1 / 36

Recent Development on Elimination Ordering Group 1

Recent Development on Elimination Ordering Group 1. Recent Development on Elimination Ordering. By Florence Lo Andrew Yan. Table of Contents. Problem Statement Background Motivations Approaches to Elimination Orderings An approximation algorithm for triangulation Q & A.

kelli
Download Presentation

Recent Development on Elimination Ordering Group 1

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Recent Development on Elimination OrderingGroup 1

  2. Recent Development on Elimination Ordering By Florence Lo Andrew Yan

  3. Table of Contents • Problem Statement • Background • Motivations • Approaches to Elimination Orderings • An approximation algorithm for triangulation • Q & A

  4. Problem Statement Minimizing Total Cost in Vertices Elimination

  5. Background BN specifies a complete joint probability distribution (JPD) over all variables. Given the JPD, one can answer all possible inference queries by summing out over irrelevant variables. Assumed all variables have 2 states and there are n variables => O(2n)

  6. Background (cont) To make inference more efficient, 2 exact inference algorithms are used Variable Elimination Algorithm (VE) Clique Tree Propagation (CTP)

  7. Variable Elimination (VE) Use the factored representation of the JPD to do marginalization efficiently. The key idea is to “push sum in” as far as possible when summing out irrelevant terms Complexity of VE is based on the cost of variables elimination (i.e. elimination orderings)

  8. Clique Tree Propagation (CTP) • Inference in BN is formulated as message passing in a junction tree • First step in CTP is triangulated graph, one way is to eliminate vertices one by one to add the extra edges (i.e. elimination orderings)

  9. Discussion • Different elimination ordering leads to different cost • NP-hard

  10. 3 Approaches to Establish Elimination Orderings • Elimination Ordering Heuristics • Triangulation • Simulated Annealing

  11. Elimination Ordering Heuristics • Maximum Cardinality Search (Tarjan et. al. 1985) • Minimum Deficiency Heuristics (Bertele et. al. 1972) • Minimum Degree Heuristics (Rose 1972)

  12. Discussions • Easy to implement • Concepts are easy • Linear time complexity • Good approximation to the optimal solution

  13. Triangulation • Objectives • Size of the maximal clique is minimum • Minimal triangulation

  14. Minimal Triangulation Computing a minimal triangulation consists in embedding a given graph with a triangulated graph by adding a set of edges (called a fill). If the set of edges added is inclusion-minimal, the fill is said to be minimal, and the corresponding triangulated graph is called a minimal triangulation

  15. Discussion • Minimal triangulation generally improves the total cost • May stuck in the local minima of the cost function • NP-hard

  16. Simulated Annealing • Stochastic optimization algorithm to find global minimum cost configuration of NP-complete combinatorial problems with cost functions having many local minima • A combination of deterministic descent search and a Monte Carlo method • Accept cost function increase with positive probability dependent on the state of the search process

  17. Pseudo-code Select an initial solution s0 Select an initial temperature t0 > 0 Select a temperature reduction function α Repeat repeat randomly select s from N(s0) δ = f(s) – f(s0) If δ < 0 then s0 = s else generate random x uniformly in the range (0, 1) if x < exp(- δ/t) then s0 = s until iteration_count = nrep Set t = α(t) Until stopping condition = true S0 is the approximation to the optimal solution

  18. 2 Categories of Decisions to make • Parameters of the annealing algorithm - t0, cooling schedule governed by nrep and α, stopping conditions 2. Problem specific and involve the choice of the space of feasible solutions, the form of the cost function, neighborhood structure employed

  19. Discussion • Time consuming • Extensive experiments on the control parameters • Enhancement and modifications to speed up computation time

  20. An approximation algorithm for triangulation • By Ann Becker and Dan Geiger in 1996 • Same state space size • Optimality criterion • Cliquewidth, k • 2+1 – approximation •  = approximation ratio for 3-way vertex cut • O(2(2+1)kn•poly(n)) • poly(n) = linear programming • Divide and conquer

  21. The algorithm Triangulate(G,W,k) If |V| < (2 +1)K then make a clique out of G Else Find a W-decomposition (X,A,B,C) of G wrt (k,) If not found Return “cliquewidth > k” WAWA, WBWB, WCWC Call Triangulate( G[AX], WA,k) Call Triangulate( G[BX], WB,k) Call Triangulate( G[CX], WC,k) Make a clique of G[W X]

  22. Trial and error Try for k = 1, 2, 3… until success!!!

  23. Example (k=3,=1)

  24. Improvement • Processing the input • Simplicial vertex • Removed repeatedly • Improves the running time

  25. Improvement • Termination condition • |V| < (2+1)k  from a clique • Junction tree instead • W as a clique • Approximation may improved

  26. Improvement • Post-processing the output • May not minimal triangulated • Remove some added edges, still triangulated • Kjaerulff’s algorithm

  27. In practice • Time complexity O(2(2+1)kn•poly(n)) • |W| < k in most case • 24.66k 22k • W consists of two subsets • No 3-way vertex cut • Step further for k Find a W-decomposition (X,A,B,C) of G wrt (k,) If not found • Return “cliquewidth > k”

  28. A tighter bound • l be the size of the largest clique in the output • Test against (2+1)k • The ratio can be smaller than 2+1 • Optimal cliquewidth can be larger than k • l/k instead of 2+1 • Instance-specific posteriori bound

  29. The weighted problem • Different state spaces of each vertex • w(v) = log2(state space size) • w(clique) = sum of w(v) in the clique • Weighted W-decomposition • When terminate in recursion, run greedy algorithm (minimum weight heuristics) • (2+1)m

  30. Results • Medianus I (43 vertices, 110 edges) • Compare with the enhanced minimum weight heuristics • Better when state space increase • l/k = 10/6 (not 3.66) • Run time: one or two minutes

  31. Discussion • O(24.66kn•poly(n)) • Polynomial for k = O(log n) • Exponential time for an arbitrary inference

  32. What is cliquewidth? • Undirected graph • The size of the largest clique in the junction free of the graph in which the size of the largest clique is minimized back

  33. What is 3-way vertex cut? • A weighted undirected graph • 3 vertices • A set of vertices of minimum weight whose removal  three vertices disconnected • 4/3 – approximation • 2 – approximation back

  34. What is W-decomposition? • An integer k  1, a real number   1 • A graph G = ( V, E ), |V|  ( 2 + 1 ) k • W  V • A decomposition ( X, A, B, C ) wrt ( k,  ) • |W| < ( + 1 )k • |X| < k • |(WA)X| < ( + 1 )k • |(WB)X| < ( + 1 )k • |(WC)X| < ( + 1 )k back

  35. What is decomposition? • A graph G = ( V, E ) • A partition ( X, A, B, C ) • A, B   • No edges between A, B, C back

More Related