approximate inference on planar graphs using loop
play

Approximate inference on planar graphs using Loop Calculus and - PowerPoint PPT Presentation

Approximate inference on planar graphs using Loop Calculus and Belief Propagation Vicen Gmez 1 Hilbert J.Kappen 1 M. Chertkov 2 1 Department of Biophysics Radboud University, Nijmegen, The Netherlands 2 Theoretical Division and Center of


  1. Approximate inference on planar graphs using Loop Calculus and Belief Propagation Vicenç Gómez 1 Hilbert J.Kappen 1 M. Chertkov 2 1 Department of Biophysics Radboud University, Nijmegen, The Netherlands 2 Theoretical Division and Center of Nonlinear Studies Los Alamos National Laboratory, Los Alamos Physics of Algorithms 2009 Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 1 / 15

  2. Outline Motivation 1 Algorithm 2 Experiments 3 Setup Full series Grids Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 2 / 15

  3. Motivation Exact inference on the Ising model defined on a planar graph is easy for zero external fields (Kasteleyn, Fisher and others, 1960s) : p ( x ) = 1 P ( i , j ) w ij x i x j Z e Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 3 / 15

  4. Motivation Exact inference on the Ising model defined on a planar graph is easy for zero external fields (Kasteleyn, Fisher and others, 1960s) : p ( x ) = 1 P ( i , j ) w ij x i x j + P i θ i x i Z e Otherwise is intractable , # P (Barahona, 82) . Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 3 / 15

  5. Motivation Exact inference on the Ising model defined on a planar graph is easy for zero external fields (Kasteleyn, Fisher and others, 1960s) : p ( x ) = 1 P ( i , j ) w ij x i x j + P i θ i x i Z e Otherwise is intractable , # P (Barahona, 82) . Recently, the Fisher & Kasteleny method has been introduced in the Machine Learning community: ◮ "Approximate inference using planar graph decomposition", Globerson A & Jaakkola T (NIPS 07) ◮ "Efficient Exact Inference in Planar Ising Models", Schraudolph N & Kamenetsky D, (NIPS 08) Both perform exact inference on an easy planar model We directly approximate Z on difficult planar graphs. Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 3 / 15

  6. Motivation Loop Calculus and Belief Propagation (BP) Exact Z of a general binary graphical model can be expressed as a finite sum of terms that can be evaluated once the BP solution is known. (Chertkov & Chernyak, 06a) � � Z = Z BP · z , � z = 1 + r C C ∈C Each term corresponds to a generalized loop (subgraph with no degree 1 vertices) Summing all terms is intractable... but truncation can provide improvements on BP (Gómez et al 07, Chertkov & Chernyak, 06b) . Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 4 / 15

  7. Motivation 2-regular loops non 2-regular loops 2-regular loop : A loop where all nodes have degree two. 2-regular part. function Z ∅ : Approximation including all 2-regular loops only. Z ∅ = Z BP · z ∅ , z ∅ = 1 + � a C | = 2 , ∀ a ∈ C r C . C ∈C s . t . | ¯ Triplet : A node with degree 3 in the Forney graph. Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 5 / 15

  8. Motivation "Belief Propagation and Loop Series for planar graphs", (Chertkov et al, 08) The 2-regular partition function Z ∅ can be expressed as a sum of weighted perfect matchings. For planar graphs , Z ∅ can be computed in polynomial time . The full loop series can be expressed as a sum over so-called Pfaffian terms , and each term may sum many loops. Contribution We develop an algorithm to compute the full Pfaffian series. Empirical analysis: ◮ Compare Loop and Pfaffian series ◮ Analyze the accuracy of the Z ∅ approximation. Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 6 / 15

  9. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  10. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  11. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  12. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  13. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  14. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . For every face, the number The 2-regular partition function is: 7 of clockwise oriented edges is odd. Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  15. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B .  The 2-regular partition function is: + µ ij if ( i , j ) ∈ E G ′ 7  ext  ˆ A ij = − µ ij if ( j , i ) ∈ E G ′ . Z ∅ = Z BP · z ∅ , ext  0 otherwise  � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  16. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B .  The 2-regular partition function is: + 1 if ( i , j ) ∈ E G ′ 7  ext  ˆ B ij = − 1 if ( j , i ) ∈ E G ′ . Z ∅ = Z BP · z ∅ , ext  0 otherwise  � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  17. Loop series for planar graphs Computing 2-regular partition function Forney graph G with binary variables and nodes with degree at most 3 . Find stationary point of BP . 1 Obtain 2-core. 2 Construct planar embedding. 3 Obtain extended graph G ext . 4 Obtain Pfaffian orientation for the 5 edges of the extended graph → G ′ ext . Construct skew-symmetric matrices ˆ 6 A and ˆ B . The 2-regular partition function is: 7 Z ∅ = Z BP · z ∅ , � ˆ · Pfaffian (ˆ � �� z ∅ = sign Pfaffian A ) . B Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 7 / 15

  18. Loop series for planar graphs Computing the full Pfaffian series Computing full loop series Denote T the set of all possible triplets in G . Consider a subset Ψ ∈ T with an even number of triplets. Loops in G including the triplets in Ψ correspond to perfect matchings on another extended graph G ext Ψ . Exact Z can be written as a sum of Pfaffian terms: � � ˆ � � � ˆ � �� z = Z Ψ , Z Ψ = z Ψ µ a ;¯ a , z Ψ = sign Pf · Pf . B Ψ A Ψ a ∈ Ψ Ψ The 2-regular partition function Z ∅ correponds to Φ = ∅ . Gómez V, Kappen HJ, Chertkov M () 1 Sept 2009 8 / 15

Recommend


More recommend