Alan Kuhnle*, Victoria G. Crawford, and My T. Thai

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

Lecture 15. Graph Algorithms
Bipartite Matching, Extremal Problems, Matrix Tree Theorem.
Efficient access to TIN Regular square grid TIN Efficient access to TIN Let q := (x, y) be a point. We want to estimate an elevation at a point q: 1. should.
Greedy Algorithms Clayton Andrews 2/26/08. What is an algorithm? “An algorithm is any well-defined computational procedure that takes some value, or set.
Lecture 24 Coping with NPC and Unsolvable problems. When a problem is unsolvable, that's generally very bad news: it means there is no general algorithm.
Introduction to Algorithms
Approximating Maximum Subgraphs Without Short Cycles Guy Kortsarz Join work with Michael Langberg and Zeev Nutov.
 Graph Graph  Types of Graphs Types of Graphs  Data Structures to Store Graphs Data Structures to Store Graphs  Graph Definitions Graph Definitions.
Lectures on Network Flows
K-structure, Separating Chain, Gap Tree, and Layered DAG Presented by Dave Tahmoush.
Approximating Maximum Edge Coloring in Multigraphs
1 Discrete Structures & Algorithms Graphs and Trees: II EECE 320.
Introduction to Approximation Algorithms Lecture 12: Mar 1.
Instructor Neelima Gupta Table of Contents Lp –rounding Dual Fitting LP-Duality.
Approximation Algorithms
An introduction to Approximation Algorithms Presented By Iman Sadeghi.
1 Introduction to Approximation Algorithms Lecture 15: Mar 5.
Lecture 11. Matching A set of edges which do not share a vertex is a matching. Application: Wireless Networks may consist of nodes with single radios,
Approximation Algorithms: Bristol Summer School 2008 Seffi Naor Computer Science Dept. Technion Haifa, Israel TexPoint fonts used in EMF. Read the TexPoint.
1 Shortest Path Calculations in Graphs Prof. S. M. Lee Department of Computer Science.
Analysis of Algorithms
V. V. Vazirani. Approximation Algorithms Chapters 3 & 22
Primal-Dual Meets Local Search: Approximating MST’s with Non-uniform Degree Bounds Author: Jochen Könemann R. Ravi From CMU CS 3150 Presentation by Dan.
1 Introduction to Approximation Algorithms. 2 NP-completeness Do your best then.
A Fast Algorithm for Enumerating Bipartite Perfect Matchings Takeaki Uno (National Institute of Informatics, JAPAN)
WAN technologies and routing Packet switches and store and forward Hierarchical addresses, routing and routing tables Routing table computation Example.
1 Introduction to Approximation Algorithms. 2 NP-completeness Do your best then.
Lecture 16 Maximum Matching. Incremental Method Transform from a feasible solution to another feasible solution to increase (or decrease) the value of.
Graph Algorithms. Definitions and Representation An undirected graph G is a pair (V,E), where V is a finite set of points called vertices and E is a finite.
DATA MINING LECTURE 13 Pagerank, Absorbing Random Walks Coverage Problems.
1/52 Overlapping Community Search Graph Data Management Lab, School of Computer Science
Christopher Moh 2005 Competition Programming Analyzing and Solving problems.
Data Structures & Algorithms Graphs
Graph Colouring L09: Oct 10. This Lecture Graph coloring is another important problem in graph theory. It also has many applications, including the famous.
Union-Find  Application in Kruskal’s Algorithm  Optimizing Union and Find Methods.
Lecture.6. Table of Contents Lp –rounding Dual Fitting LP-Duality.
Practical Message-passing Framework for Large-scale Combinatorial Optimization Inho Cho, Soya Park, Sejun Park, Dongsu Han, and Jinwoo Shin KAIST 2015.
and 6.855J March 6, 2003 Maximum Flows 2. 2 Network Reliability u Communication Network u What is the maximum number of arc disjoint paths from.
Graphs Upon completion you will be able to:
Algorithm Design and Analysis (ADA)
Approximating Buy-at-Bulk and Shallow-Light k-Steiner Trees Mohammad T. Hajiaghayi (CMU) Guy Kortsarz (Rutgers) Mohammad R. Salavatipour (U. Alberta) Presented.
The minimum cost flow problem. Solving the minimum cost flow problem.
1 The instructor will be absent on March 29 th. The class resumes on March 31 st.
::Network Optimization:: Minimum Spanning Trees and Clustering Taufik Djatna, Dr.Eng. 1.
Cohesive Subgraph Computation over Large Graphs
Finding Dense and Connected Subgraphs in Dual Networks
Computational Molecular Biology
Introduction to Approximation Algorithms
Shortest Path Problems
Introduction to Algorithms
An introduction to Approximation Algorithms Presented By Iman Sadeghi
Parallel Programming By J. H. Wang May 2, 2017.
The minimum cost flow problem
Algorithms for hard problems
Lectures on Network Flows
Greedy Algorithms / Minimum Spanning Tree Yin Tat Lee
Chapter 5. Optimal Matchings
James B. Orlin Presented by Tal Kaminker
The Importance of Communities for Learning to Influence
Graphs & Graph Algorithms 2
Lecture 16 Maximum Matching
Graphs Chapter 11 Objectives Upon completion you will be able to:
Coverage Approximation Algorithms
Linear Programming Duality, Reductions, and Bipartite Matching
Graphs.
3.3 Network-Centric Community Detection
Chapter 9 Graph algorithms
Presentation transcript:

Alan Kuhnle*, Victoria G. Crawford, and My T. Thai Scalable and Adaptive Algorithms for the Triangle Interdiction Problem on Billion-Scale Networks Alan Kuhnle*, Victoria G. Crawford, and My T. Thai Email: kuhnle@ufl.edu November 20, 2017 IEEE ICDM 2017, New Orleans, LA * Presenter

Network Transitivity 3 × Number of Triangles / Number of Connected Triplets Transitivity: 0.571 Figure degrades all triangles in multiplex by removing one edge from each layer. The smaller the objective, the less reliable the network is.

Transitivity and Reliability Transitivity allows redundant paths Bottom two nodes want to communicate

Transitivity and Social Influence Complex propagation Empirical studies (e.g. Centola1) link transitivity and influence 1 Centola, D. (2010). The Spread of Behavior in an Online Social Network Experiment. Science, 329(5996), 1194–1197.

Triangle Interdiction Problem Transitivity degradation by edge removal Triangle Interdiction Problem (TIP) Collection of triangles Triangles after removal of S

Triangle Interdiction Problem Transitivity: 0 Transitivity: 0.571

TIP Challenges TIP is an NP-hard problem Kortsarz1 introduced a 2-approximation algorithm for TIP Requires up to |E| LP solutions Does not scale to large graphs Only for static graphs Primal-dual algorithm List all triangles, iterate through them. If current triangle is unbroken, remove all of its edges. Efficient 3-approximation but performs close to its worst-case guarantee Modern networks are very large and dynamic 1 Guy Kortsarz. Approximating Maximum Subgraphs without Short Cycles. Siam Journal of Discrete Math, 24(1):255–269, 2010.

Our contributions TARL, a 5/2 approximation that solves one LP Provides speed-up of up to |E| over Kortsarz. Better worst-case guarantee than primal-dual DART, fully dynamic 3-approximation Usually within factor of 1.5 of optimal, better on sparse graphs Efficiently update its solution upon edge / vertex insertion / deletion Average update time on the order of a few microseconds Does not need to store triangle list in memory Runs on network with 4 billion triangles in only 37 GB memory

TIP IP IP formulation of TIP:

TARL Description List all triangles in G Construct and solve LP 1 optimally -> 𝑥 ∗ Break triangles containing a “low LP-weight” edge: For each 𝑒∈𝐸 If 𝑥 ∗ 𝑒 <1/5 For each unbroken triangle containing e, add edge with largest 𝑥 ∗ value into the solution. Break all other triangles: Construct large (at least ½ of edges) bipartite subgraph of residual graph Remove the complement of this graph

TARL TARL has performance guarantee of 5/2 It requires the solution of exactly one LP So provides speed-up over 2-approximation of Kortsarz of up to |E| However, the LP solution is still a large bottleneck

DART Overview Scalable algorithm with competitive ratio of 3 Maintains list of disjoint triangles 𝑈 Solution 𝑆⊆𝑈 Efficient dynamic data structure enables fast updates Vertices Edges Triangles (in 𝑈) Pointers from Edges to Triangles tell which triangle in U the edge belongs to. Pointers from Triangles back to edges tell which edges form the triangle. Pointers from Vertices to Edges tell which Edges are incident with that vertex. Easy to update since all three of these are adjacency lists K is number of layers in multiplex

DART Overview Visualization 𝑈 is maximal Disjoint triangles, 𝑈 Solution 𝑆⊆𝑈 (red edges) 𝑊=𝑈\S (green edges) 𝑈 is maximal 𝑊 comprises all edges not forced to be in 𝑆 by adjacent triangles

DART Overview All DART methods rely on the following two procedures: The AUGMENT procedure Takes an input a single edge e Ensures feasibility with respect to triangles containing e Maintains disjoint triangle list U The PRUNE procedure DART DART-BASE DART-ADD DART-REMOVE

DART Overview DART-BASE calls AUGMENT on each edge 𝑒 DART-ADD calls AUGMENT on newly inserted edge 𝑒 DART-REMOVE Determines which triangle 𝑇 in 𝑈 an edge 𝑒 belongs to, if any Removes 𝑇= {𝑒,𝑓,𝑔} from 𝑆∪𝑊 Calls AUGMENT on 𝑓,𝑔 All three DART procedures conclude with a call to PRUNE

DART Overview AUGMENT takes as input an edge e Ensures 𝑆 hits all triangles containing e Avoids adding new triangles to U if possible Otherwise, adds new disjoint triangles to 𝑈

DART Overview Have the following performance guarantee for DART

Results DART returns solution close to OPT (compared on small ER network( 30 nodes)) at much faster running times (here TARL is an alg. We formulated but unscalable).

Visualization on small ER graph

Results on smaller traces

DART Scalability In practice, DART is capable of running on networks with billions of triangles in under 2 hours It can update its solution in microseconds |S| is size of solution. Mem is max. memory used. Dynamic times are averaged over 1 million edge additions /removals. The size increase / decrease is how much solution changed after adding /removing all 1 million edges. Repeated 10 times and averaged over all.

Conclusions Provided two algorithms, TARL and DART, to approximate TIP: find minimum-size set of edges whose removal breaks all triangles. TARL Performance guarantee of 5/2 Provides significant speed-up over Kortsarz (up to 103 times faster) DART Competitive guarantee of 3 Performs close to optimal on sparse networks Fully dynamic Thank you! Questions? Contact: kuhnle@ufl.edu