1. Intro to Junction Tree propagation and adaptations for a Distributed Environment Thor Whalen Metron, Inc.

2. a b c d 11 22 33 44 conflict 55 66 77 88 This naive approach of updating the network inherits oscillation problems!

3. Idea behind the Junction Tree Algorithm a b c d a bc d clever algorithm

4. a bc de f g h Secondary Structure/ Junction Tree multi-dim. random variables joint probabilities (potentials) Bayesian Network one-dim. random variables conditional probabilities abd ade ace ceg egh def ad aece deeg

5. Write query in the form Iteratively –Move all irrelevant terms outside of innermost sum –Perform innermost sum, getting a new term –Insert the new term into the product Variable Elimination (General Idea)

6. Eaxmple of Variable Elimination The “Asia” network: Visit to Asia Smoking Lung Cancer Tuberculosis Abnormality in Chest Bronchitis X-Ray Dyspnea

7. V S L T A B XD We are interested in P(d) - Need to eliminate: v,s,x,t,l,a,b Initial factors: Brute force:

8. V S L T A B XD Eliminate variables in order: Initial factors: [ Note: f v (t) = P(t) In general, result of elimination is not necessarily a probability term ]

9. Eliminate variables in order: Initial factors: V S L T A B XD [ Note: result of elimination may be a function of several variables ]

10. Eliminate variables in order: Initial factors: V S L T A B XD [ Note: f x (a) = 1 for all values of a ]

11. Eliminate variables in order: Initial factors: V S L T A B XD

12. Eliminate variables in order: Initial factors: V S L T A B XD

13. Eliminate variables in order: Initial factors: V S L T A B XD

14. Eliminate variables in order: Initial factors: V S L T A B XD

15. Intermediate factors In our previous example: With a different ordering: V S L T A B XD Complexity is exponential in the size of these factors!

16. Notes about variable elimination Actual computation is done in the elimination steps Computation depends on the order of elimination For each query we need to compute everything again! –Many redundant calculations

17. Junction Trees The junction tree algorithm “generalizes” Variable Elimination to avoid redundant calculations The JT algorithm compiles a class of elimination orders into a data structure that supports the computation of all possible queries.

18. Building a Junction Tree DAG Moral GraphTriangulated GraphJunction TreeIdentifying Cliques

19. Step 1: Moralization a bc de f g h a bc de f g h a bc de f g h 1. For all w  V: For all u,v  pa(w) add an edge e=u-v. 2. Undirect all edges. GMGM G = ( V, E )

20. Step 2: Triangulation Add edges to G M such that there is no cycle with length  4 that does not contain a chord. NO YES a bc de f g h a bc de f g h GMGM GTGT

21. Step 2: Triangulation (cont.) Each elimination ordering triangulates the graph, not necessarily in the same way: A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G A H B D F C E G

22. Step 2: Triangulation (cont.) Intuitively, triangulations with as few fill-ins as possible are preferred –Leaves us with small cliques (small probability tables) A common heuristic: Repeat until no nodes remain: –Find the node whose elimination would require the least number of fill-ins (may be zero). –Eliminate that node, and note the need for a fill-in edge between any two non-adjacent neighbors. Add the fill-in edges to the original graph.

23. a bc de f g h a bc d e f g h a bc d e f g a bc d e f a bc d e a b d e aa e a d e vertexinduced added removedclique edges 1hegh- 2gceg- 3fdef- 4cacea-e vertexinduced added removed clique edges 5babda-d 6dade- 7eae- 8aa- GTGT GMGM Eliminate the vertex that requires least number of edges to be added.

24. Step 3: Junction Graph A junction graph for an undirected graph G is an undirected, labeled graph. The nodes are the cliques in G. If two cliques intersect, they are joined in the junction graph by an edge labeled with their intersection.

25. a b d a c e de f a de e g h c e g a bc de f g h Bayesian Network G = ( V, E ) a bc de f g h a bc de f g h Moral graph G M Triangulated graph G T abd ade ace ceg egh def adaece de eg seperators Junction graph G J (not complete) e.g. ceg  egh = eg Cliques e e e a e

26. Step 4: Junction Tree A junction tree is a sub-graph of the junction graph that –Is a tree –Contains all the cliques (spanning tree) –Satisfies the running intersection property: for each pair of nodes U, V, all nodes on the path between U and V contain

27. Running intersection? All vertices C and sepsets S along the path between any two vertices A and B contain the intersection A  B. abd ade ace ceg egh def ad aece deeg Ex: A={a,b,d}, B={a,c,e}  A  B={a} C={a,d,e}  {a}, S 1 ={a,d}  {a}, S 2 ={a,e}  {a} A B C S1S1 S2S2

28. A few useful Theorems Theorem: An undirected graph is triangulated if and only if its junction graph has a junction tree Theorem: A sub-tree of the junction graph of a triangulated graph is a junction tree if and only if it is a spanning of maximal weight (sum of number the of variables in the domain of the link).

29. Junction graph G J (not complete) abd ade ace ceg egh def adaece de eg e e e a e abd ade ace ceg egh def ad aece deeg Junction tree G JT There are several methods to find MST. Kruskal’s algorithm: choose successively a link of maximal weight unless it creates a cycle.

30. Colorful example Compute the elimination cliques (the order here is f, d, e, c, b, a). Form the complete junction graph over the maximal elimination cliques and find a maximum- weight spanning tree.

31. Principle of Inference DAG Junction Tree Inconsistent Junction Tree Initialization Consistent Junction Tree Propagation Marginalization

32. a b d a c e de f a de e g h c e g abd ade ace ceg egh def ad aece deeg sepsets In JT cliques becomes vertices G JT Ex: ceg  egh = eg

33. Potentials D EFINITION: A potential  A over a set of variables X A is a function that maps each instantiation of x A into a non-negative real number. Ex: A potential  abc over the set of vertices {a,b,c}. X a has four states, and X b and X c has three states. A joint probability is a special case of a potential where   A (x A )=1.

34. The potentials in the junction tree are not consistent with each other., i.e. if we use marginalization to get the probability distribution for a variable X u we will get different results depending on which clique we use. abd ade ace ceg egh def ad aece deeg P( X a ) =   ade = (0.02, 0.43, 0.31, 0.12) de P( X a ) =   ace = (0.12, 0.33, 0.11, 0.03) ce The potentials might not even sum to one, so they are not joint probability distributions.

35. Message Passing from clique A to clique B 1. Project the potential of A into S AB 2. Absorb the potential of S AB into B Projection Absorption Propagating potentials

36. 1. C OLLECT -E VIDENCE messages1-5 2. D ISTRIBUTE -E VIDENCE messages6-10 Global Propagation 3 2 5 14810 9 7 6 Root abd ade ace ceg egh def ad aece deeg

37. A priori distribution global propagation  potentials are consistent  Marginalizations gives probability distributions for the variables

38. Example: Create Join Tree BC AD (this BN corresponds to an HMM with 2 time steps: Junction Tree: B,C A,B C,D B C

39. Example: Initialization Variable Associated Cluster Potential function AA,B B CB,C DC,D B,C A,B C,D B C BC AD

40. Example: Collect Evidence Choose arbitrary clique, e.g. B,C, where all potential functions will be collected. Call recursively neighboring cliques for messages: 1. Call A,B: –1. Projection onto B: –2. Absorption:

41. Example: Collect Evidence (cont.) 2. Call C,D: –1. Projection: –2. Absorption: B,C A,B C,D B C

42. Example: Distribute Evidence Pass messages recursively to neighboring nodes Pass message from B,C to A,B: –1. Projection: –2. Absorption:

43. Example: Distribute Evidence (cont.) Pass message from X1,X2 to X2,Y2: –1. Projection: –2. Absorption: B,C A,B C,D B C

44. Netica’s Animal Characteristics BN

46. Subnet 1: * JTnode 1: An,En * JTnode 2: An,Sh * JTnode 3: An,Cl Subnet 2: * JTnode 4: Cl,Yo * JTnode 5: Cl,Wa Subnet 3: * JTnode 6: Cl,Bod