Deterministic Gossiping

Slides:



Advertisements
Similar presentations
Algorithms (and Datastructures) Lecture 3 MAS 714 part 2 Hartmut Klauck.
Advertisements

Chapter 8 Topics in Graph Theory
8.3 Representing Relations Connection Matrices Let R be a relation from A = {a 1, a 2,..., a m } to B = {b 1, b 2,..., b n }. Definition: A n m  n connection.
Graph-02.
Review Binary Search Trees Operations on Binary Search Tree
13 May 2009Instructor: Tasneem Darwish1 University of Palestine Faculty of Applied Engineering and Urban Planning Software Engineering Department Introduction.
1 Slides based on those of Kenneth H. Rosen Slides by Sylvia Sorkin, Community College of Baltimore County - Essex Campus Graphs.
 Graph Graph  Types of Graphs Types of Graphs  Data Structures to Store Graphs Data Structures to Store Graphs  Graph Definitions Graph Definitions.
Entropy Rates of a Stochastic Process
Applied Discrete Mathematics Week 12: Trees
Rutgers May 25, 2011 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A AA A AAA A A A A A A DIMACS Workshop on Perspectives.
Introduction to Graphs
KNURE, Software department, Ph , N.V. Bilous Faculty of computer sciences Software department, KNURE Discrete.
Discrete Mathematics Lecture 9 Alexander Bukharovich New York University.
Introduction to Flocking {Stochastic Matrices} A. S. Morse Yale University Gif – sur - Yvette May 21, 2012 TexPoint fonts used in EMF. Read the TexPoint.
A. S. Morse Yale University University of Minnesota June 4, 2014 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A.
: Appendix A: Mathematical Foundations 1 Montri Karnjanadecha ac.th/~montri Principles of.
GRAPH Learning Outcomes Students should be able to:
Linear Algebra Review 1 CS479/679 Pattern Recognition Dr. George Bebis.
Section 4.1 Using Matrices to Represent Data. Matrix Terminology A matrix is a rectangular array of numbers enclosed in a single set of brackets. The.
GRAPHS CSE, POSTECH. Chapter 16 covers the following topics Graph terminology: vertex, edge, adjacent, incident, degree, cycle, path, connected component,
1 1.3 © 2012 Pearson Education, Inc. Linear Equations in Linear Algebra VECTOR EQUATIONS.
1 CS104 : Discrete Structures Chapter V Graph Theory.
Three different ways There are three different ways to show that ρ(A) is a simple eigenvalue of an irreducible nonnegative matrix A:
Computing Eigen Information for Small Matrices The eigen equation can be rearranged as follows: Ax = x  Ax = I n x  Ax - I n x = 0  (A - I n )x = 0.
2.4 Irreducible Matrices. Reducible is reducible if there is a permutation P such that where A 11 and A 22 are square matrices each of size at least one;
Ch.6 Phylogenetic Trees 2 Contents Phylogenetic Trees Character State Matrix Perfect Phylogeny Binary Character States Two Characters Distance Matrix.
Week 11 - Monday.  What did we talk about last time?  Binomial theorem and Pascal's triangle  Conditional probability  Bayes’ theorem.
Markov Chains and Random Walks. Def: A stochastic process X={X(t),t ∈ T} is a collection of random variables. If T is a countable set, say T={0,1,2, …
A. S. Morse Yale University University of Minnesota June 2, 2014 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A.
Meeting 18 Matrix Operations. Matrix If A is an m x n matrix - that is, a matrix with m rows and n columns – then the scalar entry in the i th row and.
Relevant Subgraph Extraction Longin Jan Latecki Based on : P. Dupont, J. Callut, G. Dooms, J.-N. Monette and Y. Deville. Relevant subgraph extraction from.
An Introduction to Graph Theory
Graph Theory and Applications
CSCI 115 Chapter 8 Topics in Graph Theory. CSCI 115 §8.1 Graphs.
Graphs A graphs is an abstract representation of a set of objects, called vertices or nodes, where some pairs of the objects are connected by links, called.
GRAPHS. Graph Graph terminology: vertex, edge, adjacent, incident, degree, cycle, path, connected component, spanning tree Types of graphs: undirected,
 Quotient graph  Definition 13: Suppose G(V,E) is a graph and R is a equivalence relation on the set V. We construct the quotient graph G R in the follow.
Napa Valley August 3, M. Cao A. S. Morse B. D. O. Anderson Yale University Yale University Australian National University Vicsek’s System with Integer.
Instructor: Mircea Nicolescu Lecture 8 CS 485 / 685 Computer Vision.
Introduction to NP Instructor: Neelima Gupta 1.
UCLA March 2, 2006 IPAM Workshop on Swarming by Nature and by Design thanks to the organizers: A. Bertozzi D. Grunbaum P. S. Krishnaprasad I. Schwartz.
Theory of Computational Complexity Probability and Computing Lee Minseon Iwama and Ito lab M1 1.
Week 11 - Wednesday.  What did we talk about last time?  Graphs  Paths and circuits.
1 GRAPH Learning Outcomes Students should be able to: Explain basic terminology of a graph Identify Euler and Hamiltonian cycle Represent graphs using.
1 Objective To provide background material in support of topics in Digital Image Processing that are based on matrices and/or vectors. Review Matrices.
رياضيات متقطعة لعلوم الحاسب MATH 226. Chapter 10.
2.1 Matrix Operations 2. Matrix Algebra. j -th column i -th row Diagonal entries Diagonal matrix : a square matrix whose nondiagonal entries are zero.
More NP-Complete and NP-hard Problems
Linear Algebra Review.
Markov Chains and Random Walks
Matrix Representation of Graphs
Introduction to Graphs
Section 4.1 Eigenvalues and Eigenvectors
Graph theory Definitions Trees, cycles, directed graphs.
Flow in Network.
Discrete Structures – CNS2300
Introduction to Graphs
Graphs All tree structures are hierarchical. This means that each node can only have one parent node. Trees can be used to store data which has a definite.
Linear Equations in Linear Algebra
2. Matrix Algebra 2.1 Matrix Operations.
CS100: Discrete structures
Graph Operations And Representation
Trees.
G-v, or G-{v} When we remove a vertex v from a graph, we must remove all edges incident with the vertex v. When a edge is removed from a graph, without.
Connectivity Section 10.4.
Graph Implementation.
Graphs G = (V, E) V are the vertices; E are the edges.
Introduction to Graphs
Introduction to Graphs
Presentation transcript:

Deterministic Gossiping ARO SWARMS MURI Deterministic Gossiping A. S. Morse Yale University February 23, 2010 University of Pennsylvania

Consensus Process Consider a group of n agents labeled 1 to n The groups’ neighbor graph N is an undirected, connected graph with vertices labeled 1,2,...,n. 7 4 1 3 5 2 6 The neighbors of agent i , other than itself, correspond to those vertices which are adjacent to vertex i Each agent i controls a real-valued scalar quantity xi called a consensus variable. The goal of a consensus process is for all n agents to ultimately reach a consensus by adjusting their individual consensus variables to a common value. This is to be accomplished over time by sharing information among neighbors in a distributed manner.

Consensus Process A consensus process is a recursive process which evolves with respect to a discrete time scale. In a standard consensus process, agent i sets the value of its consensus variable at time t +1 equal to the average of the values of its neighbors’ consensus variables at time t, assuming agent i is a neighbor of itself. Average at time t of values of consensus variables of neighbors of agent i. Ni = set of indices of agent i0s neighbors. ni = number of indices in Ni A time-varying consensus process is one in which the neighbor graph N depends on t.

Gossip Process A gossip process is a consensus process in which at each clock time, each agent is allowed to average its consensus variable with the consensus variable of at most one of its neighbors. The index of the neighbor of agent i which agent i gossips with at time t. This is called a gossip and is denoted by (i, j). In the most commonly studied version of gossiping, the specific sequence of gossips which occurs during a gossiping process is determined probabilistically. In a deterministic gossiping process, the sequence of gossips which occurs is determined by a pre-specified protocol. If more than one pair of agents gossip at a given time, the event is called a multi-gossip.

A finite sequence of multi-gossips determines a subgraph M of the groups’ neighbor graph N where (i, j) is an edge in M iff (i, j) is a gossip in the sequence. A finite sequence of multi-gossips is complete if the sub-graph it determines is a spanning tree of N. An infinite multi-gossip sequence is periodic with period T if the finite sequence of multi-gossips which occur in any given period repeats itself on each successive period of length T. An infinite periodic multi-gossip sequence with period T is periodically complete if the finite sequence of multi-gossips which occur in any given period is complete. A uniformly aperiodic multi-gossip sequence is composed of an infinite sequence of successive complete subsequences, each of at length at most T.

State Space Models For a time-varying consensus process each value of M(t) is a “stochastic matrix” with positive diagonals. For a gossip process each value of M(t) is a “doubly stochastic matrix” with positive diagonals. A square matrix S is stochastic if it has only nonnegative entries and if its row sums all equal 1. A square matrix S is doubly stochastic if it has only nonnegative entries and if its row and column sums all equal 1. Stochastic S1 = 1 Doubly Stochastic S1 = 1 and S01 =1

State Space Models Reaching a consensus: Convergability A compact subset of matrices M is convergable if for each infinite sequence of matrices S1, S2, .... from M, the matrix product Si Si-1S1 ! 1c. What are conditions for convergability?

The graph of a nonnegative square matrix M, written °(M), is a directed graph on n vertices with an arc from i to j iff mji  0. Strong connectivity: There is a directed path from each vertex to each other vertex. A compact subset of matrices M is convergable if for each infinite sequence of matrices S1, S2, .... from M, the matrix product Si Si-1S1 ! 1c. Weak connectivity: There is an undirected path between each pair of vertices Rooted: For at least one vertex v, there is a directed path from v to each other vertex. Let S = set of stochastic matrices whose graphs have self arcs at all vertices. A compact subset of S is convergable iff the graphs of its matrices are all rooted. Let D = set of doubly stochastic matrices whose graphs have self arcs at all vertices. A compact subset of D is convergable iff the graphs of its matrices are all weakly connected. What about convergence rates?

Periodically Complete Multi-Gossip Sequences The convergence rate of a periodically complete multi-gossip sequence is no slower than ¸ = second largest eigenvalue of the stochastic matrix determined by a sequence of multi-gossips in a period T. For a given spanning tree T ½ N, how does ¸ depend on the order in which the gossips denoted by the edges of T are carried out? 2. For a given spanning tree T ½ N, what is the minimal value of T assuming multi-gossiping? 1. ¸ is independent of the order in which the gossips are carried out. 2. The minimal value of T is the chromatic index of T which for a tree is the degree of the tree.

Uniformly Aperiodically Complete Multi-Gossip Sequences Coefficient of Ergodicity ²(S) <1 iff S is a “scrambling matrix” A stochastic matrix is a scrambling matrix if no two rows are orthogonal. Any compact set C of scrambling matrices is convergable Convergence rate:

A compact subset of the set of stochastic matrices whose graphs have self arcs at all vertices is convergable iff the graphs of its matrices are all rooted. A compact subset of the set of doubly stochastic matrices whose graphs have self arcs at all vertices is convergable iff the graphs of its matrices are all weakly connected. Are there analogs of the coefficient of ergodicity for stochastic matrices with rooted graphs or for doubly stochastic matrices with weakly connected graphs? Any compact set C of scrambling matrices is convergable

Semi-norms of A 2 Rm£n For any p define Satisfies the triangle inequality and is thus a semi-norm Is sub-multiplicative for the set of all A satisfying A1 =1 For any doubly stochastic matrix Sn£n |S|p · 1, p=1,2 Let q be the integer quotient of n divided by 2. Then |S|1 <1 iff the number of nonzero entries in each column of S exceeds q. |S|2 = the second largest singular value of S.

A finite sequence of multi-gossips determines a subgraph M of the groups’ neighbor graph N where (i, j) is an edge in M iff (i, j) is a gossip in the sequence. A finite sequence of multi-gossips is complete if the sub-graph it determines is a spanning tree of N. A finite sequence of multi-gossips is complete if and only if the graph of the stochastic matrix it determines is weakly connected. A finite sequence of multi-gossips is complete if and only if the stochastic matrix S it determines, satisfies |S|2 < 1. |S|2 = the second largest singular value of S.

Uniformly Aperiodically Complete Multi-Gossip Sequences Assume that: Each agent gossips with at most one neighbor at one time. At any time t, there are no “uncommitted” agent pairs. An agent which has gossiped with neighbor i, gossips with all of its other neighbors exactly once before it again gossips with neighbor i If N is a graph of degree d, then a complete gossip sequence is achieved in at most 2d - 1 steps.

Uniformly Aperiodically Complete Multi-Gossip Sequences Assume N is a graph of degree d. The convergence rate of a uniformly aperiodically complete multi-gossip sequence is no slower than T = 2d -1 where: C = set of all complete multi-gossip sequences of length at most T ¸¾ = the second largest singular values of the stochastic matrix determined by multi-gossip sequence ¾2C