On Boltzmann samplers and properties of combinatorial structures joint work with Nicla Bernasconi & Kostas Panagiotou TexPoint fonts used in EMF. Read.

Slides:



Advertisements
Similar presentations
The Primal-Dual Method: Steiner Forest TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A AA A A A AA A A.
Advertisements

Vertex sparsifiers: New results from old techniques (and some open questions) Robert Krauthgamer (Weizmann Institute) Joint work with Matthias Englert,
CS 336 March 19, 2012 Tandy Warnow.
Great Theoretical Ideas in Computer Science for Some.
Planar Graphs: Coloring and Drawing
C&O 355 Lecture 23 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A A A A A A A A.
Divide and Conquer. Subject Series-Parallel Digraphs Planarity testing.
Map Overlay Algorithm. Birch forest Wolves Map 1: Vegetation Map 2: Animals.
2/14/13CMPS 3120 Computational Geometry1 CMPS 3120: Computational Geometry Spring 2013 Planar Subdivisions and Point Location Carola Wenk Based on: Computational.
Approximation, Chance and Networks Lecture Notes BISS 2005, Bertinoro March Alessandro Panconesi University La Sapienza of Rome.
Gibbs sampler - simple properties It’s not hard to show that this MC chain is aperiodic. Often is reversible distribution. If in addition the chain is.
Theory of Computing Lecture 3 MAS 714 Hartmut Klauck.
Graph Isomorphism Algorithms and networks. Graph Isomorphism 2 Today Graph isomorphism: definition Complexity: isomorphism completeness The refinement.
Small Subgraphs in Random Graphs and the Power of Multiple Choices The Online Case Torsten Mütze, ETH Zürich Joint work with Reto Spöhel and Henning Thomas.
Minimum Spanning Trees Definition Two properties of MST’s Prim and Kruskal’s Algorithm –Proofs of correctness Boruvka’s algorithm Verifying an MST Randomized.
Theory of Computing Lecture 6 MAS 714 Hartmut Klauck.
CSE 20 – Discrete Mathematics Dr. Cynthia Bailey Lee Dr. Shachar Lovett Peer Instruction in Discrete Mathematics by Cynthia Leeis licensed under a Creative.
1 The Monte Carlo method. 2 (0,0) (1,1) (-1,-1) (-1,1) (1,-1) 1 Z= 1 If  X 2 +Y 2  1 0 o/w (X,Y) is a point chosen uniformly at random in a 2  2 square.
CS774. Markov Random Field : Theory and Application Lecture 17 Kyomin Jung KAIST Nov
Random Walks Ben Hescott CS591a1 November 18, 2002.
Entropy Rates of a Stochastic Process
1 List Coloring and Euclidean Ramsey Theory TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A A Noga Alon, Tel Aviv.
Graph Sparsifiers by Edge-Connectivity and Random Spanning Trees Nick Harvey University of Waterloo Department of Combinatorics and Optimization Joint.
Coloring the edges of a random graph without a monochromatic giant component Reto Spöhel (joint with Angelika Steger and Henning Thomas) TexPoint fonts.
1 Algorithms for Large Data Sets Ziv Bar-Yossef Lecture 8 May 4, 2005
On the power of choices in random graph processes Reto Spöhel, PhD Defense February 17, 2010, ETH Zürich Examiners: Prof. Dr. Angelika Steger, ETH Zürich.
Avoiding Monochromatic Giants in Edge-Colorings of Random Graphs Henning Thomas (joint with Reto Spöhel, Angelika Steger) TexPoint fonts used in EMF. Read.
Advanced Topics in Data Mining Special focus: Social Networks.
Sampling and Approximate Counting for Weighted Matchings Roy Cagan.
Small Subgraphs in Random Graphs and the Power of Multiple Choices The Online Case Torsten Mütze, ETH Zürich Joint work with Reto Spöhel and Henning Thomas.
Quantum Algorithms II Andrew C. Yao Tsinghua University & Chinese U. of Hong Kong.
Computer Science 1 Web as a graph Anna Karpovsky.
Approximation Algorithms: Bristol Summer School 2008 Seffi Naor Computer Science Dept. Technion Haifa, Israel TexPoint fonts used in EMF. Read the TexPoint.
Graph Theory Ch6 Planar Graphs. Basic Definitions  curve, polygon curve, drawing  crossing, planar, planar embedding, and plane graph  open set  region,
UNC Chapel Hill M. C. Lin Point Location Reading: Chapter 6 of the Textbook Driving Applications –Knowing Where You Are in GIS Related Applications –Triangulation.
Orthogonal Drawings of Series-Parallel Graphs Joint work with Xiao Zhou by Tohoku University Takao Nishizeki.
Small subgraphs in the Achlioptas process Reto Spöhel, ETH Zürich Joint work with Torsten Mütze and Henning Thomas TexPoint fonts used in EMF. Read the.
Planar Graphs: Euler's Formula and Coloring Graphs & Algorithms Lecture 7 TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.:
Edge-disjoint induced subgraphs with given minimum degree Raphael Yuster 2012.
Approximation Algorithms
10 December, 2008 CIMCA2008 (Vienna) 1 Statistical Inferences by Gaussian Markov Random Fields on Complex Networks Kazuyuki Tanaka, Takafumi Usui, Muneki.
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, …
C&O 355 Mathematical Programming Fall 2010 Lecture 16 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A.
Approximate Inference: Decomposition Methods with Applications to Computer Vision Kyomin Jung ( KAIST ) Joint work with Pushmeet Kohli (Microsoft Research)
Seminar on random walks on graphs Lecture No. 2 Mille Gandelsman,
C&O 355 Lecture 24 N. Harvey TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA A A A A A A A A.
A Linear Time Algorithm for the Longest Path Problem on 2-trees joint work with Tzvetalin Vassilev and Krassimir Manev
PERTEMUAN 26. Markov Chains and Random Walks Fundamental Theorem of Markov Chains If M g is an irreducible, aperiodic Markov Chain: 1. All states are.
Great Theoretical Ideas in Computer Science for Some.
Avoiding small subgraphs in the Achlioptas process Torsten Mütze, ETH Zürich Joint work with Reto Spöhel and Henning Thomas TexPoint fonts used in EMF.
Indian Institute of Technology Kharagpur PALLAB DASGUPTA Graph Theory: Trees Pallab Dasgupta, Professor, Dept. of Computer Sc. and Engineering, IIT
Theory of Computational Complexity Probability and Computing Lee Minseon Iwama and Ito lab M1 1.
Theory of Computational Complexity Probability and Computing Ryosuke Sasanuma Iwama and Ito lab M1.
An algorithmic proof of the Lovasz Local Lemma via resampling oracles Jan Vondrak IBM Almaden TexPoint fonts used in EMF. Read the TexPoint manual before.
Visual Recognition Tutorial1 Markov models Hidden Markov models Forward/Backward algorithm Viterbi algorithm Baum-Welch estimation algorithm Hidden.
Graphs and Algorithms (2MMD30)
Sequential Algorithms for Generating Random Graphs
CSE 2331/5331 Topic 9: Basic Graph Alg.
Lecture 18: Uniformity Testing Monotonicity Testing
Algorithms and Complexity
MST in Log-Star Rounds of Congested Clique
Structural graph parameters Part 2: A hierarchy of parameters
Haim Kaplan and Uri Zwick
Bart M. P. Jansen June 3rd 2016, Algorithms for Optimization Problems
GRAPH SPANNERS.
CSE 373 Data Structures and Algorithms
Introduction Wireless Ad-Hoc Network
Advanced Topics in Data Mining Special focus: Social Networks
Treewidth meets Planarity
Presentation transcript:

On Boltzmann samplers and properties of combinatorial structures joint work with Nicla Bernasconi & Kostas Panagiotou TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AAAAA

Random Graphs Paul Erdős, Alfred Rényi On the evolution of random graphs Publ. Math. Inst. Int. Hungar. Acad. Sci., 1960 Given a set of vertices. Decide for each potential edge randomly whether edge is present in graph. edge probability p → random graph G n,p Key property: Independence of edges.

Random Partial Orders Various models were proposed and studied, see e.g.

Random Planar Graphs Colin McDiarmid, AS, Dominic Welsh Random planar graphs Journal of Combinatorial Theory, Series B, 2005  c, C: 0 < c < Prob[P n connected ] < C < 1 P n := set of all planar graphs on n (labelled) vertices P n := graph drawn randomly from P n ( → random planar graph)

Connectedness – Proof Idea Direct approach: Counting... Prob[P n connected ] = Later... [ Giménez, Noy, ] # connected planar graphs on n vertices # planar graphs on n vertices

Connectedness – Proof Idea Combinatorial approach: Markov chain... start with empty graph: G := (V, ϕ ) repeat forever: –pick two vertices u,v from V uniformly at random –if {u,v} in G: delete edge {u,v} –if {u,v} not in G and G + {u,v} planar: insert edge {u,v} –otherwise: do nothing Stationary Distribution: uniform distribution on set of all planar graphs

Connectedness – Proof Idea Combinatorial approach: Markov chain... Stationary Distribution: uniform distribution on set of all planar graphs # edges in graph ≈ c 1 n # edges that can be added ≈ c 2 n # vertices of degree one ≈ α 1 n crude counting

Questions Combinatorics: Number of (planar) graphs on n vertices Properties of random (planar) graph P n : Expected number of edges Degree distribution Expected number of substructures (of a given type)... Algorithmic: Generate a random (planar) graph P n Aim of this talk: describe some recent progress towards the development of methods for answering this kind of questions.

Counting Philippe Flajolet, Robert Sedgewick Analytic Combinatorics Cambridge University Press, to appear Generating functions: G ≈ class of objects, G n ≈ objects in G of size n G(x) = Σ n≥0 | G n | x n (unlabelled objects) G(x) = Σ n≥0 x n (labelled objects) |Gn||Gn|

Generating Functions G ≈ class of (unlabelled) objects C ≈ class of connected objects in G G(x), C(x) ≈ corresponding generating functions G(x) = Σ k≥0 = e C(x) Observation: G(x) and C(x) have same radius of convergence... and thus the same growth constant C(x) k G(x) = Σ n≥0 | G n | x n

Counting Planar Graphs Tutte connected planar maps Bender, Gao, Wormald connected planar graphs Giménez, Noy connected planar graphs, general planar graphs

Giménez, Noy P ≈ class of (labelled) planar graphs | P n | = p · n −7/2 · γ n · n! where p ≈ · 10 −6 γ ≈ and for C ≈ class of connected planar graphs | C n | = c · n −7/2 · γ n · n! where c ≈ · 10 −6

Generation of Random Objects Duchon, Flajolet, Louchard, Schaeffer Boltzmann Samplers for the Random Generation of Combinatorial Structures Combinatorics, Probability and Computing, 2004 Éric Fusy Quadratic exact-size and linear approximate-size random sampling of planar graphs Analysis of Algorithms AofA’05

Boltzmann Sampler Observations: If we condition on | ΓD (x)| = n, then ΓD (x) is a uniform sampler. Expected size of the output depends on the parameter x: E [ | ΓD (x)| ] = An algorithm ΓD (x) that generates an element D  D is called Boltzmann Sampler iff x D´(x) D(x)

From Decompositions to Samplers Duchon, Flajolet, Louchard, Schaeffer, 2004

Boltzmann Sampler ΓD (x) D (  1,  2, …,  i, …) Input: sequence of  i ‘s that are independent and identically distributed Output: object D  D s.t. size of x determines expected size of ΓD(x) Idea: establish a relation between properties of the sequence (  1,  2, …,  i, …) and properties of ΓD (x).

Random Dissections D n := set of all dissections on n (labelled) vertices D n := graph drawn randomly from D n ( → random dissection)

rooted (unlabelled) dissection labelled dissection 2 ↔ (n-1)! Observation: It suffices to consider rooted unlabelled structures.

D =  D D D D D   … Generating Function D  class of all edge-rooted dissections Bodirsky, Giménez, Kang, Noy, = c  n -3/2 δ n

Boltzmann Sampler Aim: procedure ΓD (x) that generates an element D  D such that recall: Idea: - first generate face containing the root-edge - then call procedure recursively for each edge of cycle D =  D D D D D   …

Boltzmann Sampler ΓD (x) D (  1,  2, …,  i, …) Input: sequence of  i ‘s that are independent and distributed s.t. Output: random dissection D s.t.

Sampler - How does it work? path of length α 1 -1 path of length α 2 -2 path of length α k -2 What can we deduce about the degrees ? root edge path of length Σ i (α i -2) α k+1 = 2 D2D2 DiDi D Σ (αi-2) D1D1 path of length α 1 -2 degree = 1 + min{ k : α k+1 =2 }

Sampler - How does it work? degree = min{ k : α k =2 } root edge α k+1 = 2 D2D2 DiDi D Σ (αi-1) D1D1 Observations: # of used α i ‘s = # of edges of D 1 + #{ i : α i = 2 } = # of vertices of D α 1,α 2,…,α k,2,α k+2, …,2, ….,2, …, α |E(D)|-1, 2 # blocks = # vertices - 1 pre-degree

Sampler - How does it work? root edge D2D2 DiDi D Σ (αi-1) D1D1 Key observation: # pre-deg‘( D, k) = # blocks‘ of size k deg(root, D ) = size of first block do not count root vertices do not count first block

Sampler - Summary ΓD (x) D (  1,  2, …,  i, …) If |V(D)| = n then ΓD (x) used  3n of the  i ‘s. # pre-deg ( D ;k) = number of blocks of size k among the first n blocks.

A Sampler for Dissections of Size n for i = 1 to n 3 do: if |V( D i )| = n then return D i D   (  1 1,  1 2, …,  1 3n ) Easy probabilistic arguments/Chernoff: Prob[ algorithm returns D   ]  1 – e -  (n)  k: Prob[  i s.t. (  i 1,  i 2, …,  i 3n ) contains  (1  ) c k n blocks of size k among first n blocks]  e -  (n) Easy probabilistic arguments/Chernoff: Prob[ algorithm returns D   ]  1 – e -  (n)  k: Prob[  i s.t. (  i 1,  i 2, …,  i 3n ) contains  (1  ) c k n blocks of size k among first n blocks]  e -  (n) (  2 1,  2 2, …,  2 3n ) (  n 3 1,  n 3 2, …,  n 3 3n ) ΓD(ρD)ΓD(ρD) Di  Di   (  i 1,  i 2, …,  i 3n ) c k ~ geometric distribution

A Sampler for Dissections of Size n D   (  1 1,  1 2, …,  1 3n ) Properties: Prob[ algorithm returns D   ]  1 – e -  (n)  k: Prob[ #pre-deg( D, k)  (1  ) c k n]  e -  (n)  k: Prob[ #post-deg( D, k)  (1  ) c k n]  e -  (n)  k, : Prob[ #pre-deg( D, k)  (1  ) c k n & #post-deg( D, )  (1  ) c n]  e -  (n)  k: Prob[ #deg( D, k)  (1  ) d k n]  e -  (n) Properties: Prob[ algorithm returns D   ]  1 – e -  (n)  k: Prob[ #pre-deg( D, k)  (1  ) c k n]  e -  (n)  k: Prob[ #post-deg( D, k)  (1  ) c k n]  e -  (n)  k, : Prob[ #pre-deg( D, k)  (1  ) c k n & #post-deg( D, )  (1  ) c n]  e -  (n)  k: Prob[ #deg( D, k)  (1  ) d k n]  e -  (n) (  2 1,  2 2, …,  2 3n ) (  n 3 1,  n 3 2, …,  n 3 3n ) ΓD‘ (ρ D ) k = k(n) is also possible Δ(D n ) = Θ(log n) Similar results for subdisections

Extension: Outerplanar Graphs biconnected outerplanar graphs (dissections) connected outerplanar graphs outerplanar graphs McDiarmid, St., Welsh (2005): degree sequence of a random planar graph ≈ degree sequence of random planar connected graph

Let C  be the class of labelled rooted connected (outerplanar) graphs: Z is the class consisting of one single element, B  is the class of rooted biconnected (outerplanar) graphs, and B ‘ the derivated class From Biconnected to Connected

A Sampler for Rooted Connected Outerplanar Graphs Γ C  (x) C ( λ 1, λ 2, …, λ i, …) (B 1, B 2, …, B i, …) List of parameters indep. distributed according to Po( B‘ (C  (x)) ). List of vertex rooted dissections according to Γ B ‘(x). Similarly as before: If we run Γ C  (ρ C ) n 3 times then one of the graphs will have size exactly n C with probability 1 – e -  (n). Need to study properties of one run only …

A Sampler for Rooted Connected Outerplanar Graphs Intuitively :  Every vertex is born with a certain degree  It then receives a certain number of new neighbors  Every vertex is born with a certain degree  It then receives a certain number of new neighbors d = P [ born with degree ] p k- = P [ receive k- more neighbors later ]  inner vertex of dissection  Poisson many copies of a root of a dissection  inner vertex of dissection  Poisson many copies of a root of a dissection ( )

Summary and Outlook dissections and triangulations of a polygon connected outerplanar graphs → general scheme to lift properties from 2-connected graphs to 1-connected graphs 1 outerplanar graphs → same as 1-connected due to McDiarmid, St, Welsh Degree sequence/substructures of random … 2 3 Extensions: series-parallel graphs Work in progress: general scheme, other graph classes, … [Drmota, Giménez, Noy 2008+]