Lasserre Hierarchy, Higher Eigenvalues and Approximation Schemes for Graph Partitioning and PSD QIP Ali Kemal Sinop (joint work with Venkatesan Guruswami)

Slides:



Advertisements
Similar presentations
Geometry and Expansion: A survey of some results Sanjeev Arora Princeton ( touches upon: S. A., Satish Rao, Umesh Vazirani, STOC04; S. A., Elad Hazan,
Advertisements

Network Design with Degree Constraints Guy Kortsarz Joint work with Rohit Khandekar and Zeev Nutov.
Routing in Undirected Graphs with Constant Congestion Julia Chuzhoy Toyota Technological Institute at Chicago.
Cuts, Trees, and Electrical Flows Aleksander Mądry.
Approximation algorithms for geometric intersection graphs.
Compressing Forwarding Tables Ori Rottenstreich (Technion, Israel) Joint work with Marat Radan, Yuval Cassuto, Isaac Keslassy (Technion, Israel) Carmi.
How to Round Any CSP Prasad Raghavendra University of Washington, Seattle David Steurer, Princeton University (In Principle)
Submodular Set Function Maximization via the Multilinear Relaxation & Dependent Rounding Chandra Chekuri Univ. of Illinois, Urbana-Champaign.
The Unique Games Conjecture and Graph Expansion School on Approximability, Bangalore, January 2011 Joint work with S Prasad Raghavendra Georgia Institute.
Subexponential Algorithms for Unique Games and Related Problems School on Approximability, Bangalore, January 2011 David Steurer MSR New England Sanjeev.
Inapproximability of MAX-CUT Khot,Kindler,Mossel and O ’ Donnell Moshe Ben Nehemia June 05.
Approximability & Sums of Squares Ryan O’Donnell Carnegie Mellon.
Satyen Kale (Yahoo! Research) Joint work with Sanjeev Arora (Princeton)
Graph Partitioning Problems Lecture 18: March 14 s1 s3 s4 s2 T1 T4 T2 T3 s1 s4 s2 s3 t3 t1 t2 t4 A region R1 R2 C1 C2.
Approximation Algorithms
Approximation Algorithms Chapter 14: Rounding Applied to Set Cover.
Approximation Algoirthms: Graph Partitioning Problems Lecture 17: March 16 s1 s3 s4 s2 T1 T4 T2 T3.
Online Social Networks and Media. Graph partitioning The general problem – Input: a graph G=(V,E) edge (u,v) denotes similarity between u and v weighted.
Optimal Column-Based Low-Rank Matrix Reconstruction SODA’12 Ali Kemal Sinop Joint work with Prof. Venkatesan Guruswami.
Approximation Algorithms for Unique Games Luca Trevisan Slides by Avi Eyal.
Approximation Algorithms for Non-Uniform Buy-at-Bulk Network Design Problems Mohammad R. Salavatipour Department of Computing Science University of Alberta.
Introduction to Approximation Algorithms Lecture 12: Mar 1.
Approximation Algoirthms: Semidefinite Programming Lecture 19: Mar 22.
Yi Wu IBM Almaden Research Joint work with Alina Ene and Jan Vondrak.
A Linear Round Lower Bound for Lovasz-Schrijver SDP relaxations of Vertex Cover Grant Schoenebeck Luca Trevisan Madhur Tulsiani UC Berkeley.
Sparsest Cut S S  G) = min |E(S, S)| |S| S µ V G = (V, E) c- balanced separator  G) = min |E(S, S)| |S| S µ V c |S| ¸ c ¢ |V| Both NP-hard.
Semidefinite Programming
1 Optimization problems such as MAXSAT, MIN NODE COVER, MAX INDEPENDENT SET, MAX CLIQUE, MIN SET COVER, TSP, KNAPSACK, BINPACKING do not have a polynomial.
Approximation Algorithms
Job Scheduling Lecture 19: March 19. Job Scheduling: Unrelated Multiple Machines There are n jobs, each job has: a processing time p(i,j) (the time to.
Prize Collecting Cuts Daniel Golovin Carnegie Mellon University Lamps of ALADDIN 2005 Joint work with Mohit Singh & Viswanath Nagarajan.
Colored Necklace Bisection Victor Kostyuk Advisor: Michael Capalbo.
Graph-Cut Algorithm with Application to Computer Vision Presented by Yongsub Lim Applied Algorithm Laboratory.
(work appeared in SODA 10’) Yuk Hei Chan (Tom)
Tight Integrality Gaps for Lovász-Schrijver LP relaxations of Vertex Cover Grant Schoenebeck Luca Trevisan Madhur Tulsiani UC Berkeley.
Finding Almost-Perfect
Dana Moshkovitz, MIT Joint work with Subhash Khot, NYU.
Outline Introduction The hardness result The approximation algorithm.
Integrality Gaps for Sparsest Cut and Minimum Linear Arrangement Problems Nikhil R. Devanur Subhash A. Khot Rishi Saket Nisheeth K. Vishnoi.
Domain decomposition in parallel computing Ashok Srinivasan Florida State University COT 5410 – Spring 2004.
1 Introduction to Approximation Algorithms. 2 NP-completeness Do your best then.
Edge Covering problems with budget constrains By R. Gandhi and G. Kortsarz Presented by: Alantha Newman.
Approximating Minimum Bounded Degree Spanning Tree (MBDST) Mohit Singh and Lap Chi Lau “Approximating Minimum Bounded DegreeApproximating Minimum Bounded.
1 Introduction to Approximation Algorithms. 2 NP-completeness Do your best then.
Princeton University COS 423 Theory of Algorithms Spring 2001 Kevin Wayne Approximation Algorithms These lecture slides are adapted from CLRS.
1 On Completing Latin Squares Iman Hajirasouliha Joint work with Hossein Jowhari, Ravi Kumar, and Ravi Sundaram.
Yuan Zhou Carnegie Mellon University Joint works with Boaz Barak, Fernando G.S.L. Brandão, Aram W. Harrow, Jonathan Kelner, Ryan O'Donnell and David Steurer.
Learning Spectral Clustering, With Application to Speech Separation F. R. Bach and M. I. Jordan, JMLR 2006.
1/19 Minimizing weighted completion time with precedence constraints Nikhil Bansal (IBM) Subhash Khot (NYU)
Subexponential Algorithms for Unique Games and Related Problems Approximation Algorithms, June 2011 David Steurer MSR New England Sanjeev Arora Princeton.
Lecture.6. Table of Contents Lp –rounding Dual Fitting LP-Duality.
Unique Games Approximation Amit Weinstein Complexity Seminar, Fall 2006 Based on: “Near Optimal Algorithms for Unique Games" by M. Charikar, K. Makarychev,
Maximizing Symmetric Submodular Functions Moran Feldman EPFL.
Yuan Zhou, Ryan O’Donnell Carnegie Mellon University.
Boaz Barak (MSR New England) Fernando G.S.L. Brandão (Universidade Federal de Minas Gerais) Aram W. Harrow (University of Washington) Jonathan Kelner (MIT)
TU/e Algorithms (2IL15) – Lecture 12 1 Linear Programming.
Instructor: Shengyu Zhang 1. Optimization Very often we need to solve an optimization problem.  Maximize the utility/payoff/gain/…  Minimize the cost/penalty/loss/…
Approximation Algorithms based on linear programming.
Yuan Zhou Carnegie Mellon University Joint works with Boaz Barak, Fernando G.S.L. Brandão, Aram W. Harrow, Jonathan Kelner, Ryan O'Donnell and David Steurer.
Introduction to Approximation Algorithms
Finding Almost-Perfect
Polynomial integrality gaps for
Generalized Sparsest Cut and Embeddings of Negative-Type Metrics
Approximation Algorithms
Bo-Young Kim Applied Algorithm Lab, KAIST 月
Sum of Squares, Planted Clique, and Pseudo-Calibration
Approximation Algorithms
Structural Properties of Low Threshold Rank Graphs
My relationship with correlation clustering started in 2016
Embedding Metrics into Geometric Spaces
Presentation transcript:

Lasserre Hierarchy, Higher Eigenvalues and Approximation Schemes for Graph Partitioning and PSD QIP Ali Kemal Sinop (joint work with Venkatesan Guruswami) Carnegie Mellon University

Outline Introduction – Sample Problem: Minimum Bisection – Approximation Algorithms – Our Motivation and Results Overview Results – Graph Spectrum – Related Work and Our Results Case Study: Minimum Bisection – Lasserre Hierarchy Formulation – Rounding Algorithm – Analysis 24:49 PM

Minimum Bisection Given graph G=(V,E,W), find subset of size n/2 which cuts as few edges as possible. Canonical problem for graph partitioning by allowing arbitrary size: – Small Set Expansion (weight each node by its degree) – Uniform Sparsest Cut (try out all partition sizes in small increments) – Etc… NP-hard Cost=2 µ 4:49 PM

Approximation Algorithms Find an α-factor approximation. – If minimum cost = OPT, Algorithm always finds a solution with value α OPT. (This work) Round a convex relaxation. OPTAlgorithmα OPT 0 Relaxation 4 1 4:49 PM

Motivation For many graph partitioning problems (including minimum bisection), huge gap between hardness and approximation results. Best known algorithms have factor Whereas no 1.1 factor hardness is known. We want to close the gap. 54:49 PM

Our Results: Overview For graph partitioning problems including: – Minimum bisection, – Small set expansion, – Uniform sparsest cut, – Minimum uncut, – Their k-way generalizations, etc… We give approximation schemes whose running time is dependent on graph spectrum. 64:49 PM

Outline Introduction – Sample Problem: Minimum Bisection – Approximation Algorithms – Our Motivation and Results Overview Results – Graph Spectrum – Related Work and Our Results Case Study: Minimum Bisection – Lasserre Hierarchy Formulation – Rounding Algorithm – Analysis 74:49 PM

Graph Spectrum and Eigenvalues rows and cols indexed by V 8 λ 2 : Measures expansion of the graph through Cheegers inequality. λ r : Related to small set expansion [Arora, Barak, Steurer10], [Gharan, Trevisan11]. 0 = λ 1 λ 2 … λ n 2 and λ 1 + λ 2 + … + λ n = n, 4:49 PM

Related Previous Work (Minimization form of) Unique Games (k- labeling with permutation constraints): – [AKKSTV08], [Makarychev, Makarychev10] Constant factor approximation for Unique Games on expanders in polynomial time. – [Kolla10] Constant factor when λ r is large. [Arora, Barak, Steurer10] – For Unique Games and Small Set Expansion, factor in time – For Sparsest Cut, factor assuming 94:49 PM

Our Results (1) In time we obtain Why approximation scheme? – 0 = λ 1 λ 2 … λ n 2 and λ 1 + λ 2 + … + λ n = n, 10 Minimum Bisection* Small Set Expansion* Uniform Sparsest Cut Their k-way generalizations* Minimum Bisection* Small Set Expansion* Uniform Sparsest Cut Their k-way generalizations* Independent Set * Satisfies constraints within factor of For r=n, λ r >1, λ n-r <1 Minimum Uncut 4:49 PM

Our Results for Unique Games For Unique Games, a direct bound will involve spectrum of lifted graph, whereas we want to bound using spectrum of original graph. – We give a simple embedding and work directly on the original graph. We obtain factor in time. Concurrent to our work, [Barak, Steurer, Raghavendra11] obtained factor in time using a similar rounding. 114:49 PM

Outline Introduction – Sample Problem: Minimum Bisection – Approximation Algorithms – Our Motivation and Results Overview Results – Graph Spectrum – Related Work and Our Results Case Study: Minimum Bisection – Lasserre Hierarchy – Rounding Algorithm – Analysis 124:49 PM

Case Study: Minimum Bisection We will present an approximation algorithm for minimum bisection problem on d-regular unweighted graphs. We will show that it achieves factor. Obtaining factor requires some additional ideas. 134:49 PM

Lasserre Hierarchy Basic idea: Rounding a convex relaxation of minimum bisection. [Lasserre01] Strongest known SDP-relaxation. – (Relaxation of) For each subset S of size r and each possible labeling of S, – An indicator vector which is 1 if S is labeled with f – 0 else. And all implied consistency constraints. 144:49 PM

Previous Work on Lasserre Hierarchy Few algorithmic results known before, including: – [Chlamtac07], [Chlamtac, Singh08] n(1) approximation for 3-coloring and independent set on 3-uniform hypergraphs, – [Karlin, Mathieu, Nguyen10] (1+1/r) approximation of knapsack for r-rounds. Known integrality gaps are: – [Schoenebeck08], [Tulsiani09] Most NP-hardness results carry over to (n) rounds of Lasserre. – [Guruswami, S, Zhou11] Factor (1+α) integrality gap for(n) rounds of min-bisection and max-cut. Not ruled out yet: 5-rounds of Lasserre relaxation disproves Unique Games Conjecture. 154:49 PM

Why So Few Positive Results? For regular SDP [Goemans, Williamson95] showed that with hyperplane rounding: Prior to our work, no analogue for Lasserre solution. 164:49 PM

Consistency Lasserre Relaxation for Minimum Bisection 17 Relaxation for consistent labeling of all subsets of size < r: Marginalization Distribution Partition SIze Cut cost 4:49 PM

Rounding Algorithm Choose S with probability – [Deshpande, Rademacher, Vempala, Wang06] Volume sampling. Label S by choosing f with probability. Propagate to other nodes: – For each node v, With probability include v in U. – Inspired by [AKKSTTV08] which used propagation from a single node chosen uniformly at random. Return U. 4:49 PM18

Analysis Partition Size – Each node is chosen into U independently – By Chernoff, with high probability Number of Edges Cut – After arithmetization, we have the following bound: 19 Normalized Vector for x S (f) OPT 19 4:49 PM

Matrix Π S Remember {x S (f)} f are orthogonal. is a projection matrix onto span{x S (f)} f. For any 20 Let P S be the corresponding projection matrix. 4:49 PM

Low Rank Matrix Reconstruction The final bound is: For any S of size r this is lower bounded by: [Guruswami, S11] Volume sampling columns yield – And this bound is tight. best rank-r approximation of X 214:49 PM

Relating Reconstruction Error to Graph Spectrum Best rank-r approximation is obtained by top r-eigenvectors. Using Courant-Fischer theorem, Therefore 224:49 PM

Summary Gave a randomized rounding algorithm based on propagation from a seed set S so that: Related choosing S to low rank matrix reconstruction error. Bounded low rank matrix reconstruction error in terms of λ r. 234:49 PM

Questions? Thanks. 244:49 PM