DOCSLIB.ORG

Higher Order Learning with Graphs

Total Page：16

File Type：pdf, Size：1020Kb

Download full-text PDF Read full-text
Higher Order Learning with Graphs Higher Order Learning with Graphs Sameer Agarwal [email protected] Kristin Branson [email protected] Serge Belongie [email protected] Department of Computer Science and Engineering, University of California San Diego, La Jolla, CA 92093, USA Abstract assumption that it is possible to measure similarity Recently there has been considerable inter- between pairs of points. Consider a k-lines algorithm d est in learning with higher order relations which clusters points in R into k clusters where ele- (i.e., three-way or higher) in the unsupervised ments in each cluster are well-approximated by a line. and semi-supervised settings. Hypergraphs As every pair of data points trivially define a line, and tensors have been proposed as the nat- there is no useful measure of similarity between pairs ural way of representing these relations and of points for this problem. However, it is possible to their corresponding algebra as the natural define measures of similarity over triplets of points that tools for operating on them. In this paper indicate how close they are to being collinear. This we argue that hypergraphs are not a natural analogy can be extended to any model-based clustering representation for higher order relations, in- task where the fitting error of a set of points to a model deed pairwise as well as higher order relations can be considered a measure of the dissimilarity among can be handled using graphs. We show that them. We refer to similarity/dissimilarity measured various formulations of the semi-supervised over triples or more of points as higher order relations. and the unsupervised learning problem on hy- A number of questions that have been addressed in do- pergraphs result in the same graph theoretic mains with pairwise relations can now be asked for the problem and can be analyzed using existing case of higher order relations. How does one perform tools. clustering in such a domain? How can one formulate and solve the semi-supervised and unsupervised learn- ing problems in this setting? 1. Introduction Hypergraphs are a generalization of graphs in which Given a data set, it is common practice to represent the edges are arbitrary non-empty subsets of the vertex the similarity relation between its elements using a set. Instead of having edges between pairs of vertices, weighted graph. A number of machine learning meth- hypergraphs have edges that connect sets of two or ods for unsupervised and semi-supervised learning can more vertices. While our understanding of hypergraph then be formulated in terms of operations on this graph. spectral methods relative to that of graphs is very In some cases like spectral clustering, the relation be- limited, a number of authors have considered extensions tween the structural and the spectral properties of the of spectral graph theoretic methods to hypergraphs. graph can be exploited to construct matrix theoretic methods that are also graph theoretic. The most com- Another possible representation of higher order rela- monly used matrix in these methods is the Laplacian tions is a tensor. Tensors are a generalization of ma- of the graph (Chung, 1997). In the same manner that trices to higher dimensional arrays, and they can be the Laplace-Beltrami operator is used to analyze the analyzed with multilinear algebra. geometry of continuous manifolds, the Laplacian of a Recently a number of authors have considered the graph is used to study the structure of the graph and problem of unsupervised and semi-supervised learning functions defined on it. in domains with higher order relations (Agarwal et al., A fundamental constraint in this formulation is the 2005; Govindu, 2005; Shashua & Hazan, 2005; Shashua et al., 2006; Zhou et al., 2005). The success of graph rd Appearing in Proceedings of the 23 International Confer- and matrix theoretic representations have prompted ence on Machine Learning, Pittsburgh, PA, 2006. Copyright 2006 by the author(s)/owner(s). researchers to extend these representations to the case Higher Order Learning with Graphs of higher order relations. where S is the |V | × |V | adjacency matrix with entry (u, v) equal to the weight of the edge (u, v) if they are In this paper we focus on spectral graph and hyper- connected, 0 otherwise. An important variant is the graph theoretic methods for learning with higher order normalized Laplacian, relations. We survey a number of approaches from machine learning, VLSI CAD and graph theory that L = I − D−1/2SD−1/2 (3) have been proposed for analyzing the structure of hy- v v pergraphs. We show that despite significant differences For future reference it is useful to rewrite the above ex- in how previous authors have approached the problem, pressions in terms of the vertex-edge incidence relation there are two basic graph constructions that underlie > all these studies. Furthermore, we show that these L = 2Dv − HWH (4) constructions are essentially the same when viewed 1 L = I − D−1/2HWH>D−1/2 (5) through the lens of the normalized Laplacian. 2 v v The paper is organized as follows. Section 2 defines our notation. Section 3 reviews the properties of the graph 3. The Graph Laplacian Laplacian from a machine learning perspective. Sec- The graph Laplacian is the discrete analog of the tion 4 considers the algebraic generalization of Lapla- Laplace-Beltrami operator on compact Riemannian cian to higher order structures and shows why it is not manifolds (Belkin & Niyogi, 2003; Rosenberg, 1997; useful for machine learning tasks. Section 5 presents Chung, 1997). It has been used extensively in machine a survey of graph constructions and linear operators learning, initially for the unsupervised case and then re- related to hypergraphs that various studies have used cently for semi-supervised learning (Zhu, 2005). In this for analyzing the structure of hypergraphs and for unsu- section we highlight some of the properties of the Lapla- pervised and semi-supervised learning. In Section 6 we cian from a machine learning perspective, and motivate show how all these constructions can be reduced to two the search for similar operators on hypergraphs. graph constructions and their associated Laplacians. Finally, in Section 7 we conclude with a summary and Perhaps the earliest use of the graph Laplacian was the discussion of key results. development of spectral clustering algorithms which considered continuous relaxations of graph partitioning 2. Notation problems (Alpert et al., 1999; Shi & Malik, 2000; Ng et al., 2002). The relaxation converted the optimiza- Let G(V, E) denote a hypergraph with vertex set V tion problem into a generalized eigenvalue problems and edge set E. The edges are arbitrary subsets of V involving the Laplacian matrix of the graph. with weight w(e) associated with edge e. The degree In (Zhou & Sch¨olkopf, 2005), the authors develop a dis- d(v) of a vertex is d(v) = P w(e). The degree e∈E|v∈e crete calculus on 2-graphs by treating them as discrete of an edge e is denoted by δ(e) = |e|. For k-uniform analogs of compact Riemannian manifolds. As one of hypergraphs, the degrees of each edge are the same, the consequences of this development they argue that, δ(e) = k. In particular, for the case of ordinary graphs in analogy to the continuous case, the graph Laplacian or “2-graphs,” δ(e) = 2. The vertex-edge incidence be defined as an operator L : H(V ) → H(V ), matrix H is |V | × |E| where the entry h(v, e) is 1 if v ∈ e and 0 otherwise. By these definitions, we have: 1 Lf := div(∇f) (6) X X 2 d(v) = w(e)h(v, e) and δ(e) = h(v, e) (1) e∈E v∈V Zhou et al. also argue that there exists a family of regularization operators on the 2-graphs, the Laplacian De and Dv are the diagonal matrices consisting of being one of them, that can be used for transduction, edge and vertex degrees, respectively. W is the di- i.e., given a partial labeling of the graph vertices y, use agonal matrix of edge weights, w(·). A number of the geometric structure of the graph to induce a labeling different symbols have been used in the literature to f on the unlabeled vertices. The vertex label y(v) denote the Laplacian of graph. We follow the conven- is +1, −1 for positive and negative valued examples, tion in (Chung, 1997) and use L for the combinatorial respectively, and 0 if no information is available about Laplacian and L for the normalized Laplacian. L is the label. They consider the regularized least squares also known as the unnormalized Laplacian of a graph problem and is usually written as 2 arg min hf, Lfi + µkf − yk2 (7) f L = Dv − S (2) Higher Order Learning with Graphs While the discrete version of the problem where f(v) ∈ lines, triangles, etc. Algebraic topologists refer to these {+1, −1} is a hard combinatorial problem, relaxing the functions as p-chains, where p is size of the simplices range of f to the real line R results in a simple linear on which they are defined. Thus vertex functions are least squares problem, solved as f = µ(µI + ∆)−1y.A 0-chains, edge functions are 1-chains, and so on. In similar formulation is considered by (Belkin & Niyogi, each case one can ask the question, how does one mea- 2003). In the absence of any labels, the problem re- sure the variation in these functions with respect to duces to that of clustering, and the eigenmodes of the the geometry of the hypergraph or its corresponding Laplacian are used to label the vertices (Shi & Malik, simplex? 2000; Ng et al., 2002).
Load more

Recommended publications
  • Practical Parallel Hypergraph Algorithms
    Practical Parallel Hypergraph Algorithms Julian Shun [email protected] MIT CSAIL Abstract v While there has been signicant work on parallel graph pro- 0 cessing, there has been very surprisingly little work on high- e0 performance hypergraph processing. This paper presents v0 v1 v1 a collection of ecient parallel algorithms for hypergraph processing, including algorithms for betweenness central- e1 ity, maximal independent set, k-core decomposition, hyper- v2 trees, hyperpaths, connected components, PageRank, and v2 v3 e single-source shortest paths. For these problems, we either 2 provide new parallel algorithms or more ecient implemen- v3 tations than prior work. Furthermore, our algorithms are theoretically-ecient in terms of work and depth. To imple- (a) Hypergraph (b) Bipartite representation ment our algorithms, we extend the Ligra graph processing Figure 1. An example hypergraph representing the groups framework to support hypergraphs, and our implementa- , , , , , , and , , and its bipartite repre- { 0 1 2} { 1 2 3} { 0 3} tions benet from graph optimizations including switching sentation. between sparse and dense traversals based on the frontier size, edge-aware parallelization, using buckets to prioritize processing of vertices, and compression. Our experiments represented as hyperedges, can contain an arbitrary number on a 72-core machine and show that our algorithms obtain of vertices. Hyperedges correspond to group relationships excellent parallel speedups, and are signicantly faster than among vertices (e.g., a community in a social network). An algorithms in existing hypergraph processing frameworks. example of a hypergraph is shown in Figure 1a. CCS Concepts • Computing methodologies → Paral- Hypergraphs have been shown to enable richer analy- lel algorithms; Shared memory algorithms.
    [Show full text]
  • Three Conjectures in Extremal Spectral Graph Theory
    Three conjectures in extremal spectral graph theory Michael Tait and Josh Tobin June 6, 2016 Abstract We prove three conjectures regarding the maximization of spectral invariants over certain families of graphs. Our most difficult result is that the join of P2 and Pn−2 is the unique graph of maximum spectral radius over all planar graphs. This was conjectured by Boots and Royle in 1991 and independently by Cao and Vince in 1993. Similarly, we prove a conjecture of Cvetkovi´cand Rowlinson from 1990 stating that the unique outerplanar graph of maximum spectral radius is the join of a vertex and Pn−1. Finally, we prove a conjecture of Aouchiche et al from 2008 stating that a pineapple graph is the unique connected graph maximizing the spectral radius minus the average degree. To prove our theorems, we use the leading eigenvector of a purported extremal graph to deduce structural properties about that graph. Using this setup, we give short proofs of several old results: Mantel's Theorem, Stanley's edge bound and extensions, the K}ovari-S´os-Tur´anTheorem applied to ex (n; K2;t), and a partial solution to an old problem of Erd}oson making a triangle-free graph bipartite. 1 Introduction Questions in extremal graph theory ask to maximize or minimize a graph invariant over a fixed family of graphs. Perhaps the most well-studied problems in this area are Tur´an-type problems, which ask to maximize the number of edges in a graph which does not contain fixed forbidden subgraphs. Over a century old, a quintessential example of this kind of result is Mantel's theorem, which states that Kdn=2e;bn=2c is the unique graph maximizing the number of edges over all triangle-free graphs.
    [Show full text]
  • Manifold Learning
    MANIFOLD LEARNING Kavé Salamatian- LISTIC Learning Machine Learning: develop algorithms to automatically extract ”patterns” or ”regularities” from data (generalization) Typical tasks Clustering: Find groups of similar points Dimensionality reduction: Project points in a lower dimensional space while preserving structure Semi-supervised: Given labelled and unlabelled points, build a labelling function Supervised: Given labelled points, build a labelling function All these tasks are not well-defined Clustering Identify the two groups Dimensionality Reduction 4096 Objects in R ? But only 3 parameters: 2 angles and 1 for illumination. They span a 3- dimensional sub-manifold of R4096! Semi-Supervised Learning Assign labels to black points Supervised learning Build a function that predicts the label of all points in the space Formal definitions Clustering: Given build a function Dimensionality reduction: Given build a function Semi-supervised: Given and with m << n, build a function Supervised: Given , build Geometric learning Consider and exploit the geometric structure of the data Clustering: two ”close” points should be in the same cluster Dimensionality reduction: ”closeness” should be preserved Semi-supervised/supervised: two ”close” points should have the same label Examples: k-means clustering, k-nearest neighbors. Manifold ? A manifold is a topological space that is locally Euclidian Around every point, there is a neighborhood that is topologically isomorph to unit ball Regularization Adopt a functional viewpoint
    [Show full text]
  • Contemporary Spectral Graph Theory
    CONTEMPORARY SPECTRAL GRAPH THEORY GUEST EDITORS Maurizio Brunetti, University of Naples Federico II, Italy, [email protected] Raffaella Mulas, The Alan Turing Institute, UK, [email protected] Lucas Rusnak, Texas State University, USA, [email protected] Zoran Stanić, University of Belgrade, Serbia, [email protected] DESCRIPTION Spectral graph theory is a striking theory that studies the properties of a graph in relationship to the characteristic polynomial, eigenvalues and eigenvectors of matrices associated with the graph, such as the adjacency matrix, the Laplacian matrix, the signless Laplacian matrix or the distance matrix. This theory emerged in the 1950s and 1960s. Over the decades it has considered not only simple undirected graphs but also their generalizations such as multigraphs, directed graphs, weighted graphs or hypergraphs. In the recent years, spectra of signed graphs and gain graphs have attracted a great deal of attention. Besides graph theoretic research, another major source is the research in a wide branch of applications in chemistry, physics, biology, electrical engineering, social sciences, computer sciences, information theory and other disciplines. This thematic special issue is devoted to the publication of original contributions focusing on the spectra of graphs or their generalizations, along with the most recent applications. Contributions to the Special Issue may address (but are not limited) to the following aspects: f relations between spectra of graphs (or their generalizations) and their structural properties in the most general shape, f bounds for particular eigenvalues, f spectra of particular types of graph, f relations with block designs, f particular eigenvalues of graphs, f cospectrality of graphs, f graph products and their eigenvalues, f graphs with a comparatively small number of eigenvalues, f graphs with particular spectral properties, f applications, f characteristic polynomials.
    [Show full text]
  • Learning on Hypergraphs: Spectral Theory and Clustering
    Learning on Hypergraphs: Spectral Theory and Clustering Pan Li, Olgica Milenkovic Coordinated Science Laboratory University of Illinois at Urbana-Champaign March 12, 2019 Learning on Graphs Graphs are indispensable mathematical data models capturing pairwise interactions: k-nn network social network publication network Important learning on graphs problems: clustering (community detection), semi-supervised/active learning, representation learning (graph embedding) etc. Beyond Pairwise Relations A graph models pairwise relations. Recent work has shown that high-order relations can be significantly more informative: Examples include: Understanding the organization of networks (Benson, Gleich and Leskovec'16) Determining the topological connectivity between data points (Zhou, Huang, Sch}olkopf'07). Graphs with high-order relations can be modeled as hypergraphs (formally defined later). Meta-graphs, meta-paths in heterogeneous information networks. Algorithmic methods for analyzing high-order relations and learning problems are still under development. Beyond Pairwise Relations Functional units in social and biological networks. High-order network motifs: Motif (Benson’16) Microfauna Pelagic fishes Crabs & Benthic fishes Macroinvertebrates Algorithmic methods for analyzing high-order relations and learning problems are still under development. Beyond Pairwise Relations Functional units in social and biological networks. Meta-graphs, meta-paths in heterogeneous information networks. (Zhou, Yu, Han'11) Beyond Pairwise Relations Functional units in social and biological networks. Meta-graphs, meta-paths in heterogeneous information networks. Algorithmic methods for analyzing high-order relations and learning problems are still under development. Review of Graph Clustering: Notation and Terminology Graph Clustering Task: Cluster the vertices that are \densely" connected by edges. Graph Partitioning and Conductance A (weighted) graph G = (V ; E; w): for e 2 E, we is the weight.
    [Show full text]
  • Multilevel Hypergraph Partitioning with Vertex Weights Revisited
    Multilevel Hypergraph Partitioning with Vertex Weights Revisited Tobias Heuer ! Karlsruhe Institute of Technology, Karlsruhe, Germany Nikolai Maas ! Karlsruhe Institute of Technology, Karlsruhe, Germany Sebastian Schlag ! Karlsruhe Institute of Technology, Karlsruhe, Germany Abstract The balanced hypergraph partitioning problem (HGP) is to partition the vertex set of a hypergraph into k disjoint blocks of bounded weight, while minimizing an objective function defined on the hyperedges. Whereas real-world applications often use vertex and edge weights to accurately model the underlying problem, the HGP research community commonly works with unweighted instances. In this paper, we argue that, in the presence of vertex weights, current balance constraint definitions either yield infeasible partitioning problems or allow unnecessarily large imbalances and propose a new definition that overcomes these problems. We show that state-of-the-art hypergraph partitioners often struggle considerably with weighted instances and tight balance constraints (even with our new balance definition). Thus, we present a recursive-bipartitioning technique that isable to reliably compute balanced (and hence feasible) solutions. The proposed method balances the partition by pre-assigning a small subset of the heaviest vertices to the two blocks of each bipartition (using an algorithm originally developed for the job scheduling problem) and optimizes the actual partitioning objective on the remaining vertices. We integrate our algorithm into the multilevel hypergraph
    [Show full text]
  • On Decomposing a Hypergraph Into K Connected Sub-Hypergraphs
    Egervary´ Research Group on Combinatorial Optimization Technical reportS TR-2001-02. Published by the Egrerv´aryResearch Group, P´azm´any P. s´et´any 1/C, H{1117, Budapest, Hungary. Web site: www.cs.elte.hu/egres . ISSN 1587{4451. On decomposing a hypergraph into k connected sub-hypergraphs Andr´asFrank, Tam´asKir´aly,and Matthias Kriesell February 2001 Revised July 2001 EGRES Technical Report No. 2001-02 1 On decomposing a hypergraph into k connected sub-hypergraphs Andr´asFrank?, Tam´asKir´aly??, and Matthias Kriesell??? Abstract By applying the matroid partition theorem of J. Edmonds [1] to a hyper- graphic generalization of graphic matroids, due to M. Lorea [3], we obtain a gen- eralization of Tutte's disjoint trees theorem for hypergraphs. As a corollary, we prove for positive integers k and q that every (kq)-edge-connected hypergraph of rank q can be decomposed into k connected sub-hypergraphs, a well-known result for q = 2. Another by-product is a connectivity-type sufficient condition for the existence of k edge-disjoint Steiner trees in a bipartite graph. Keywords: Hypergraph; Matroid; Steiner tree 1 Introduction An undirected graph G = (V; E) is called connected if there is an edge connecting X and V X for every nonempty, proper subset X of V . Connectivity of a graph is equivalent− to the existence of a spanning tree. As a connected graph on t nodes contains at least t 1 edges, one has the following alternative characterization of connectivity. − Proposition 1.1. A graph G = (V; E) is connected if and only if the number of edges connecting distinct classes of is at least t 1 for every partition := V1;V2;:::;Vt of V into non-empty subsets.P − P f g ?Department of Operations Research, E¨otv¨osUniversity, Kecskem´etiu.
    [Show full text]
  • Hypergraph Packing and Sparse Bipartite Ramsey Numbers
    Hypergraph packing and sparse bipartite Ramsey numbers David Conlon ∗ Abstract We prove that there exists a constant c such that, for any integer ∆, the Ramsey number of a bipartite graph on n vertices with maximum degree ∆ is less than 2c∆n. A probabilistic argument due to Graham, R¨odland Ruci´nskiimplies that this result is essentially sharp, up to the constant c in the exponent. Our proof hinges upon a quantitative form of a hypergraph packing result of R¨odl,Ruci´nskiand Taraz. 1 Introduction For a graph G, the Ramsey number r(G) is defined to be the smallest natural number n such that, in any two-colouring of the edges of Kn, there exists a monochromatic copy of G. That these numbers exist was proven by Ramsey [19] and rediscovered independently by Erd}osand Szekeres [10]. Whereas the original focus was on finding the Ramsey numbers of complete graphs, in which case it is known that t t p2 r(Kt) 4 ; ≤ ≤ the field has broadened considerably over the years. One of the most famous results in the area to date is the theorem, due to Chvat´al,R¨odl,Szemer´ediand Trotter [7], that if a graph G, on n vertices, has maximum degree ∆, then r(G) c(∆)n; ≤ where c(∆) is just some appropriate constant depending only on ∆. Their proof makes use of the regularity lemma and because of this the bound it gives on the constant c(∆) is (and is necessarily [11]) very bad. The situation was improved somewhat by Eaton [8], who proved, by using a variant of the regularity lemma, that the function c(∆) may be taken to be of the form 22c∆ .
    [Show full text]
  • Hypernetwork Science: from Multidimensional Networks to Computational Topology∗
    Hypernetwork Science: From Multidimensional Networks to Computational Topology∗ Cliff A. Joslyn,y Sinan Aksoy,z Tiffany J. Callahan,x Lawrence E. Hunter,x Brett Jefferson,z Brenda Praggastis,y Emilie A.H. Purvine,y Ignacio J. Tripodix March, 2020 Abstract cations, and physical infrastructure often afford a rep- resentation as such a set of entities with binary rela- As data structures and mathematical objects used tionships, and hence may be analyzed utilizing graph for complex systems modeling, hypergraphs sit nicely theoretical methods. poised between on the one hand the world of net- Graph models benefit from simplicity and a degree of work models, and on the other that of higher-order universality. But as abstract mathematical objects, mathematical abstractions from algebra, lattice the- graphs are limited to representing pairwise relation- ory, and topology. They are able to represent com- ships between entities, while real-world phenomena plex systems interactions more faithfully than graphs in these systems can be rich in multi-way relation- and networks, while also being some of the simplest ships involving interactions among more than two classes of systems representing topological structures entities, dependencies between more than two vari- as collections of multidimensional objects connected ables, or properties of collections of more than two in a particular pattern. In this paper we discuss objects. Representing group interactions is not possi- the role of (undirected) hypergraphs in the science ble in graphs natively, but rather requires either more of complex networks, and provide a mathematical complex mathematical objects, or coding schemes like overview of the core concepts needed for hypernet- “reification” or semantic labeling in bipartite graphs.
    [Show full text]
  • Spectral Geometry for Structural Pattern Recognition
    Spectral Geometry for Structural Pattern Recognition HEWAYDA EL GHAWALBY Ph.D. Thesis This thesis is submitted in partial fulfilment of the requirements for the degree of Doctor of Philosophy. Department of Computer Science United Kingdom May 2011 Abstract Graphs are used pervasively in computer science as representations of data with a network or relational structure, where the graph structure provides a flexible representation such that there is no fixed dimensionality for objects. However, the analysis of data in this form has proved an elusive problem; for instance, it suffers from the robustness to structural noise. One way to circumvent this problem is to embed the nodes of a graph in a vector space and to study the properties of the point distribution that results from the embedding. This is a problem that arises in a number of areas including manifold learning theory and graph-drawing. In this thesis, our first contribution is to investigate the heat kernel embed- ding as a route to computing geometric characterisations of graphs. The reason for turning to the heat kernel is that it encapsulates information concerning the distribution of path lengths and hence node affinities on the graph. The heat ker- nel of the graph is found by exponentiating the Laplacian eigensystem over time. The matrix of embedding co-ordinates for the nodes of the graph is obtained by performing a Young-Householder decomposition on the heat kernel. Once the embedding of its nodes is to hand we proceed to characterise a graph in a geometric manner. With the embeddings to hand, we establish a graph character- ization based on differential geometry by computing sets of curvatures associated ii Abstract iii with the graph nodes, edges and triangular faces.
    [Show full text]
  • CS 598: Spectral Graph Theory. Lecture 10
    CS 598: Spectral Graph Theory. Lecture 13 Expander Codes Alexandra Kolla Today Graph approximations, part 2. Bipartite expanders as approximations of the bipartite complete graph Quasi-random properties of bipartite expanders, expander mixing lemma Building expander codes Encoding, minimum distance, decoding Expander Graph Refresher We defined expander graphs to be d- regular graphs whose adjacency matrix eigenvalues satisfy |훼푖| ≤ 휖푑 for i>1, and some small 휖. We saw that such graphs are very good approximations of the complete graph. Expanders as Approximations of the Complete Graph Refresher Let G be a d-regular graph whose adjacency eigenvalues satisfy |훼푖| ≤ 휖푑. As its Laplacian eigenvalues satisfy 휆푖 = 푑 − 훼푖, all non-zero eigenvalues are between (1 − ϵ)d and 1 + ϵ 푑. 푇 푇 Let H=(d/n)Kn, 푠표 푥 퐿퐻푥 = 푑푥 푥 We showed that G is an ϵ − approximation of H 푇 푇 푇 1 − ϵ 푥 퐿퐻푥 ≼ 푥 퐿퐺푥 ≼ 1 + ϵ 푥 퐿퐻푥 Bipartite Expander Graphs Define them as d-regular good approximations of (some multiple of) the complete bipartite 푑 graph H= 퐾 : 푛 푛,푛 푇 푇 푇 1 − ϵ 푥 퐿퐻푥 ≼ 푥 퐿퐺푥 ≼ 1 + ϵ 푥 퐿퐻푥 ⇒ 1 − ϵ 퐻 ≼ 퐺 ≼ 1 + ϵ 퐻 G 퐾푛,푛 U V U V The Spectrum of Bipartite Expanders 푑 Eigenvalues of Laplacian of H= 퐾 are 푛 푛,푛 휆1 = 0, 휆2푛 = 2푑, 휆푖 = 푑, 0 < 푖 < 2푛 푇 푇 푇 1 − ϵ 푥 퐿퐻푥 ≼ 푥 퐿퐺푥 ≼ 1 + ϵ 푥 퐿퐻푥 Says we want d-regular graph G that has evals 휆1 = 0, 휆2푛 = 2푑, 휆푖 ∈ ( 1 − 휖 푑, 1 + 휖 푑), 0 < 푖 < 2푛 Construction of Bipartite Expanders Definition (Double Cover).
    [Show full text]
  • Notes on Elementary Spectral Graph Theory Applications to Graph Clustering Using Normalized Cuts
    University of Pennsylvania ScholarlyCommons Technical Reports (CIS) Department of Computer & Information Science 1-1-2013 Notes on Elementary Spectral Graph Theory Applications to Graph Clustering Using Normalized Cuts Jean H. Gallier University of Pennsylvania, [email protected] Follow this and additional works at: https://repository.upenn.edu/cis_reports Part of the Computer Engineering Commons, and the Computer Sciences Commons Recommended Citation Jean H. Gallier, "Notes on Elementary Spectral Graph Theory Applications to Graph Clustering Using Normalized Cuts", . January 2013. University of Pennsylvania Department of Computer and Information Science Technical Report No. MS-CIS-13-09. This paper is posted at ScholarlyCommons. https://repository.upenn.edu/cis_reports/986 For more information, please contact [email protected]. Notes on Elementary Spectral Graph Theory Applications to Graph Clustering Using Normalized Cuts Abstract These are notes on the method of normalized graph cuts and its applications to graph clustering. I provide a fairly thorough treatment of this deeply original method due to Shi and Malik, including complete proofs. I include the necessary background on graphs and graph Laplacians. I then explain in detail how the eigenvectors of the graph Laplacian can be used to draw a graph. This is an attractive application of graph Laplacians. The main thrust of this paper is the method of normalized cuts. I give a detailed account for K = 2 clusters, and also for K > 2 clusters, based on the work of Yu and Shi. Three points that do not appear to have been clearly articulated before are elaborated: 1. The solutions of the main optimization problem should be viewed as tuples in the K-fold cartesian product of projective space RP^{N-1}.
    [Show full text]