Karush-Kuhn-Tucker (KKT) Conditions: Lecture 11: Convex Optimization

STAT 598Y Statistical Learning Theory
Instructor: Jian Zhang
Lecture 11: Convex Optimization

In this lecture we first give some background about convex optimization including the KKT condition and
duality. We then derive the SVM dual optimization problem.
Consider the constrained minimization problem of the form
minimize
f (x)
subject to:
gi (x) = 0 i = 1, . . . , m n
hj (x) 0 j = 1, . . . , p.
(1)
gi s are equality constraints and hj s are inequality constraints and usually they are assumed to be within
the class C 2 . A point that satisfies all constraints is said to be a feasible point. An inequality constraint is
said to be active at a feasible point x if hj (x) = 0 and inactive if hj (x) < 0. Equality constraints are always
active at any feasible point. To simplify notation we write h = [h1 , . . . , hp ] and g = [g1 , . . . , gm ], and the
constraints now become g(x) = 0 and h(x) 0.
Karush-Kuhn-Tucker (KKT) Conditions

KKT conditions (a.k.a. Kuhn-Tucker conditions) are necessary conditions for the local minimum solutions
of problem (1). Let x be a local minimum point for Problem (1) and suppose x is a regular point for the
constraints. Then there is a vector Rm and a vector Rp with 0 such that
f (x ) + T h(x ) + T g(x )
g(x )
j hj (x )
= 0
= 0
(2)
(3)
= 0 (j = 1, . . . , p)
(4)
Convince yourself why the above conditions hold geometrically. It is convenient to introduce the Lagrangian
associated with the problem as
L(x, , ) = f (x) + T h(x) + T g(x)
where Rm , Rp and 0 are Lagrange multipliers. Note that equation (2), (3) and (4) together
give a total of n + m + p equations in the n + m + p variables x , and .
From now on we assume that we only have inequality constrains for simplicity. The case with equality
constraints can be done in a similar way, except that does not have the nonnegative constraint as . So in
our case we have the following optimization problem:
min f (x) s.t. h(x) 0.
Weak Duality and Strong Duality

Consider the Lagrangian L(x, ) for the above optimization problem. Then we have the following two types
of dualities:
weak duality: we have obviously for any 0 that
inf L(x, ) inf sup L(x, )
x
x 0
and thus
sup inf L(x, ) inf sup L(x, ).
0 x
x 0
Such a relation always holds.

strong duality: suppose in addition there exist x and 0 such that
L(x , ) L(x , ) L(x, )
for all feasible x and 0. Then we have
inf sup L(x, )
sup L(x , )
x 0
L(x , )
inf L(x, )
sup inf L(x, ).
0 x
Thus we have
inf sup L(x, ) = sup inf L(x, ).
x 0
0 x
The point (x , ) is called the saddle point. One example is the function L(x, ) = x2 2 , with
saddle point (0, 0) as shown in Figure 1.
Weak duality always holds, and strong duality holds if f and hj s are convex and there exists at least one
feasible point which is an interior point. The Lagrange dual function D() is defined as
$
D() := inf L(x, ) = inf f (x) +
j hj (x)
x
x
j=1
and we define the dual optimization problem as:
max D() s.t. 0.

Note that (1) D() is a concave function; (2) for any feasible and x we have D() f (x). In fact if we
define p to be the minimum solution of the primal optimization problem (primal solution), and d to the
maximum of the dual problem d = sup0 D() (dual solution). Then the weak duality says d p . The
quantity p d is known as the duality gap, which can be a useful criteria for convergence.
Now we illustrate this duality relationship with a simple example where we only have one inequality constraint:
min f (x) s.t. h(x) 0.
Define (z) = inf{f (x) : h(x) z} for z R. Then it is easy to observe that (z) is monotone on each
coordinate of z. The duality can be illustrated by the fact that the primal solution p is the intercept of
(z) with the vertical axis z = 0, and it is an upperbound of the maximum intercept with the vertical axis
of all hyperplanes that lie below (.). Such hyperplanes have the form l (z) = T z + inf x {f (x) + h(x)}
with 0. An example is shown in Figure 2.
SVM Dual Problem

The SVM primal problem can be written as
minw,b,
T
1
n
(n
i=1 i
+ &w&2
s.t. yi (w xi + b) 1 i ; i 0 i
Saddle Point
Duality
8
7
6
(z)
L(x, )
2
4
0
3
2
2
p*
4
2
1
0
2
0
1
1.5
0.5
0.5
1.5
Figure 1: Left: Saddle point (0, 0) of L(x, ) = x2 2 ; Right: Geometric interpretation of duality.
Now the Lagrangian can be written as
L(w, b, , , ) =
n
n
n
$
$
1$
i i
i + wT w +
i (1 i yi wT xi yi b)
n i=1
i=1
i=1
where the Lagrange multiplers 0 and 0. We want to remove the primal variables w, b, by
maximization, i.e. set the following derivatives to zero:
n
L
=0
w
w=
L
=0
b
n
$
L
=0
1 $
i yi xi
2 i=1
i yi = 0
i=1
i + i =
1
.
n
Plugging in and we obtain the dual:

D(, ) =
n
$
i=1
1 $
i j yi yj xTi xj .
4 i,j
Since we have i 0 and i 0 and i + i = 1/n, thus we have 0 i 1/n. So the dual optimization
problem becomes
(n
1 (
T
max
i=1 i 4
i,j i j yi yj xi xj
s.t.
0 i 1/n.
which is a quadratic programming problem. Note that due to the constraints, the dual solution is in general
sparse, i.e. we have many #i s equal to 0. We have the following observations:
1. If i > 0: we have yi (wT xi + b) = 1 i 1. So the example is either at or on the wrong side of the
margin. Such examples for i > 0 are called support vectors.
2. If i = 0: we have i = 1/n and thus i = 0. So yi (wT xi + b) 1. Such examples are on the correct
side of the margin.
3. If yi (wT xi + b) < 1: we have > 0 and thus i = 0 and i = 1/n. So if an example causes margin
error then its dual variable i will take at the right boundary 1/n.
4. It is possible that for examples which are on the correct side of the margin, their i s are nonzero.
5. In the objective xi s appear always in the form of inner product xTi xj . So if we first map xi into a
feature vector (xi ), then we could replace xTi xj by )(xi ), (xj )*. This leads to the introduction of
reproducing kernel Hilbert space in SVM.

Karush-Kuhn-Tucker (KKT) Conditions: Lecture 11: Convex Optimization

Uploaded by

Copyright:

Available Formats

Karush-Kuhn-Tucker (KKT) Conditions: Lecture 11: Convex Optimization

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Karush-Kuhn-Tucker (KKT) Conditions: Lecture 11: Convex Optimization

Uploaded by

Copyright:

Available Formats

STAT 598Y Statistical Learning Theory

Instructor: Jian Zhang

Lecture 11: Convex Optimization

Karush-Kuhn-Tucker (KKT) Conditions

Weak Duality and Strong Duality

Such a relation always holds.

sup inf L(x, ).

and we define the dual optimization problem as:

max D() s.t. 0.

SVM Dual Problem

Plugging in and we obtain the dual:

You might also like