
optimality
w e
K j = (x i T) (x j) = (x i, x j)
2.11.
Multiclass SVM
Multiclass SVM
multiclass
(i) (ONE - - all) (ii) (ONE - one) -
- (
) - one
- Max
3.1.
(Artificial neural network)
(neural network neural net)
(connectionist)
(Pattern
Recognition) ( Knowledge deduction)
(bioelectric network)
(synapses)
"" (Dendrite) input
"" (Axon) output
3.1.1.
3.1.2.
Neurons input output
input weight input neuron threshold
input output neurons
neuron
3.1.3.
Neural networks input network input weight
input neuron threshold
threshold neuron output output
input neuron network threshold output
weight threshold
pattern "back propagation"
feed-forward neural networks back-propagation
(network weight)
(output)
input 9
3x3 3x3
1,2,3,4,6,7,8,9
neural network
( inputs outputs
inputs outputs )
()
(Model Predictive Control)
HVAC
4.1.
Support Vector Machines
Support Vector Machines 100%
Support Vector Machines
Support Vector Machines
Support Vector Machines
http://www.support-vector.net/icml-tutorial.pdf
http://www.cs.columbia.edu/~kathy/cs4701/.../jason_svm_tutorial.pdf
http://www.svms.org/tutorials
http://www.autonlab.org/tutorials/svm.html
http://www.dtreg.com/svm.htm
http://www.cs.brown.edu/courses/archive/2006-2007/.../Burges98.pdf
http://www.support-vector-machines.org/SVM_review.html
http://www.springerlink.com/index/km7krm46802r2114.pdf