Presentation is loading. Please wait.

Presentation is loading. Please wait.

Spectral Graph Theory (Basics)

Similar presentations


Presentation on theme: "Spectral Graph Theory (Basics)"— Presentation transcript:

1 Spectral Graph Theory (Basics)
Charalampos (Babis) Tsourakakis CMU Charalampos E. Tsourakakis

2 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

3 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

4 Matrix Representations of G(V,E)
Faloutsos, Tong Matrix Representations of G(V,E) Associate a matrix to a graph: Adjacency matrix Laplacian Normalized Laplacian Main focus Charalampos E. Tsourakakis

5 Charalampos E. Tsourakakis
Faloutsos, Tong Matrix as an operator The image of the unit circle (sphere) under any mxn matrix is an ellipse (hyperellipse). e.g., Charalampos E. Tsourakakis

6 More Reminders Let M be a symmetric nxn matrix.
Faloutsos, Tong More Reminders Let M be a symmetric nxn matrix. λ eigenvalue x eigenvector Charalampos E. Tsourakakis

7 Charalampos E. Tsourakakis
Faloutsos, Tong More Reminders 1-Dimensional Invariant Subspaces Diagonal: No rotation y x (λ,u) Ax Ay Charalampos E. Tsourakakis

8 Charalampos E. Tsourakakis
Faloutsos, Tong Keep in mind! For the rest of slides we are talking for square nxn matrices and unless noticed symmetric ones, i.e, Charalampos E. Tsourakakis

9 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Cheeger Inequality and Sparsest Cut: Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

10 Charalampos E. Tsourakakis
Faloutsos, Tong Adjacency matrix Undirected 4 1 A= 2 3 Charalampos E. Tsourakakis

11 Charalampos E. Tsourakakis
Faloutsos, Tong Adjacency matrix Undirected Weighted 4 10 1 4 0.3 A= 2 3 2 Charalampos E. Tsourakakis

12 Charalampos E. Tsourakakis
Faloutsos, Tong Adjacency matrix Directed 4 1 Observation If G is undirected, A = AT 2 3 Charalampos E. Tsourakakis

13 Charalampos E. Tsourakakis
Faloutsos, Tong Spectral Theorem Theorem [Spectral Theorem] If M=MT, then where Reminder 2: xi i-th principal axis λi length of i-th principal axis Reminder 1: xi,xj orthogonal λi λj xi xj Charalampos E. Tsourakakis

14 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

15 Faloutsos, Tong Bipartite Graphs Any graph with no cycles of odd length is bipartite e.g., all trees are bipartite K3,3 1 4 2 5 Can we check if a graph is bipartite via its spectrum? Can we get the partition of the vertices in the two sets of nodes? 3 6 Charalampos E. Tsourakakis

16 Charalampos E. Tsourakakis
Faloutsos, Tong Bipartite Graphs Adjacency matrix K3,3 1 4 where 2 5 3 6 Why λ1=-λ2=3? Recall: Ax=λx, (λ,x) eigenvalue-eigenvector Eigenvalues: Λ=[3,-3,0,0,0,0] Charalampos E. Tsourakakis

17 Faloutsos, Tong Bipartite Graphs 1 1 3=3x1 2 1 3 1 1 4 1 4 1 1 1 2 5 5 1 1 1 3 6 6 Repeat same argument for the other nodes Charalampos E. Tsourakakis

18 Bipartite Graphs 1 -1 -3=(-3)x1 -2 -1 -3 -1 1 -1 -1 1 -1 -1 1 4 1 4 2
Faloutsos, Tong Bipartite Graphs 1 -1 -3=(-3)x1 -2 -1 -3 -1 1 4 1 4 1 -1 -1 2 5 5 1 -1 -1 3 6 6 Repeat same argument for the other nodes Charalampos E. Tsourakakis

19 Faloutsos, Tong Bipartite Graphs Observation u2 gives the partition of the nodes in the two sets S, V-S! S V-S Question: Were we just “lucky”? Answer: No Theorem: λ2=-λ1 iff G bipartite. u2 gives the partition. Charalampos E. Tsourakakis

20 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

21 Charalampos E. Tsourakakis
Faloutsos, Tong Walks A walk of length r in a directed graph: where a node can be used more than once. Closed walk when: 4 4 1 1 Closed walk of length 3 Walk of length 2 2-1-4 2 2 3 3 Charalampos E. Tsourakakis

22 Charalampos E. Tsourakakis
Faloutsos, Tong Walks Theorem G(V,E) directed graph, adjacency matrix A. The number of walks from node u to node v in G with length r is (Ar)uv Proof Induction on k. See Doyle-Snell, p.165 (i,j) (i, i1),(i1,j) (i,i1),..,(ir-1,j) Charalampos E. Tsourakakis

23 Charalampos E. Tsourakakis
Faloutsos, Tong Walks 4 1 2 3 4 i=3, j=3 i=2, j=4 4 1 1 2 3 2 3 Charalampos E. Tsourakakis

24 Charalampos E. Tsourakakis
Faloutsos, Tong Walks 4 1 2 3 Always 0, node 4 is a sink 4 1 2 3 Charalampos E. Tsourakakis

25 Charalampos E. Tsourakakis
Faloutsos, Tong Walks Corollary A adjacency matrix of undirected G(V,E) (no self loops), e edges and t triangles. Then the following hold: a) trace(A) = 0 b) trace(A2) = 2e c) trace(A3) = 6t 1 Computing Ar is a bad idea: High memory requirements, expensive! 1 2 1 2 3 Charalampos E. Tsourakakis

26 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

27 Why is Triangle Counting important? From the Graph Mining Perspective
Clustering coefficient Transitivity ratio Social Network Analysis fact: “Friends of friends are friends” A C B Other applications include: Hidden Thematic Structure of the Web Motif Detection, e.g., biological networks Web Spam Detection Charalampos E. Tsourakakis

28 Theorem [EigenTriangle]
Theorem δ(G) = # triangles in graph G(V,E) = eigenvalues of adjacency matrix A Charalampos E. Tsourakakis

29 Theorem[EigenTriangleLocal]
δ(i) = #Δs vertex i participates at. = i-th eigenvector = j-th entry of Charalampos E. Tsourakakis

30 Charalampos E. Tsourakakis
Algorithm’s idea Almost symmetric around 0! Political Blogs Omit! Keep only 3! 3 Charalampos E. Tsourakakis

31 Results: Edges vs. Speedup
Observe the trend Charalampos E. Tsourakakis

32 Charalampos E. Tsourakakis
#Eigenvalues vs. ϱ 2-3 eigenvalues almost ideal results! Charalampos E. Tsourakakis

33 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

34 Charalampos E. Tsourakakis
Faloutsos, Tong Laplacian 4 1 L= D-A= 2 3 Diagonal matrix, dii=di Charalampos E. Tsourakakis

35 Charalampos E. Tsourakakis
Faloutsos, Tong Weighted Laplacian 4 10 1 4 0.3 2 3 2 Charalampos E. Tsourakakis

36 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

37 Charalampos E. Tsourakakis
Faloutsos, Tong Connected Components Lemma Let G be a graph with n vertices and c connected components. If L is the Laplacian of G, then rank(L)=n-c. Proof see p.279, Godsil-Royle Charalampos E. Tsourakakis

38 Charalampos E. Tsourakakis
Faloutsos, Tong Connected Components G(V,E) 1 2 3 L= 4 6 #zeros = #components 7 5 eig(L)= Charalampos E. Tsourakakis

39 Charalampos E. Tsourakakis
Faloutsos, Tong Connected Components G(V,E) 1 2 3 L= 0.01 4 6 #zeros = #components Indicates a “good cut” 7 5 eig(L)= Charalampos E. Tsourakakis

40 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

41 Adjacency vs. Laplacian Intuition
Faloutsos, Tong Adjacency vs. Laplacian Intuition V-S Let x be an indicator vector: S Consider now y=Lx k-th coordinate Charalampos E. Tsourakakis

42 Adjacency vs. Laplacian Intuition
Faloutsos, Tong Adjacency vs. Laplacian Intuition G30,0.5 S Consider now y=Lx k Charalampos E. Tsourakakis

43 Adjacency vs. Laplacian Intuition
Faloutsos, Tong Adjacency vs. Laplacian Intuition G30,0.5 S Consider now y=Lx k Charalampos E. Tsourakakis

44 Adjacency vs. Laplacian Intuition
Faloutsos, Tong Adjacency vs. Laplacian Intuition G30,0.5 S Consider now y=Lx k k Laplacian: connectivity, Adjacency: #paths Charalampos E. Tsourakakis

45 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

46 Charalampos E. Tsourakakis
Faloutsos, Tong Why Sparse Cuts? Clustering, Community Detection And more: Telephone Network Design, VLSI layout, Sparse Gaussian Elimination, Parallel Computation cut 4 8 1 5 9 2 3 6 7 Charalampos E. Tsourakakis

47 Charalampos E. Tsourakakis
Faloutsos, Tong Quality of a Cut Edge expansion/Isoperimetric number φ 4 1 2 3 Charalampos E. Tsourakakis

48 Charalampos E. Tsourakakis
Faloutsos, Tong Quality of a Cut Edge expansion/Isoperimetric number φ 4 1 and thus 2 3 Charalampos E. Tsourakakis

49 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? V-S Characteristic Vector x S Edges across cut Then: Charalampos E. Tsourakakis

50 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? S V-S cut 4 8 1 5 9 2 3 6 7 x=[1,1,1,1,0,0,0,0,0]T xTLx=2 Charalampos E. Tsourakakis

51 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? Ratio cut Sparsest ratio cut NP-hard Relax the constraint: ? Normalize: Charalampos E. Tsourakakis

52 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? Sparsest ratio cut NP-hard Relax the constraint: λ2 Normalize: because of the Courant-Fisher theorem (applied to L) Charalampos E. Tsourakakis

53 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? OSCILLATE x1 xn Each ball 1 unit of mass Node id Eigenvector value Charalampos E. Tsourakakis

54 Charalampos E. Tsourakakis
Faloutsos, Tong Why λ2? Fundamental mode of vibration: “along” the separator Charalampos E. Tsourakakis

55 Charalampos E. Tsourakakis
Faloutsos, Tong Cheeger Inequality Step 1: Sort vertices in non-decreasing order according to their assigned by the second eigenvector value. Step 2: Decide where to cut. Bisection Best ratio cut Two common heuristics Charalampos E. Tsourakakis

56 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

57 Example: Spectral Partitioning
Faloutsos, Tong Example: Spectral Partitioning K500 K500 dumbbell graph A = zeros(1000); A(1:500,1:500)=ones(500)-eye(500); A(501:1000,501:1000)= ones(500)-eye(500); myrandperm = randperm(1000); B = A(myrandperm,myrandperm); In social network analysis, such clusters are called communities Charalampos E. Tsourakakis

58 Example: Spectral Partitioning
Faloutsos, Tong Example: Spectral Partitioning This is how adjacency matrix of B looks spy(B) Charalampos E. Tsourakakis

59 Example: Spectral Partitioning
Faloutsos, Tong Example: Spectral Partitioning This is how the 2nd eigenvector of B looks like. L = diag(sum(B))-B; [u v] = eigs(L,2,'SM'); plot(u(:,1),’x’) Not so much information yet… Charalampos E. Tsourakakis

60 Example: Spectral Partitioning
Faloutsos, Tong Example: Spectral Partitioning This is how the 2nd eigenvector looks if we sort it. [ign ind] = sort(u(:,1)); plot(u(ind),'x') But now we see the two communities! Charalampos E. Tsourakakis

61 Example: Spectral Partitioning
Faloutsos, Tong Example: Spectral Partitioning This is how adjacency matrix of B looks now spy(B(ind,ind)) Community 1 Cut here! Observation: Both heuristics are equivalent for the dumbell Community 2 Charalampos E. Tsourakakis

62 Charalampos E. Tsourakakis
Faloutsos, Tong Outline Reminders Adjacency matrix Intuition behind eigenvectors: Bipartite Graphs Walks of length k Case Study: Triangles Laplacian Connected Components Intuition: Adjacency vs. Laplacian Sparsest Cut and Cheeger Inequality : Derivation, intuition Example Normalized Laplacian Charalampos E. Tsourakakis

63 Where does it go from here?
Faloutsos, Tong Where does it go from here? Normalized Laplacian Ng, Jordan, Weiss Spectral Clustering Laplacian Eigenmaps for Manifold Learning Computer Vision and many more applications… Standard reference: Spectral Graph Theory Monograph by Fan Chung Graham Charalampos E. Tsourakakis

64 Why Normalized Laplacian
Faloutsos, Tong Why Normalized Laplacian K500 K500 The only weighted edge! Cut here Cut here φ= φ= > So, φ is not good here… Charalampos E. Tsourakakis

65 Why Normalized Laplacian
Faloutsos, Tong Why Normalized Laplacian K500 K500 The only weighted edge! Cut here Cut here Optimize Cheeger constant h(G), balanced cuts φ= φ= > where Charalampos E. Tsourakakis

66 Charalampos E. Tsourakakis
Faloutsos, Tong Conclusions Spectrum tells us a lot about the graph. What to remember What is an eigenvector (f:Nodes Reals) Adjacency: #Paths Laplacian: Sparsest Cut and Intuition Normalized Laplacian: Normalized cuts, tend to avoid unbalanced cuts Charalampos E. Tsourakakis

67 Charalampos E. Tsourakakis
Faloutsos, Tong References A list of references is on my web site, in the KDD tutorial web page Charalampos E. Tsourakakis

68 Charalampos E. Tsourakakis
Faloutsos, Tong Thank you! Charalampos E. Tsourakakis


Download ppt "Spectral Graph Theory (Basics)"

Similar presentations


Ads by Google