43
Probabilistic Inference Lecture 4 – Part 2 M. Pawan Kumar [email protected] es available online http://cvc.centrale-ponts.fr/personnel/pa

Probabilistic Inference Lecture 4 – Part 2

  • Upload
    wardah

  • View
    55

  • Download
    0

Embed Size (px)

DESCRIPTION

Probabilistic Inference Lecture 4 – Part 2. M. Pawan Kumar [email protected]. Slides available online http:// cvc.centrale-ponts.fr /personnel/ pawan /. Outline. Integer Programming Formulation LP Relaxation and its Dual Convergent Solution for Dual Properties and Computational Issues. - PowerPoint PPT Presentation

Citation preview

Page 1: Probabilistic Inference Lecture 4 – Part 2

Probabilistic InferenceLecture 4 – Part 2

M. Pawan [email protected]

Slides available online http://cvc.centrale-ponts.fr/personnel/pawan/

Page 2: Probabilistic Inference Lecture 4 – Part 2

• Integer Programming Formulation

• LP Relaxation and its Dual

• Convergent Solution for Dual

• Properties and Computational Issues

Outline

Page 3: Probabilistic Inference Lecture 4 – Part 2

Things to Remember

• Forward-pass computes min-marginals of root

• BP is exact for trees

• Every iteration provides a reparameterization

Page 4: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

Va Vb

Label l0

Label l12

5

4

2

0

1 1

0

2Unary Potentials

a;0 = 5

a;1 = 2

b;0 = 2

b;1 = 4

Labellingf(a) = 1

f(b) = 0

ya;0 = 0 ya;1 = 1

yb;0 = 1 yb;1 = 0

Any f(.) has equivalent boolean variables ya;i

Page 5: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

Va Vb

2

5

4

2

0

1 1

0

2Unary Potentials

a;0 = 5

a;1 = 2

b;0 = 2

b;1 = 4

Labellingf(a) = 1

f(b) = 0

ya;0 = 0 ya;1 = 1

yb;0 = 1 yb;1 = 0

Find the optimal variables ya;i

Label l0

Label l1

Page 6: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

Va Vb

2

5

4

2

0

1 1

0

2Unary Potentials

a;0 = 5

a;1 = 2

b;0 = 2

b;1 = 4

Sum of Unary Potentials

∑a ∑i a;i ya;i

ya;i {0,1}, for all Va, li∑i ya;i = 1, for all Va

Label l0

Label l1

Page 7: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

Va Vb

2

5

4

2

0

1 1

0

2Pairwise Potentials

ab;00 = 0

ab;10 = 1

ab;01 = 1

ab;11 = 0

Sum of Pairwise Potentials

∑(a,b) ∑ik ab;ik ya;iyb;k

ya;i {0,1}

∑i ya;i = 1

Label l0

Label l1

Page 8: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

Va Vb

2

5

4

2

0

1 1

0

2Pairwise Potentials

ab;00 = 0

ab;10 = 1

ab;01 = 1

ab;11 = 0

Sum of Pairwise Potentials

∑(a,b) ∑ik ab;ik yab;ik

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

Label l0

Label l1

Page 9: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

min ∑a ∑i a;i ya;i + ∑(a,b) ∑ik ab;ik yab;ik

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

Page 10: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

min Ty

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

= [ … a;i …. ; … ab;ik ….]y = [ … ya;i …. ; … yab;ik ….]

Page 11: Probabilistic Inference Lecture 4 – Part 2

One variable, two labels

ya;0

ya;1

ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1

y = [ ya;0 ya;1] = [ a;0 a;1]

Page 12: Probabilistic Inference Lecture 4 – Part 2

Two variables, two labels

= [ a;0 a;1 b;0 b;1

ab;00 ab;01 ab;10 ab;11]y = [ ya;0 ya;1 yb;0 yb;1

yab;00 yab;01 yab;10 yab;11]

ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1

yb;0 {0,1} yb;1 {0,1} yb;0 + yb;1 = 1

yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1

yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1

Page 13: Probabilistic Inference Lecture 4 – Part 2

In General

Marginal Polytope

Page 14: Probabilistic Inference Lecture 4 – Part 2

In General

R(|V||L| + |E||L|2)

y {0,1}(|V||L| + |E||L|2)

Number of constraints

|V||L| + |V| + |E||L|2

ya;i {0,1} ∑i ya;i = 1 yab;ik = ya;i yb;k

Page 15: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

min Ty

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

= [ … a;i …. ; … ab;ik ….]y = [ … ya;i …. ; … yab;ik ….]

Page 16: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

min Ty

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

Solve to obtain MAP labelling y*

Page 17: Probabilistic Inference Lecture 4 – Part 2

Integer Programming Formulation

min Ty

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

But we can’t solve it in general

Page 18: Probabilistic Inference Lecture 4 – Part 2

• Integer Programming Formulation

• LP Relaxation and its Dual

• Convergent Solution for Dual

• Properties and Computational Issues

Outline

Page 19: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i {0,1}

∑i ya;i = 1

yab;ik = ya;i yb;k

Two reasons why we can’t solve this

Page 20: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

yab;ik = ya;i yb;k

One reason why we can’t solve this

Page 21: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

∑k yab;ik = ∑kya;i yb;k

One reason why we can’t solve this

Page 22: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

One reason why we can’t solve this

= 1∑k yab;ik = ya;i∑k yb;k

Page 23: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

∑k yab;ik = ya;i

One reason why we can’t solve this

Page 24: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

∑k yab;ik = ya;i

No reason why we can’t solve this *

*memory requirements, time complexity

Page 25: Probabilistic Inference Lecture 4 – Part 2

One variable, two labels

ya;0

ya;1

ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1

y = [ ya;0 ya;1] = [ a;0 a;1]

Page 26: Probabilistic Inference Lecture 4 – Part 2

One variable, two labels

ya;0

ya;1

ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1

y = [ ya;0 ya;1] = [ a;0 a;1]

Page 27: Probabilistic Inference Lecture 4 – Part 2

Two variables, two labels

= [ a;0 a;1 b;0 b;1

ab;00 ab;01 ab;10 ab;11]y = [ ya;0 ya;1 yb;0 yb;1

yab;00 yab;01 yab;10 yab;11]

ya;0 {0,1} ya;1 {0,1} ya;0 + ya;1 = 1

yb;0 {0,1} yb;1 {0,1} yb;0 + yb;1 = 1

yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1

yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1

Page 28: Probabilistic Inference Lecture 4 – Part 2

Two variables, two labels

= [ a;0 a;1 b;0 b;1

ab;00 ab;01 ab;10 ab;11]y = [ ya;0 ya;1 yb;0 yb;1

yab;00 yab;01 yab;10 yab;11]

ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1

yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1

yab;00 = ya;0 yb;0 yab;01 = ya;0 yb;1

yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1

Page 29: Probabilistic Inference Lecture 4 – Part 2

Two variables, two labels

= [ a;0 a;1 b;0 b;1

ab;00 ab;01 ab;10 ab;11]y = [ ya;0 ya;1 yb;0 yb;1

yab;00 yab;01 yab;10 yab;11]

ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1

yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1

yab;00 + yab;01 = ya;0

yab;10 = ya;1 yb;0 yab;11 = ya;1 yb;1

Page 30: Probabilistic Inference Lecture 4 – Part 2

Two variables, two labels

= [ a;0 a;1 b;0 b;1

ab;00 ab;01 ab;10 ab;11]y = [ ya;0 ya;1 yb;0 yb;1

yab;00 yab;01 yab;10 yab;11]

ya;0 [0,1] ya;1 [0,1] ya;0 + ya;1 = 1

yb;0 [0,1] yb;1 [0,1] yb;0 + yb;1 = 1

yab;00 + yab;01 = ya;0

yab;10 + yab;11 = ya;1

Page 31: Probabilistic Inference Lecture 4 – Part 2

In General

Marginal Polytope

LocalPolytope

Page 32: Probabilistic Inference Lecture 4 – Part 2

In General

R(|V||L| + |E||L|2)

y [0,1](|V||L| + |E||L|2)

Number of constraints

|V||L| + |V| + |E||L|

Page 33: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

min Ty

ya;i [0,1]

∑i ya;i = 1

∑k yab;ik = ya;i

No reason why we can’t solve this

Page 34: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

Extensively studied

Optimization

Schlesinger, 1976

Koster, van Hoesel and Kolen, 1998

Theory

Chekuri et al, 2001 Archer et al, 2004

Machine Learning

Wainwright et al., 2001

Page 35: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

Many interesting properties

• Global optimal MAP for trees

Wainwright et al., 2001

But we are interested in NP-hard cases

• Preserves solution for reparameterization

• Global optimal MAP for submodular energy

Chekuri et al., 2001

Page 36: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

• Large class of problems

• Metric Labelling• Semi-metric Labelling

Many interesting properties - Integrality Gap

Manokaran et al., 2008

• Most likely, provides best possible integrality gap

Page 37: Probabilistic Inference Lecture 4 – Part 2

Linear Programming Relaxation

• A computationally useful dual

Many interesting properties - Dual

Optimal value of dual = Optimal value of primal

Page 38: Probabilistic Inference Lecture 4 – Part 2

Dual of the LP RelaxationWainwright et al., 2001

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

min Ty

ya;i [0,1]

∑i ya;i = 1

∑k yab;ik = ya;i

Page 39: Probabilistic Inference Lecture 4 – Part 2

Dual of the LP RelaxationWainwright et al., 2001

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

1

2

3

4 5 6

1

2

3

4 5 6

ii =

i ≥ 0

Page 40: Probabilistic Inference Lecture 4 – Part 2

Dual of the LP RelaxationWainwright et al., 2001

1

2

3

4 5 6

q*(1)

ii =

q*(2)

q*(3)

q*(4) q*(5) q*(6)

i q*(i)

Dual of LP

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

i ≥ 0

max

Page 41: Probabilistic Inference Lecture 4 – Part 2

Dual of the LP RelaxationWainwright et al., 2001

1

2

3

4 5 6

q*(1)

ii

q*(2)

q*(3)

q*(4) q*(5) q*(6)

Dual of LP

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

Va Vb Vc

Vd Ve Vf

Vg Vh Vi

i ≥ 0

i q*(i)max

Page 42: Probabilistic Inference Lecture 4 – Part 2

Dual of the LP RelaxationWainwright et al., 2001

ii

max i q*(i)

I can easily compute q*(i)

I can easily maintain reparam constraint

So can I easily solve the dual?

Page 43: Probabilistic Inference Lecture 4 – Part 2

Continued in Lecture 5 …