Optimization Theory Primal Optimization Problem subject to: Primal Optimal Value:

Optimization Theory

Primal Optimization Problem

minx

pih

mif

i

i

,2,1,0)(

,2,1,0)(

x

x

)(0 xf

subject to:

*)(* 0 xfp Primal Optimal Value:

Optimization Theory

Convex Optimization Problem

minx

bAx

x

mifi ,2,1,0)(

)(0 xf

subject to:

)(),(),(),( 210 xxxx mffff : convex functions

Optimization Theory

Primal Lagrangian function

minx

pih

mif

i

i

,2,1,0)(

,2,1,0)(

x

x

)(0 xf

subject to:

)()()(

)()()(),,(

0

110

xhβxfαx

xxxβαx

TT

i

p

iii

m

ii

f

hffL

)0( i

Optimization Theory

Kuhn-Tucker Theory

)()()(

)()()(),,(

0

110

xhβxfαx

xxxβαx

TT

i

p

iii

m

ii

f

hffL

0x

L

0β

L

0)(,0

0)(,0 0)(

x

xx

ii

iiii f

ff

0)( xif

0i

KKT Complementarity Condition

Optimization Theory

Dual Lagrangian Function

inf),(x

βα ),,( βαxL

0

xx

LL

Optimization Theory

Dual Optimization problem

minx

pih

mif

i

i

,2,1,0)(

,2,1,0)(

x

xsubject to:

*)(* 0 xfp

)(0 xf

Primal

max,βα

),( βα

subject to: mii ,2,1,0

*)*,(* βαd

Dual

For convex optimization problem: ** dp

Support Vector Machine (SVM)

SVM

• Classification

• Regression

• Linear SVM

• Nonlinear SVM

Support Vector Machine (SVM)

Linear SVM

• Training

• Prediction

Support Vector Machine (SVM)

Linear SVM Training

)},(),,(),,{( 2211 kk yyy xxx Training dataset:

ki ,2,1 Label:}1,1{

Attribute :

i

Ni

y

Rx

bf T xwx)(Optimal Separating Hyperplane:

Support Vector Machine (SVM)

Linear SVM Prediction

},,{ 21 lxxx Testing dataset:

0)(1

0)(1)))(( Label

x

xx

f

ffsign

Support Vector Machine (SVM)

Linear SVM: Separable case

• The optimal hyperplane is obtained by maximizing the margin

• Support vectors

Support Vector Machine (SVM)

Linear SVM: Separable case

Primal Problem

Support Vector Machine (SVM)

Linear SVM: Separable case

Support Vector Machine (SVM)

Linear SVM: Separable case

Support Vector Machine (SVM)

Linear SVM: Separable case

Support Vector Machine (SVM)

Linear SVM: Separable case

Dual Problem

Support Vector Machine (SVM)

Linear SVM: Separable case

Support Vector Machine (SVM)

Linear SVM: Non-separable case

Support Vector Machine (SVM)

Linear SVM: Non-separable case

Support Vector Machine (SVM)

Linear SVM: Non-separable case

Support Vector Machine (SVM)

Linear SVM: Non-separable case (Primal Problem)

Subject to:

Support Vector Machine (SVM)

Linear SVM: Non-separable case (Primal Problem)

Support Vector Machine (SVM)

Linear SVM: Non-Separable case

Support Vector Machine (SVM)

Linear SVM: Non-separable case (Implementation)

Quadratic programming Problem