Title: An Analysis of Convex Relaxations (PART I) Minimizing Higher Order Energy Functions (PART 2)
1An Analysis of Convex Relaxations (PART
I)Minimizing Higher Order Energy Functions
(PART 2)
Work in collaboration with Pushmeet Kohli,
Srikumar Ramalingam M. Pawan Kumar, Lubor
Ladický, Vladimir Kolmogorov
2An Analysis of Convex Relaxations
for MAP Estimation
- M. Pawan Kumar
- Vladimir Kolmogorov
- Philip Torr
3Aim
- To analyze convex relaxations for MAP estimation
6
3
0
1
2
4
0
Label 1
1
2
4
1
1
3
Label 0
1
0
5
3
7
0
2
V2
V3
V4
V1
Random Variables V V1, ... ,V4
Label Set L 0, 1
Labelling m 1, 0, 0, 1
4Aim
- To analyze convex relaxations for MAP estimation
6
3
0
1
2
4
0
Label 1
1
2
4
1
1
3
Label 0
1
0
5
3
7
0
2
V2
V3
V4
V1
Cost(m) 2 1 2 1 3 1 3 13
Minimum Cost Labelling? NP-hard problem
Which approximate algorithm is the best?
5Aim
- To analyze convex relaxations for MAP estimation
6
3
0
1
2
4
0
Label 1
1
2
4
1
1
3
Label 0
1
0
5
3
7
0
2
V2
V3
V4
V1
Objectives
- Compare existing convex relaxations LP, QP and
SOCP - Develop new relaxations based on the comparison
6Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
7Integer Programming Formulation
2
4
0
Unary Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
2 4
2
Unary Cost Vector u 5
8Integer Programming Formulation
2
4
0
Unary Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
2 4 T
2
Unary Cost Vector u 5
Label vector x -1
1
1 -1 T
Recall that the aim is to find the optimal x
9Integer Programming Formulation
2
4
0
Unary Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
2 4 T
2
Unary Cost Vector u 5
Label vector x -1
1
1 -1 T
1
Sum of Unary Costs
?i ui (1 xi)
2
10Integer Programming Formulation
2
4
0
Pairwise Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
0
3
0
11Integer Programming Formulation
2
4
0
Pairwise Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
Sum of Pairwise Costs
1
?ij Pij (1 xi)(1xj)
0
3
0
4
12Integer Programming Formulation
2
4
0
Pairwise Cost
Label 1
1
3
Label 0
5
0
2
V2
V1
Labelling m 1 , 0
Sum of Pairwise Costs
1
?ij Pij (1 xi xj xixj)
0
3
0
4
X x xT
Xij xi xj
13Integer Programming Formulation
Constraints
xi ?-1,1
X x xT
14Integer Programming Formulation
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
Convex
xi ?-1,1
X x xT
15Outline
- Integer Programming Formulation
- Existing Relaxations
- Linear Programming (LP-S)
- Semidefinite Programming (SDP-L)
- Second Order Cone Programming (SOCP-MS)
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
16LP-S
Schlesinger, 1976
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
Relax Non-Convex Constraint
xi ?-1,1
X x xT
17LP-S
Schlesinger, 1976
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
Relax Non-Convex Constraint
X x xT
18LP-S
Schlesinger, 1976
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
LP-S
Xij ?-1,1
1 xi xj Xij 0
19Outline
- Integer Programming Formulation
- Existing Relaxations
- Linear Programming (LP-S)
- Semidefinite Programming (SDP-L)
- Second Order Cone Programming (SOCP-MS)
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
- Experiments
20SDP-L
Lasserre, 2000
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
Relax Non-Convex Constraint
xi ?-1,1
X x xT
21SDP-L
Lasserre, 2000
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
Relax Non-Convex Constraint
X x xT
22SDP-L
Lasserre, 2000
Retain Convex Part
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
Xii 1
Inefficient
Accurate
23Outline
- Integer Programming Formulation
- Existing Relaxations
- Linear Programming (LP-S)
- Semidefinite Programming (SDP-L)
- Second Order Cone Programming (SOCP-MS)
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
24SOCP Relaxation
Derive SOCP relaxation from the SDP relaxation
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
Xii 1
Further Relaxation
252-D Example
X11
X12
X
X21
X22
262-D Example
(X - xxT)
1 - x12
X12-x1x2
X12-x1x2
1 - x22
(x1 x2)2 ? 2 2X12
SOC of the form v 2 ? st
272-D Example
(X - xxT)
1 - x12
X12-x1x2
X12-x1x2
1 - x22
(x1 - x2)2 ? 2 - 2X12
SOC of the form v 2 ? st
28SOCP Relaxation
C1 . ? 0
Kim and Kojima, 2000
UTx 2 ? X . C1
(X - xxT)
SOC of the form v 2 ? st
Continue for C2, C3, , Cn
29SOCP Relaxation
How many constraints for SOCP SDP ?
Infinite.
Specify constraints similar to the 2-D example
Xij
xi
xj
(xi xj)2 ? 2 2Xij
(xi xj)2 ? 2 - 2Xij
30SOCP-MS
Muramatsu and Suzuki, 2003
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
Xii 1
31SOCP-MS
Muramatsu and Suzuki, 2003
1
1
? Pij (1 xi xj Xij)
x argmin
? ui (1 xi)
4
2
xi ?-1,1
(xi xj)2 ? 2 2Xij
(xi - xj)2 ? 2 - 2Xij
32Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
33Dominating Relaxation
A
B
For all MAP Estimation problem (u, P)
A dominates B
Dominating relaxations are better
34Equivalent Relaxations
Strictly Dominating Relaxation
35SOCP-MS
min ?ij Pij Xij
(xi xj)2 ? 2 2Xij
(xi - xj)2 ? 2 - 2Xij
SOCP-MS is a QP
Same as QP by Ravikumar and Lafferty, 2006
SOCP-MS QP-RL
36LP-S vs. SOCP-MS
Differ in the way they relax X xxT
37LP-S vs. SOCP-MS
- LP-S strictly dominates SOCP-MS
- LP-S strictly dominates QP-RL
- Where have we gone wrong?
38Recap of SOCP-MS
Can we use a different subgraph ??
Can we use different C matrices ??
39Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- SOCP Relaxations on Trees
- SOCP Relaxations on Cycles
- Two New SOCP Relaxations
40SOCP Relaxations on Trees
Choose any arbitrary tree
41SOCP Relaxations on Trees
Choose any arbitrary C 0
Repeat over trees to get relaxation SOCP-T
LP-S strictly dominates SOCP-T
LP-S strictly dominates QP-T
42Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- SOCP Relaxations on Trees
- SOCP Relaxations on Cycles
- Two New SOCP Relaxations
43SOCP Relaxations on Cycles
Choose an arbitrary even cycle
Pij 0
OR
Pij 0
44SOCP Relaxations on Cycles
Choose any arbitrary C 0
Repeat over even cycles to get relaxation SOCP-E
LP-S strictly dominates SOCP-E
LP-S strictly dominates QP-E
45SOCP Relaxations on Cycles
- True for odd cycles with Pij 0
- True for odd cycles with Pij 0 for only one
edge
- True for odd cycles with Pij 0 for only one
edge
- True for all combinations of above cases
46Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
- The SOCP-C Relaxation
- The SOCP-Q Relaxation
47The SOCP-C Relaxation
Include all LP-S constraints
True SOCP
a
b
Cycle of size 4
c
d
Define SOCP Constraint using appropriate C
SOCP-C strictly dominates LP-S
Which SOCP is strictly dominated by cycle
inequalities?
Open Question !!!
48Outline
- Integer Programming Formulation
- Existing Relaxations
- Comparison
- Generalization of Results
- Two New SOCP Relaxations
- The SOCP-C Relaxation
- The SOCP-Q Relaxation
49The SOCP-Q Relaxation
Include all cycle inequalities
True SOCP
a
b
Clique of size n
c
d
Define an SOCP Constraint using C 1
SOCP-Q strictly dominates LP-S
SOCP-Q strictly dominates cycle inequalities
50Conclusions
- Large class of SOCP/QP dominated by LP-S
- New SOCP relaxations dominate LP-S
- Preliminary experiments conform with analysis
51Future Work
- Comparison with cycle inequalities
- Determine best SOC constraints
- Develop efficient algorithms for new relaxations
52Part 2
534-Neighbourhood MRF
Test SOCP-C
50 binary MRFs of size 30x30
u N (0,1)
P N (0,s2)
544-Neighbourhood MRF
s 2.5
558-Neighbourhood MRF
Test SOCP-Q
50 binary MRFs of size 30x30
u N (0,1)
P N (0,s2)
568-Neighbourhood MRF
s 1.125
57Equivalent Relaxations
A
B
For all MAP Estimation problem (u, P)
A dominates B
B dominates A
58Strictly Dominating Relaxation
gt
A
B
For at least one MAP Estimation problem (u, P)
A dominates B
B does not dominate A