1. An Analysis of Convex Relaxations (PART I) Minimizing Higher Order Energy Functions (PART 2) Philip Torr Work in collaboration with: Pushmeet Kohli, Srikumar Ramalingam M. Pawan Kumar, Ľubor Ladický, Vladimir Kolmogorov

2. An Analysis of Convex Relaxations M. Pawan Kumar Vladimir Kolmogorov Philip Torr for MAP Estimation

3. Aim 2 5 4 2 6 3 3 7 0 1 1 0 0 2 3 1 1 41 0 V1V1 V2V2 V3V3 V4V4 Label ‘0’ Label ‘1’ Labelling m = {1, 0, 0, 1} Random Variables V = {V 1,...,V 4 } Label Set L = {0, 1} To analyze convex relaxations for MAP estimation

4. Aim 2 5 4 2 6 3 3 7 0 1 1 0 0 2 3 1 1 41 0 V1V1 V2V2 V3V3 V4V4 Label ‘0’ Label ‘1’ Cost(m) = 2 + 1 + 2 + 1 + 3 + 1 + 3 = 13 Which approximate algorithm is the best? Minimum Cost Labelling? NP-hard problem To analyze convex relaxations for MAP estimation

5. Aim 2 5 4 2 6 3 3 7 0 1 1 0 0 2 3 1 1 41 0 V1V1 V2V2 V3V3 V4V4 Label ‘0’ Label ‘1’ Objectives Compare existing convex relaxations – LP, QP and SOCP Develop new relaxations based on the comparison To analyze convex relaxations for MAP estimation

6. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results Two New SOCP Relaxations

7. Integer Programming Formulation 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Unary Cost Unary Cost Vector u = [ 5 Cost of V 1 = 0 2 Cost of V 1 = 1 ; 2 4 ] Labelling m = {1, 0}

8. 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Unary Cost Unary Cost Vector u = [ 5 2 ; 2 4 ] T Labelling m = {1, 0} Label vector x = [ -1 V 1  0 1 V 1 = 1 ; 1 -1 ] T Recall that the aim is to find the optimal x Integer Programming Formulation

9. 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Unary Cost Unary Cost Vector u = [ 5 2 ; 2 4 ] T Labelling m = {1, 0} Label vector x = [ -11; 1 -1 ] T Sum of Unary Costs = 1 2 ∑ i u i (1 + x i ) Integer Programming Formulation

10. 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Pairwise Cost Labelling m = {1, 0} 0 Cost of V 1 = 0 and V 1 = 0 0 00 0 Cost of V 1 = 0 and V 2 = 0 3 Cost of V 1 = 0 and V 2 = 1 10 00 00 10 30 Pairwise Cost Matrix P Integer Programming Formulation

11. 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Pairwise Cost Labelling m = {1, 0} Pairwise Cost Matrix P 00 00 0 3 10 00 00 10 30 Sum of Pairwise Costs 1 4 ∑ ij P ij (1 + x i )(1+x j ) Integer Programming Formulation

12. 2 5 4 2 0 1 3 0 V1V1 V2V2 Label ‘0’ Label ‘1’ Pairwise Cost Labelling m = {1, 0} Pairwise Cost Matrix P 00 00 0 3 10 00 00 10 30 Sum of Pairwise Costs 1 4 ∑ ij P ij (1 + x i +x j + x i x j ) 1 4 ∑ ij P ij (1 + x i + x j + X ij )= X = x x T X ij = x i x j Integer Programming Formulation

13. Constraints Uniqueness Constraint ∑ x i = 2 - |L| i  V a Integer Constraints x i  {-1,1} X = x x T Integer Programming Formulation

14. x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  {-1,1} X = x x T Convex Non-Convex Integer Programming Formulation

15. Outline 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

16. LP-S x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  {-1,1} X = x x T Retain Convex Part Schlesinger, 1976 Relax Non-Convex Constraint

17. LP-S x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] X = x x T Retain Convex Part Schlesinger, 1976 Relax Non-Convex Constraint

18. LP-S x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a Retain Convex Part Schlesinger, 1976 X ij  [-1,1] 1 + x i + x j + X ij ≥ 0 ∑ X ij = (2 - |L|) x i j  V b LP-S x i  [-1,1]

19. Outline 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

20. SDP-L x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  {-1,1} X = x x T Retain Convex Part Lasserre, 2000 Relax Non-Convex Constraint

21. SDP-L x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] X = x x T Retain Convex Part Relax Non-Convex Constraint Lasserre, 2000

22. SDP-L x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] Retain Convex Part X ii = 1 X - xx T 0 Accurate Inefficient Lasserre, 2000

23. Outline 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

24. SOCP Relaxation x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] X ii = 1 X - xx T 0 Derive SOCP relaxation from the SDP relaxation Further Relaxation

25. 2-D Example X 11 X 12 X 21 X 22 1X 12 1 = X = x1x1x1x1 x1x2x1x2 x2x1x2x1 x2x2x2x2 xx T = x12x12 x1x2x1x2 x1x2x1x2 = x22x22

26. 2-D Example (X - xx T ) 1 - x 1 2 X 12 -x 1 x 2 1 - x 2 2 C 1  0 C 1 0 (x 1 + x 2 ) 2  2 + 2X 12 SOC of the form || v || 2  st  0 11 11

27. 2-D Example (X - xx T ) 1 - x 1 2 X 12 -x 1 x 2 1 - x 2 2 C 2  0 C 2 0 (x 1 - x 2 ) 2  2 - 2X 12 SOC of the form || v || 2  st  0 1 1

28. SOCP Relaxation Consider a matrix C 1 = UU T 0 (X - xx T ) ||U T x || 2  X. C 1 C 1.  0 Continue for C 2, C 3, …, C n SOC of the form || v || 2  st Kim and Kojima, 2000

29. SOCP Relaxation How many constraints for SOCP = SDP ? Infinite. Specify constraints similar to the 2-D example xixi xjxj X ij (x i + x j ) 2  2 + 2X ij (x i + x j ) 2  2 - 2X ij

30. SOCP-MS x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] X ii = 1 X - xx T 0 Muramatsu and Suzuki, 2003

31. SOCP-MS x* = argmin 1 2 ∑ u i (1 + x i ) + 1 4 ∑ P ij (1 + x i + x j + X ij ) ∑ x i = 2 - |L| i  V a x i  [-1,1] (x i + x j ) 2  2 + 2X ij (x i - x j ) 2  2 - 2X ij Specified only when P ij  0 Muramatsu and Suzuki, 2003

32. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results Two New SOCP Relaxations

33. Dominating Relaxation For all MAP Estimation problem (u, P) A dominates B A B ≥ Dominating relaxations are better

34. Equivalent Relaxations A dominates B B dominates A Strictly Dominating Relaxation A dominates B B does not dominate A

35. SOCP-MS (x i + x j ) 2  2 + 2X ij (x i - x j ) 2  2 - 2X ij min  ij P ij X ij P ij ≥ 0 (x i + x j ) 2 2 - 1 X ij = P ij < 0 (x i - x j ) 2 2 1 - X ij = SOCP-MS is a QP Same as QP by Ravikumar and Lafferty, 2006 SOCP-MS ≡ QP-RL

36. LP-S vs. SOCP-MS Differ in the way they relax X = xx T X ij  [-1,1] 1 + x i + x j + X ij ≥ 0 ∑ X ij = (2 - |L|) x i j  V b LP-S (x i + x j ) 2  2 + 2X ij (x i - x j ) 2  2 - 2X ij SOCP-MS F(LP-S) F(SOCP-MS)

37. LP-S vs. SOCP-MS LP-S strictly dominates SOCP-MS LP-S strictly dominates QP-RL Where have we gone wrong? A Quick Recap !

38. Recap of SOCP-MS xixi xjxj X ij Can we use different C matrices ?? Can we use a different subgraph ?? 1 1 C = (x i - x j ) 2  2 - 2X ij 11 11 C = (x i + x j ) 2  2 + 2X ij

39. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results –SOCP Relaxations on Trees –SOCP Relaxations on Cycles Two New SOCP Relaxations

40. SOCP Relaxations on Trees Choose any arbitrary tree

41. SOCP 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

42. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results –SOCP Relaxations on Trees –SOCP Relaxations on Cycles Two New SOCP Relaxations

43. SOCP Relaxations on Cycles Choose an arbitrary even cycle P ij ≥ 0 P ij ≤ 0OR

44. SOCP 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

45. SOCP Relaxations on Cycles True for odd cycles with P ij ≤ 0 True for odd cycles with P ij ≤ 0 for only one edge True for odd cycles with P ij ≥ 0 for only one edge True for all combinations of above cases

46. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results Two New SOCP Relaxations –The SOCP-C Relaxation –The SOCP-Q Relaxation

47. The SOCP-C Relaxation Include all LP-S constraintsTrue SOCP ab cd Cycle of size 4 Define SOCP Constraint using appropriate C SOCP-C strictly dominates LP-S Which SOCP is strictly dominated by cycle inequalities? Open Question !!!

48. Outline Integer Programming Formulation Existing Relaxations Comparison Generalization of Results Two New SOCP Relaxations –The SOCP-C Relaxation –The SOCP-Q Relaxation

49. The SOCP-Q Relaxation Include all cycle inequalitiesTrue SOCP ab cd Clique of size n Define an SOCP Constraint using C = 1 SOCP-Q strictly dominates LP-S SOCP-Q strictly dominates cycle inequalities

50. Conclusions Large class of SOCP/QP dominated by LP-S New SOCP relaxations dominate LP-S Preliminary experiments conform with analysis

51. Future Work Comparison with cycle inequalities Determine best SOC constraints Develop efficient algorithms for new relaxations

52. Part 2

53. 4-Neighbourhood MRF Test SOCP-C 50 binary MRFs of size 30x30 u ≈ N (0,1) P ≈ N (0,σ 2 )

54. 4-Neighbourhood MRF σ = 2.5

55. 8-Neighbourhood MRF Test SOCP-Q 50 binary MRFs of size 30x30 u ≈ N (0,1) P ≈ N (0,σ 2 )

56. 8-Neighbourhood MRF σ = 1.125

57. Equivalent Relaxations A dominates B A B = B dominates A For all MAP Estimation problem (u, P)

58. Strictly Dominating Relaxation A dominates B A B > B does not dominate A For at least one MAP Estimation problem (u, P)