The Laplacian Matrices of Graphs Daniel A Spielman
- Slides: 81
The Laplacian Matrices of Graphs Daniel A. Spielman ISIT, July 12, 2016
Outline Laplacians Interpolation on graphs Resistor networks Spring networks Graph drawing Clustering Linear programming Sparsification Solving Laplacian Equations Best results The simplest algorithm
Interpolation on Graphs (Zhu, Ghahramani, Lafferty ’ 0 Interpolate values of a function at all vertices from given values at a few vertices. Minimize Subject to given values CDC 20 0 CDC 27 ANAPC 10 ANAPC 2 ANAPC 5 UBE 2 C 1
Interpolation on Graphs (Zhu, Ghahramani, Lafferty ’ 0 Interpolate values of a function at all vertices from given values at a few vertices. Minimize Subject to given values 0. 51 0 CDC 27 0. 30 ANAPC 10 ANAPC 2 ANAPC 5 1 0. 53 UBE 2 C 0. 61
Interpolation on Graphs (Zhu, Ghahramani, Lafferty ’ 0 Interpolate values of a function at all vertices from given values at a few vertices. Minimize Subject to given values 0. 51 0 CDC 27 0. 30 ANAPC 10 ANAPC 2 ANAPC 5 1 0. 53 UBE 2 C 0. 61
Interpolation on Graphs (Zhu, Ghahramani, Lafferty ’ 0 Interpolate values of a function at all vertices from given values at a few vertices. Minimize Subject to given values 0. 51 0 CDC 27 0. 30 ANAPC 10 ANAPC 2 ANAPC 5 1 0. 53 UBE 2 C 0. 61 Take derivatives. Minimize by solving Laplacian
The Laplacian Quadratic Form of 0. 51 1 0. 53 0 0. 61 0. 30
The Laplacian Quadratic Form of 0. 51 2 ) 1 (0. 5 . (0 0 (0 (0. 49)2 02 (0. 53)2 )2 2 ) 7 4 (0. 0. 53 (0. 0 . 3 0) 2 0. 30 1 (0. 31)2 8) 2 (0. 39)2 0. 61
Graphs with positive edge weights
Resistor Networks View edges as resistors connecting vertices Apply voltages at some vertices. Measure induced voltages and current flow. 1 V 0 V
Resistor Networks Induced voltages minimize subject to constraints. 1 V 0 V
Resistor Networks Induced voltages minimize subject to constraints. 1 V 1 V 0. 5 V 0 V 0 V 0. 5 V 0. 375 V 0. 625 V
Resistor Networks Induced voltages minimize subject to constraints. 1 V 1 V 0. 5 V 0 V 0 V 0. 5 V 0. 375 V 0. 625 V
Resistor Networks Induced voltages minimize subject to constraints. Effective resistance = 1/(current flow at one volt) 1 V 0. 5 5 0 V 0. 3 0. 5 75 1 V 0. 5 0. 1 2 25 5 1 0. 25 0 V 0. 375 V 0. 625 V
Spring Networks View edges as rubber bands or ideal linear springs Nail down some vertices, let rest settle When stretched to length potential energy is
Spring Networks Nail down some vertices, let rest settle Physics: position minimizes total potential energy subject to boundary constraints (nails)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks If the graph is planar, then the spring drawing has no crossing edges! (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Drawing by Spring Networks (Tutte ’ 63)
Spectral Graph Theory A n-by-n symmetric matrix has n real eigenvalues and eigenvectors such that These eigenvalues and eigenvectors tell us a lot about a graph!
Spectral Graph Theory A n-by-n symmetric matrix has n real eigenvalues and eigenvectors such that These eigenvalues and eigenvectors tell us a lot about a graph! (excluding )
Spectral Graph Drawing 1 2 3 5 4 6 8 7 9 Original Drawing (Hall ’ 70)
Spectral Graph Drawing (Hall ’ 70) Plot vertex at draw edges as straight lines 9 1 2 3 5 5 4 1 6 8 7 4 6 8 9 7 Original Drawing 3 2 Spectral Drawing
Spectral Graph Drawing Original Drawing (Hall ’ 70) Spectral Drawing
Spectral Graph Drawing Original Drawing (Hall ’ 70) Spectral Drawing
Dodecahedron Best embedded by first three eigenvectors
Erdos’s co-authorship graph
When there is a “nice” drawing Most edges are short Vertices are spread out and don’t clump too much is close to 0 When is big, say there is no nice picture of the graph
Measuring boundaries of sets Boundary: edges leaving a set S S
Measuring boundaries of sets Boundary: edges leaving a set Characteristic Vector of S: 0 0 0 1 1 S S 1 1 0 0 0
Measuring boundaries of sets Boundary: edges leaving a set Characteristic Vector of S: 0 0 0 1 1 S S 1 1 0 0 0
Spectral Clustering and Partitioning Find large sets of small boundary Heuristic to find x with small 0. 7 Consider the level sets S S 1. 3 0. 2 0. 5 1. 3 1. 6 Compute eigenvector -0. 4 0. 9 1. 0 1. 1 0. 4 -1. 1 -0. 20 0. 8 0. 5 -0. 3
Spectral Partitioning (Donath-Hoffman ‘ 72, Barnes ‘ 82, Hagen-Kahng ’ 92) for some Cheeger’s inequality implies good approximation
Spectral Partitioning (Donath-Hoffman ‘ 72, Barnes ‘ 82, Hagen-Kahng ’ 92) for some Cheeger’s inequality implies good approximation
The Laplacian Matrix of a Graph 2 1 6 4 3 5 Symmetric Non-positive off-diagonals Diagonally dominant
The Laplacian Matrix of a Graph
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers Where m is number of non-zeros and n is dimension
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers Koutis, Miller, Peng ’ 11: Low-stretch trees and sampling Where m is number of non-zeros and n is dimension
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers Koutis, Miller, Peng ’ 11: Low-stretch trees and sampling Cohen, Kyng, Pachocki, Peng, Rao ’ 14: Where m is number of non-zeros and n is dimension
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers Koutis, Miller, Peng ’ 11: Low-stretch trees and sampling Cohen, Kyng, Pachocki, Peng, Rao ’ 14: Good code: LAMG (lean algebraic multigrid) – Livne-Brandt CMG (combinatorial multigrid) – Koutis
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers An -accurate solution to is an satisfying where
Quickly Solving Laplacian Equations S, Teng ’ 04: Using low-stretch trees and sparsifiers An -accurate solution to is an satisfying Allows fast computation of eigenvectors corresponding to small eigenvalues.
Laplacians in Linear Programming Laplacians appear when solving Linear Programs on on graphs by Interior Point Methods Maximum and Min-Cost Flow Shortest Paths Isotonic Regression (Daitch, S ’ 08, Mądry ‘ 13) (Cohen, Mądry, Sankowski, Vladu ‘ 16) (Kyng, Rao, Sachdeva ‘ 15) Lipschitz Learning : regularized interpolation on graphs (Kyng, Rao, Sachdeva, S ‘ 15)
Interior Point Method for Maximum s-t Flow maximize subject to /1 /3 t /1 s /1 /4
Interior Point Method for Maximum s-t Flow maximize subject to 1/1 2/3 3 t 1/1 s 1/1 2/4 3
Interior Point Method for Maximum s-t Flow maximize subject to Multiple calls with varying weights maximize subject to
Spectral Sparsification Every graph can be approximated by a sparse graph with a similar Laplacian
Approximating Graphs A graph H is an -approximation of G if for all x
Approximating Graphs A graph H is an -approximation of G if for all x Preserves boundaries of every set
Approximating Graphs A graph H is an -approximation of G if for all x Solutions to linear equations are similar As are effective resistances
Expanders Sparsify Complete Graphs Yield good LDPC codes Every set of vertices has large boundary is large Random regular graphs are usually expanders
Sparsification by Random Sampling Assign a probability pa, b to each edge (a, b) Include edge (a, b) in H with probability pa, b. If include edge (a, b), give it weight wa, b /pa, b
Sparsification by Random Sampling Choose pa, b to be wa, b times the effective resistance between a and b. Low resistance between a and b means there are many alternate routes for current to flow and that the edge is not critical. Proof by random matrix concentration bounds (Rudelson, Ahlswede-Winter, Tropp, etc. ) Only need edges (S, Srivastava ‘ 08)
Optimal Graph Sparsification? For every , there is a s. t. and Is within a factor of 2 of how well Ramanujan expanders approximate complete graphs (Batson, S, Srivastava ‘ 09)
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) Gaussian Elimination: compute upper triangular U so that Approximate Gaussian Elimination: compute sparse upper triangular U so that
Additive view of Gaussian Elimination •
Additive view of Gaussian Elimination Find the rank-1 matrix that agrees on the first row and column.
Additive view of Gaussian Elimination Subtract the rank 1 matrix. We have eliminated the first variable.
Additive view of Gaussian Elimination
Additive view of Gaussian Elimination Find the rank-1 matrix that agrees on the next row and column.
Additive view of Gaussian Elimination Subtract the rank 1 matrix. We have eliminated the second variable.
Additive view of Gaussian Elimination Running time proportional to sum of squares of number of non-zeros in these vectors.
Additive view of Gaussian Elimination
Additive view of Gaussian Elimination
Gaussian Elimination of Laplacians If this is a Laplacian, then so is this
Gaussian Elimination of Laplacians If this is a Laplacian, then so is this When eliminate a node, add a clique on its neighbors 4 2 4 1 8 3 5 1 2 2 4 6 2 2 3 5 4 6
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) 1. when eliminate a node, add a clique on its neighbors 4 2 4 1 8 3 5 1 2 2 4 6 5 2 2 4 3 2. Sparsify that clique, without ever constructing it 6
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) 1. When eliminate a node of degree d, add d edges at random between its neighbors, sampled with probability proportional to the weight of the edge to the eliminated node 1
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) 0. Initialize by randomly permuting vertices, and making copies of every edge 1. When eliminate a node of degree d, add d edges at random between its neighbors, sampled with probability proportional to the weight of the edge to the eliminated node Total time is
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) 0. Initialize by randomly permuting vertices, and making copies of every edge 1. When eliminate a node of degree d, add d edges at random between its neighbors, sampled with probability proportional to the weight of the edge to the eliminated node Total time is Can be improved by sacrificing some simplicity
Approximate Gaussian Elimination (Kyng & Sachdeva ‘ 16) Analysis by Random Matrix Theory: Write UTU as a sum of random matrices. Random permutation and copying control the variances of the random matrices Apply Matrix Freedman inequality (Tropp ‘ 11)
Recent Developments Other families of linear systems (Kyng, Lee, Peng, Sachdeva, S ‘ 16) complex-weighted Laplacians connection Laplacians. jl
To learn more My web page on: Laplacian linear equations, sparsification, local graph clustering, low-stretch spanning trees, and so on. My class notes from “Graphs and Networks” and “Spectral Graph Theory” Lx = b, by Nisheeth Vishnoi
- Spectral graph theory and its applications
- Ingrid spielman
- Stephanie spielman osu
- Spectral graph theory spielman
- G = (v e)
- Laplacian operator
- Laplacian of gaussian
- Laplacian filter
- Lnxn
- Definition of a conservative vector field
- Graph laplacian regularization
- Laplace transfer function table
- Laplace equation spherical
- Laplacian operator
- Project outline
- Laplacian of gaussian
- Laplacian of a vector
- Polynomial function form
- State graphs in software testing
- Comparing distance/time graphs to speed/time graphs
- Graphs that enlighten and graphs that deceive
- Cái miệng nó xinh thế
- Mật thư tọa độ 5x5
- Bổ thể
- Tư thế ngồi viết
- Thẻ vin
- Ví dụ về giọng cùng tên
- Thể thơ truyền thống
- Hát lên người ơi alleluia
- Các châu lục và đại dương trên thế giới
- Từ ngữ thể hiện lòng nhân hậu
- Khi nào hổ con có thể sống độc lập
- Diễn thế sinh thái là
- Vẽ hình chiếu vuông góc của vật thể sau
- Phép trừ bù
- Tỉ lệ cơ thể trẻ em
- Lời thề hippocrates
- đại từ thay thế
- Quá trình desamine hóa có thể tạo ra
- Công thức tiính động năng
- Các môn thể thao bắt đầu bằng từ đua
- Thế nào là mạng điện lắp đặt kiểu nổi
- Hình ảnh bộ gõ cơ thể búng tay
- Khi nào hổ con có thể sống độc lập
- Dạng đột biến một nhiễm là
- Nguyên nhân của sự mỏi cơ sinh 8
- Vẽ hình chiếu đứng bằng cạnh của vật thể
- độ dài liên kết
- Voi kéo gỗ như thế nào
- Thiếu nhi thế giới liên hoan
- điện thế nghỉ
- Một số thể thơ truyền thống
- Trời xanh đây là của chúng ta thể thơ
- Thế nào là hệ số cao nhất
- Slidetodoc
- Số nguyên tố là
- Tia chieu sa te
- đặc điểm cơ thể của người tối cổ
- Các châu lục và đại dương trên thế giới
- Chụp phim tư thế worms-breton
- Hệ hô hấp
- ưu thế lai là gì
- Tư thế ngồi viết
- Vectors and matrices
- Dr frost matrices
- Matrices simultaneous equations worksheet
- Clasificacion de matrices
- Unit 1 algebra basics homework 6 matrices
- Igualdad de matriz
- Matrices 2 bachillerato
- Google docs matrices
- Matrix transpose times matrix
- Matrices diagonal
- Leslie matrix
- Distance matrices
- Matriz nxm
- Matrix multiplication associative property
- Stata vector
- Vectores y matrices en java
- Subtract matrices calculator
- Matrices
- Eigenvectors for dummies