final program & abstracts · membership and customer service: (800) 447-7426 (us & canada)...

FINAL PROGRAM & AbstRActs

SODA is jointly sponsored by the ACM Special Interest Group on Algorithms and Computation Theory and the SIAM Activity Group on Discrete Mathematics

The SIAG on Discrete Mathematics focuses on combinatorics, graph theory, cryptography, discrete optimization, mathematical programming, coding theory, information theory, game theory, and theoretical computer science, including algorithms, complexity, circuit design, robotics, and parallel processing. This activity group provides an opportunity to unify pure discrete mathematics and areas of applied research such as computer science, operations research, combinatorics, and the social sciences. I t organizes a biennial conference on discrete mathematics; co-sponsors, with ACM SIGACT, the annual Symposium on Discrete Algorithms; and sponsors minisymposia at SIAM meetings and confer ences. The activity group also runs DM-Net, an electronic forum; publishes an electronic newsletter; and maintains a website and a member directory.

Society for Industrial and Applied Mathematics3600 Market Street, 6th Floor

Philadelphia, PA 19104-2688 USATelephone: +1-215-382-9800 Fax: +1-215-386-7999

Conference E-mail: [email protected] Conference Web: www.siam.org/meetings/

Membership and Customer Service: (800) 447-7426 (US & Canada) or +1-215-382-9800 (worldwide)

http://www.siam.org/meetings/da13/http://www.siam.org/meetings/alenex13/http://www.siam.org/meetings/analco13/

2 SODA, ALENEX and ANALCO 2013 Conference Program

table of contentsGeneral Information ............................... 2Get-togethers .......................................... 5 Invited Plenary Presentations .................. 6 Program-at-a-Glance ............................... 7 Program Schedule ................................... 8 Abstracts................................................ 23 Speaker Index ........................................ 61 Hotel Meeting Room Map ...... Back Cover

sODA cOMMIttees

Program committee chairsanjeev KhannaUniversity of Pennsylvania, USA

Program committeeGlencora borradaileOregon State University, USANiv buchbinder Open University, IsraelDeeparnab chakrabarty Microsoft Research India, IndiaXi chenColumbia University, USAJoseph cheriyanUniversity of Waterloo, CanadaMaria chudnovsky Columbia University, USAedith elkind Nanyang Technological University, Singaporecyril Gavoille University of Bordeaux, FranceAshish GoelStanford University, USANick HarveyUniversity of British Columbia, CanadaJohn Iacono Polytechnic Institute of New York University, USALap chi LauThe Chinese University of Hong Kong, Hong KongAleksander Madry EPFL, Switzerland, and Microsoft Research New England, USA

Konstantin MakarychevMicrosoft Research, Redmond, USAconrado Martínez Universitat Politècnica de Catalunya, SpainDániel Marx MTA SZTAKI, Hungarian Academy of Sciences, HungaryViswanath NagarajanIBM T.J. Watson Research Center, USAJelani NelsonPrinceton University, USA, and IAS, Princeton, USAKrzysztof OnakCarnegie Mellon University, USAKonstantinos PanagiotouLudwig-Maximilians-Universität München, Germanyseth PettieUniversity of Michigan, USAKirk PruhsUniversity of Pittsburgh, USAAaron RothUniversity of Pennsylvania, USAMohammad salavatipourUniversity of Alberta, CanadaPeter sandersKarlsruhe Institute for Technology, GermanyPiotr sankowskiUniversity of Warsaw, Polandbruce shepherdMcGill University, CanadaDaniel ŠtefankoviUniversity of Rochester, USAKavitha telikepalliTata Institute of Fundamental Research, IndiaKasturi Varadarajan University of Iowa, USAVirginia Vassilevska Williams University of California, Berkeley, and Stanford University, USAsuresh Venkatasubramanian University of Utah, USADorothea WagnerKarlsruhe Institute of Technology, Germany

steering committeeMoses charikarPrinceton University, USADavid JohnsonAT&T Labs – Research, USAclaire MathieuBrown University, USAMike MolloyUniversity of Toronto, CanadaAngelika stegerETH Zürich, Switzerland

ALeNeX cOMMIttees

Program committee co-chairsPeter sandersKarlsruhe Institute of Technology, GermanyNorbert Zeh Dalhousie University, Canada

Program committeeGraham cormode AT&T Labs Research, USADaniel Delling Microsoft Research, USARolf FagerbergUniversity of Southern Denmark, DenmarkIrene FinocchiUniversity of Rome “La Sapienza,” Italybruce Hendrickso Sandia National Laboratories, USAAlejandro López-OrtizUniversity of Waterloo, CanadaGonzalo NavarroUniversity of Chile, ChileKnut ReinertFreie Universität Berlin, GermanyMartin skutella Technische Universität Berlin, GermanyKavitha telikepalliTata Institute of Fundamental Research, Indiatakeaki UnoNational Institute of Informatics, JapanKe YiHong Kong University of Science and Technology, Hong Kong

General Information

Conference Program SODA, ALENEX and ANALCO 2013 3

steering committee chaircliff stein Columbia University, USA

steering committeeDavid ApplegateAT&T Labs, USALars ArgeAarhus University, DenmarkDavid bader Georgia Institute of Technology, USAGuy blellochCarnegie Mellon University, USAGerth stølting brodalAarhus University, DenmarkIrene FinocchiSapienza University of Rome, ItalyMike GoodrichUniversity of California, Irvine, USAJohn HershbergerMentor Graphics, USA David JohnsonAT&T Labs Research, USAbernard MoretÉcole Polytechnique Fédérale de Lausanne, France Petra Mutzel Technische Universität Dortmund, Germany Matthias Müller-HannemannMartin-Luther-Universität, Halle-Wittenberg, Germany Rajeev Raman University of Leicester, United Kingdom Dorothea WagnerKarlsruhe Institute of Technology, Germany Renato WerneckMicrosoft Research, USA

ANALcO committees

Program committee co-chairsMarkus NebelUniversity of Kaiserslautern, GermanyWojciech szpankowskiPurdue University, USA

Program committeeJacek cichonWroclaw University of Technology, PolandAmin coja-OghlanUniversity of Warwick, United KingdomMichael Drmota Technische Universität Wien, AustriaMordecai Golin Hong Kong University of Science & Technology, Hong KongAlan FriezeCarnegie Mellon University, USAHosam MahmoudThe George Washington University, USAchristian Reidys University of Southern Denmark Odense, Denmarkbruno salvy INRIA Rocquencourt, France

steering committeeMordecai GolinHong Kong University of Science & Technology, Hong KongH.K. HwangInstitute of Statistical Science, Academia Sinica, Taiwanconrado MartinezUniversitat Politècnica de Catalunya, SpainDaniel Panario Carleton University, CanadaRobert sedgewickPrinceton University, USAWojciech szpankowskiPurdue University, USA

sIAM Registration Desk The SIAM registration desk is located in the Grand Ballroom Foyer - 2nd Floor. It is open during the following hours:

Saturday, January 55:00 PM - 8:00 PM

Sunday, January 68:00 AM - 5:00 PM

Monday, January 78:00 AM - 5:00 PM

Tuesday, January 8

8:00 AM - 5:00 PM

Hotel Information Astor Crowne Plaza, French Quarter739 Canal Street at BourbonNew Orleans, Louisiana 70130Direct Telephone: +1-504-692-0500Toll Free Reservations (US and Canada): +1-888-696-4806Fax: +1-504-962-0501 Hotel web address: http://www.astorneworleans.com/

Hotel telephone NumberTo reach an attendee or to leave a message, call +1-504-692-0500. The hotel operator can either connect you with the SIAM registration desk or to the attendee’s room. Messages taken at the SIAM registration desk will be posted to the message boar d located in the registration area.

Hotel check-in and check-out timesCheck-in time is 3:00 PM.Check-out time is 12:00 PM.


child careThe Astor Crowne Plaza does not offer on-site childcare. For information regarding local childcare services, please contact the concierge for recommendations at +1-504-962-0500.

corporate Members and AffiliatesSIAM corporate members provide their employees with knowledge about, access to, and contacts in the applied mathematics and computational sciences community through their membership benefits. Corporate membership is more than just a bundle of tangible products and services; it is an expression of support for SIAM and its programs. SIAM is pleased to acknowledge its corporate members and sponsors. In recognition of their support, non-member attendees who are employed by the following organizations are entitled to the SIAM member registration rate.

Corporate/Institutional Members

The Aerospace Corporation Air Force Office of Scientific Research AT&T Laboratories - Research Bechtel Marine Propulsion Laboratory The Boeing Company CEA/DAM Department of National Defence (DND/CSEC) DSTO- Defence Science and Technology Organisation Hewlett-Packard IBM Corporation IDA Center for Communications Research, La Jolla IDA Center for Communications Research, Princeton Institute for Computational and Experimental Research in Mathematics (ICERM) Institute for Defense Analyses, Center for Computing Sciences Lawrence Berkeley National Laboratory

Lockheed Martin Los Alamos National LaboratoryMathematical Sciences Research Institute Max-Planck-Institute for Dynamics of Complex Technical Systems Mentor Graphics National Institute of Standards and Technology (NIST) National Security Agency (DIRNSA) Oak Ridge National Laboratory, managed by UT-Battelle for the Department of Energy Sandia National Laboratories Schlumberger-Doll Research Tech X Corporation U.S. Army Corps of Engineers, Engineer Research and Development Center United States Department of Energy

List current November 2012.

Leading the applied mathematics community…Join SIAM and save!

SIAM members save up to $130 on full registration for the ACM-SIAM Symposium on Discrete Algorithms (SODA13) and its associated meetings, Analytic Algorithmics and Combinatorics (ANALCO13) and Algorithm Engineering and Experiments (ALENEX13)! Join your peers in supporting the premier professional society for applied mathematicians and computational scientists. SIAM members receive subscriptions to SIAM Review and SIAM News and enjoy substantial discounts on SIAM books, journal subscriptions, and conference registrations. If you are a SIAM member, it only costs $10 to join the SIAM Activity Group on Discrete Mathematics (SIAG/DM).

Free Student Memberships are available to students who attend an institution that is an Academic Member of SIAM, are members of Student Chapters of SIAM, or are nominated by a Regular Member of SIAM.

Join onsite at the registration desk, go to www.siam.org/joinsiam to join online or download an application form, or contact SIAM Customer Service: Telephone: +1-215-382-9800 (worldwide); or 800-447-7426 (U.S. and Canada only) Fax: +1-215-386-7999 E-mail: [email protected] Postal mail: Society for Industrial and Applied Mathematics, 3600 Market Street, 6th floor, Philadelphia, PA 19104-2688 USA

standard Audio/Visual set-Up in Meeting Rooms SIAM does not provide computers for any speaker. When giving an electronic presentation, speakers must provide their own computers. SIAM is not r esponsible for the safety and security of speakers’ computers.

The Plenary Session Room will have one (1) overhead projector, two (2) screens and one (1) data projector. Cables or adaptors for Apple computers are not supplied, as they vary for each model. Please bring your own cable/adaptor if using a Mac computer. All other concurrent/breakout rooms will have one (1) screen and one (1) data projector. Cables or adaptors for Apple computers are not supplied, as they vary for each model. Please bring your own cable/adaptor if using a Mac computer. Overhead projectors will be provided only when requested.

If you have questions regarding availability of equipment in the meeting room of your presentation, or to request an overhead projector for your session, please see a SIAM staff member at the registration desk.


congratulations to the following sODA award winners!Best Papers (awarded jointly to two papers)A Simple Algorithm for the Graph Minor Decomposition - Logic meets Structural Graph Theory

Martin Grohe, Ken-Ichi Kawarabayashi and Bruce Reed

Dynamic Graph Connectivity in Polylogarithmic Worst Case Time

Bruce Kapron, Valerie King and Ben Mountjoy

Best Student Papers (awarded jointly to two papers)New Additive Spanners

Shiri Chechik

Simple, Fast and Deterministic Gossip and Rumor Spreading

Bernhard Haeupler

Recording of PresentationsAudio and video recording of presentations at SIAM meetings is prohibited without the written permission of the presenter and SIAM.

social MediaSIAM is promoting the use of social media, such as Facebook and Twitter, in order to enhance scientific discussion at its meetings and enable attendees to connect with each other prior to, during and after conferences. If you are tweeting about a conference, please use the designated hashtag to enable other attendees to keep up with the Twitter conversation and to allow better archiving of our conference discussions. The hashtag for SODA is #SIAMDA13. The hashtag for ALENEX is #ALENEX13, and ANALCO is #ANALCO13.

conference sponsorsSIAM and the conference organizing committee wish to extend their thanks and appreciation to the following sponsors for their support of this conference.

Name badgesA space for emergency contact information is provided on the back of your name badge. Help us help you in the event of an emergency!

comments?Comments about SIAM meetings are encouraged! Please send to:Sven Leyffer, SIAM Vice President for Programs ([email protected]).

Get-togethersWelcome ReceptionSaturday, January 5 6:00 PM – 8:00 PM

ALENEX/ANALCO Business Meeting

Sunday, January 68:00 PM – 9:00 PM

SODA Business Meeting and Awards Presentation (open to SIAG/DM members)Monday, January 78:00 PM – 9:00 PMComplimentary beer and wine will be served.

Please NoteSIAM is not responsible for the safety and security of attendees’ computers. Do not leave your laptop computers unattended. Please remember to turn off your cell phones, pagers, etc., during sessions.

e-mail AccessComplimentary Internet access is available in the lobby and meeting space.Conference attendees booked within the SIAM room block will also have complimentary wireless Internet access in their guest rooms. SIAM will provide a limited number of email stations for attendees during registration hours.

Registration Fee Includes• Admission to all technical sessions• ANALCO/ALENEX Business Meeting • Coffee breaks daily• Continental Breakfasts daily• Luncheon on Sunday, January 6, 2013• Proceedings (SODA CD distributed

onsite; ALENEX, ANALCO and SODA posted online)

• Room set-ups and audio/visual equipment

• SODA Business Meeting (open to SIAG/DM members)• Welcome Reception

Job PostingsPlease check with the SIAM registration desk regarding the availability of job postings or visit http://jobs.siam.org.

sIAM books and JournalsDisplay copies of books and complimentary copies of journals are available on site. SIAM books are available at a discounted price during the conference. If a SIAM books representative is not available, completed order forms and payment (credit cards are preferred) may be taken to the SIAM registration desk. The books table will close at Tuesday, January 8, at 2:00 PM.

table top DisplaysCambridge University PressPrinceton University Press SIAM


Invited Plenary speakers

** All Invited Plenary Presentations will take place in Grand Ballroom A/B/C - 2nd Floor**

sunday, January 6

11:30 AM - 12:30 PM

IP1 “If You Can Specify It, You Can Analyze It” ---

The Lasting Legacy of Philippe Flajolet

Robert sedgewick, Princeton University, USA

Monday, January 7

11:30 AM - 12:30 PM

IP2 Submodular Functions and Their Applications

Jan Vondrak, IBM Almaden Research Center, USA

tuesday, January 8

11:30 AM - 12:30 PM

IP3 On Graph Property Testing, Arithmetic Progressions and Communication

Noga Alon, Tel Aviv University, Israel


sODA, ALeNeX and ANALcO Program-at-a-Glance

Saturday, January 5

5:00 PM - 8:00 PM Registration Grand Ballroom Foyer - 2nd Floor

6:00 PM - 8:00 PM Welcome Reception Grand Ballroom A/B/C-2nd Floor

Sunday, January 6

8:00 AM - 5:00 PM Registration Grand Ballroom Foyer - 2nd Floor

8:30 AM Continental Breakfast Grand Ballroom Foyer - 2nd Floor

9:00 AM - 11:05 AM Concurrent SessionsANALCO: Session 1 St. Charles A - 1st Floor

CP1 SODA Session 1A Grand Ballroom A/B - 2nd Floor

CP2 SODA Session 1BGrand Ballroom C - 2nd Floor

CP3 SODA Session 1C Ballroom D - 2nd Floor

11:05 AM - 11:30 AM Coffee Break Grand Ballroom Foyer - 2nd Floor

11:30 AM - 12:30 PM IP1 “If You Can Specify It, You Can Analyze It” ---The Lasting Legacy of Philippe FlajoletRobert Sedgewick, Princeton University, USAGrand Ballroom A/B/C - 2nd Floor

12:30 PM - 2:00 PM Luncheon **Ticketed Event** Astor Ballroom - 2nd Floor

2:00 PM - 4:05 PM Concurrent SessionsANALCO: Session 2 St. Charles A - 1st Floor


CP5 SODA Session 2B Grand Ballroom C - 2nd Floor


4:05 PM - 4:30 PM Coffee Break Grand Ballroom Foyer - 2nd Floor

4:30 PM - 6:35 PM Concurrent SessionsANALCO: Session 3 St. Charles A - 1st Floor




6:35 PM - 8:00 PM Dinner Break Attendees on their own

8:00 PM - 9:00 PM ALENEX & ANALCO Business Meeting St. Charles A - 1st Floor

Monday, January 7



9:00 AM - 11:05 AM Concurrent SessionsALENEX: Session 1 St. Charles A - 1st Floor





11:30 AM - 12:30 PM IP2 Submodular Functions and Their ApplicationsJan Vondrak, IBM Almaden Research Center, USA Grand Ballroom A/B/C - 2nd Floor

12:30 PM - 2:00 PM Lunch Break Attendees on their own

2:00 PM - 4:05 PM Concurrent SessionsALENEX: Session 2 St. Charles A - 1st Floor





4:30 PM - 6:35 PM Concurrent SessionsALENEX: Session 3 St. Charles A - 1st Floor




6:35 PM - 8:00 PM Dinner Break Attendees on their own

8:00 PM - 9:00 PM SODA Business Meeting and Awards Presentation Grand Ballroom C - 2nd Floor

Complimentary wine and beer will be served.

Tuesday, January 8



9:00 AM - 11:05 AM Concurrent SessionsCP19 SODA Session 7A Grand Ballroom A/B - 2nd Floor




11:30 AM - 12:30 PM IP3 On Graph Property Testing, Arithmetic Progressions and CommunicationNoga Alon, Tel Aviv University, Israel Grand Ballroom A/B/C - 2nd Floor

12:30 PM - 2:00 PM Lunch Break Attendees on their own

2:00 PM - 4:05 PM Concurrent SessionsCP22 SODA Session 8A Grand Ballroom A/B - 2nd Floor




4:30 PM - 6:35 PM Concurrent SessionsCP25 SODA Session 9A Grand Ballroom A/B - 2nd Floor



= Award Presentation

= Business Meeting

= Coffee Break

= Continental Breakfast and Luncheon

= Refreshments

Key to abbreviations and symbols


sODA, ALeNeX and ANALcO Program


Sunday, January 6

Registration8:00 AM-5:00 PMRoom: Grand Ballroom Foyer - 2nd Floor

continental breakfast8:30 AMRoom: Grand Ballroom Foyer - 2nd Floor

Saturday, January 5

Registration5:00 PM-8:00 PMRoom: Grand Ballroom Foyer - 2nd Floor

Welcome Reception6:00 PM-8:00 PMRoom: Grand Ballroom A/B/C - 2nd Floor

Final Program

Sunday, January 6

ANALcO: session 19:00 AM-10:40 AMRoom: St. Charles A - 1st FloorChair: Mordecai Golin, Hong Kong University of Science and Technology, Hong Kong9:00-9:20 extremal Parameters in sub-critical Graph classesMichael Drmota, Technische Universitat Vienna, Austria; Marc Noy, Universidad Politecnica de Catalunya, Spain9:25-9:45 the Number of Ways to Assemble a GraphMiklos Bona and Andrew Vince, University of

Florida, USA

9:50-10:10 bootstrap Percolation on Random Geometric GraphsMilan Bradonjic and Iraj Saniee, Bell Labs,

Alcatel-Lucent, USA

10:15-10:35 Approximating Fault-tolerant Domination in General GraphsKlaus-Tycho Förster, ETH Zürich, Switzerland


Sunday, January 6

cP1sODA: session 1A9:00 AM-11:05 AMRoom: Grand Ballroom A/B - 2nd Floor

Chair: John Iacono, Polytechnic Institute of New York University, USA

9:00-9:20 Mixing times of Markov chains for self-Organizing Lists and biased Permutations

Prateek Bhakta, Sarah Miracle, and Dana Randall, Georgia Institute of Technology, USA; Amanda Streib, National Institute of Standards and Technology, USA

9:25-9:45 balls into bins Via Local searchPaul Bogdan, Carnegie Mellon University, USA,

Thomas Sauerwald, Max Planck Institute for Computer Science, Germany; Alexandre Stauffer, Microsoft Research, USA; He Sun, Max Planck Institute for Informatics, Germany

9:50-10:10 convergence of Multivariate belief Propagation, with Applications to cuckoo Hashing and Load balancingMathieu Leconte, Technicolor - INRIA, France;

Marc Lelarge, INRIA, France; Laurent Massoulié, Microsoft Research-INRIA Joint Centre, France

10:15-10:35 Approximate counting Via correlation Decay on Planar GraphsYitong Yin, Nanjing University, China; Chihao

Zhang, Shanghai Jiaotong University, China

10:40-11:00 correlation Decay Up to Uniqueness in spin systemsLiang Li, Peking University, China; Pinyan Lu,

Microsoft Research Asia; Yitong Yin, Nanjing University, China

Sunday, January 6

cP2sODA: session 1b9:00 AM-11:05 AMRoom: Grand Ballroom C - 2nd Floor

Chair: Aleksander Madry, Microsoft Research New England, USA

9:00-9:20 Online Mixed Packing and covering

Yossi Azar, Tel Aviv University, Israel; Umang Bhaskar, California Institute of Technology, USA; Lisa Fleischer, Dartmouth College, USA; Debmalya Panigrahi, Microsoft Research, USA

9:25-9:45 Randomized Primal-Dual Analysis of Ranking for Online bipartite MatchingNikhil R. Devanur, Microsoft Research, USA;

Kamal Jain, eBay Research Labs, USA; Robert Kleinberg, Cornell University, USA

9:50-10:10 Matroid secretary for Regular and Decomposable MatroidsMichael Dinitz, Weizmann Institute of Science,

Israel; Guy Kortsarz, Rutgers University, USA

10:15-10:35 A New Approach to Online scheduling: Approximating the Optimal competitive RatioAndreas Wiese, MPII Saarbrücken, Germany;

Elisabeth Günther, Olaf Maurer, and Nicole Megow, Technische Universität Berlin, Germany

10:40-11:00 Weighted Flowtime on capacitated MachinesKyle Fox, University of Illinois at Urbana-

Champaign, USA; Madhukar Korupolu, Google Research, USA

Sunday, January 6

cP3sODA: session 1c9:00 AM-11:05 AMRoom:Ballroom D - 2nd Floor

Chair: Suresh Venkatasubramanian, University of Utah, USA

9:00-9:20 Approximate shortest Descending PathsSiu-Wing Cheng and Jiongxin Jin, Hong Kong

University of Science and Technology, Hong Kong

9:25-9:45 computing the Discrete Fréchet Distance in subquadratic time

Pankaj Agarwal, Duke University, USA; Rinat Ben Avraham and Haim Kaplan, Tel Aviv University, Israel; Micha Sharir, Tel Aviv University, Israel, and New York University, USA

9:50-10:10 the complexity of Detecting taut Angle structures on triangulationsBenjamin A. Burton and Jonathan Spreer,

University of Queensland, Australia

10:15-10:35 An Infinite class of sparse-Yao spannersMirela Damian and Matthew Bauer, Villanova

University, USA

10:40-11:00 Weighted Graph Laplace Operator under topological NoiseTamal Dey, Pawas Ranjan, and Yusu Wang,

Ohio State University, USA

coffee break11:05 AM-11:30 AMRoom: Grand Ballroom Foyer - 2nd Floor


ANALcO: session 22:00 PM-3:40 PMRoom: St. Charles A - 1st Floor

Chair: Michael Drmota, Technische Universitat Vienna, Austria

2:00-2:20 Analysis of Parameters of trees corresponding to Huffman codes and sums of Unit FractionsClemens Heuberger, Alpen-Adria-Universität

Klagenfurt, Austria; Daniel Krenn, Technische Universität, Graz, Austria; Stephan Wagner, Stellenbosch University, South Africa

2:25-2:45 the Variance of the Number of 2-Protected Nodes in a trieJeffrey B. Gaither and Mark Daniel Ward,

Purdue University, USA

2:50-3:10 exact-size sampling for Motzkin trees in Linear time Via boltzmann samplers and Holonomic specification

Axel Bacher, Olivier Bodini, and Alice Jacquot, Université Paris 13, France

3:15-3:35 Perpetuities in Fair Leader election AlgorithmsHosam M. Mahmoud and Ravi Kalpathy,

George Washington University, USA

Sunday, January 6Luncheon **ticketed event**12:30 PM-2:00 PMRoom: Astor Ballroom - 2nd Floor

Please visit the SIAM Registration Desk if you require a ticket.

Sunday, January 6Sunday, January 6

IP1“If You can specify It, You can Analyze It” ---the Lasting Legacy of Philippe Flajolet11:30 AM-12:30 PMRoom: Grand Ballroom A/B/C - 2nd Floor

Chair: Peter Sanders, Karlsruhe Institute of Technology, Germany

The “Flajolet School” of the analysis of algorithms and combinatorial structures is centered on an effective calculus, known as analytic combinatorics, for the development of mathematical models that are sufficiently accurate and precise that they can be validated through scientific experimentation. It is based on the generating function as the central object of study, first as a formal object that can translate a specification into mathematical equations, then as an analytic object whose properties as a function in the complex plane yield the desired quantitative results. Universal laws of sweeping generality can be proven within the framework, and easily applied. Standing on the shoulders of Cauchy, Polya, de Bruijn, Knuth, and many others, Philippe Flajolet and scores of collaborators developed this theory and demonstrated its effectiveness in a broad range of scientific applications. Flajolet’s legacy is a vibrant field of research that holds the key not just to understanding the properties of algorithms and data structures, but also to understanding the properties of discrete structures that arise as models in all fields of science. This talk will survey Flajolet’s story and its implications for future research.

Robert SedgewickPrinceton University, USA


cP5sODA: session 2b2:00 PM-4:05 PMRoom: Grand Ballroom C - 2nd Floor

Chair: Konstantin Makarychev, IBM T.J. Watson Research Center, USA

2:00-2:20 better balance by being biased: A 0.8776-Approximation for Max bisectionKonstantinos Georgiou, University of

Waterloo, Canada; Per Austrin, KTH Royal Institute of Technology, Sweden; Siavosh Benabbas, University of Toronto, Canada

2:25-2:45 Approximating Non-Uniform sparsest cut Via Generalized spectraVenkatesan Guruswami, Carnegie Mellon

University, USA; Ali K. Sinop, Princeton University, USA

2:50-3:10 Local Distribution and the symmetry Gap: Approximability of Multiway Partitioning ProblemsAlina Ene, University of Illinois at Urbana-

Champaign, USA; Jan Vondrak, IBM Almaden Research Center, USA; Yi Wu, Purdue University, USA

3:15-3:35 Poly-Logarithmic Approximation for Maximum Node Disjoint Paths with constant congestionAlina Ene and Chandra Chekuri, University of

Illinois at Urbana-Champaign, USA

3:40-4:00 How to sell Hyperedges: the Hypermatching Assignment ProblemFabrizio Grandoni, Marek Cygan, and Monaldo Mastrolilli, University of Lugano, Switzerland

Sunday, January 6 Sunday, January 6 Sunday, January 6

cP4sODA: session 2A2:00 PM-4:05 PMRoom: Grand Ballroom A/B - 2nd Floor


2:00-2:20 twisted tabulation Hashing

Mihai Patrascu, AT&T Labs - Research, USA; Mikkel Thorup, University of Copenhagen, Denmark and AT&T Labs – Research, USA

2:25-2:45 compressed static Functions with ApplicationsDjamal Belazzougui, Université Paris-Diderot,

France; Rossano Venturini, University of Pisa, Italy

2:50-3:10 Adaptive and Approximate Orthogonal Range countingBryan T. Wilkinson, Aarhus University,

Denmark; Timothy Chan, University of Waterloo, Canada

3:15-3:35 the space complexity of 2-Dimensional Approximate Range countingZhewei Wei, Aarhus University, Denmark; Ke

Yi, Hong Kong University of Science and Technology, Hong Kong

3:40-4:00 Near-Optimal Range Reporting structures for categorical DataKasper G. Larsen and Freek Van Walderveen,

Aarhus University, Denmark

cP6sODA: session 2c2:00 PM-4:05 PMRoom:Ballroom D - 2nd Floor

Chair: Kavitha Telikepalli, Tata Institute of Fundamental Research, India

2:00-2:20 shortest Non-trivial cycles in Directed and Undirected surface GraphsKyle Fox, University of Illinois at Urbana-

Champaign, USA

2:25-2:45 Packing Directed cycles through a specified Vertex setKen-ichi Kawarabayashi, National Institute of

Informatics, Japan; Daniel Kral, University of Warwick, United Kingdom; Marek Krcal, Charles University, Czech Republic; Stephan Kreutzer, Technical University Berlin, Germany

2:50-3:10 4-connected Projective-Planar Graphs Are Hamiltonian-connectedKenta Ozeki and Ken-ichi Kawarabayashi,

National Institute of Informatics, Japan

3:15-3:35 Jungles, bundles, and Fixed-Parameter tractabilityFedor Fomin and Michal Pilipczuk, University

of Bergen, Norway

3:40-4:00 A simple Algorithm for the Graph Minor Decomposition - Logic Meets structural Graph theory -Martin Grohe, RWTH Aachen University,

Germany; Ken-ichi Kawarabayashi, National Institute of Informatics, Japan; Bruce Reed, McGill University, Canada

coffee break4:05 PM-4:30 PMRoom: Grand Ballroom Foyer - 2nd Floor


Sunday, January 6 Sunday, January 6 Sunday, January 6


Chair: Krzysztof Onak, Carnegie Mellon University, USA

4:30-4:50 Restricted Isometry of Fourier Matrices and List Decodability of Random Linear codesMahdi Cheraghchi, Venkatesan Guruswami,

and Ameya Velingker, Carnegie Mellon University, USA

4:55-5:15 Fast Algorithms for Interactive codingZvika Brakerski, Stanford University, USA;

Moni Naor, Weizmann Institute of Science, Israel

5:20-5:40 shift Finding in sub-Linear timeHaitham Hassanieh, Massachusetts Institute

of Technology, USA; Alexandr Andoni, Microsoft Research, USA; Piotr Indyk and Dina Katabi, Massachusetts Institute of Technology, USA

5:45-6:05 the Fast cauchy transform and Faster Robust Linear RegressionMichael Mahoney, Stanford University, USA;

Kenneth Clarkson, IBM Corporation, USA; Petros Drineas and Malik Magdon-Ismail, Rensselaer Polytechnic Institute, USA; Xiangrui Meng, Stanford University, USA; David Woodruff, IBM Almaden Research Center, USA

6:10-6:30 Generalized Perron-Frobenius theorem for Multiple choice Matrices, and ApplicationsChen Avin, Ben-Gurion University, Israel;

Michael Borokhovich, Ben Gurion University, Israel; Yoram Haddad, Jerusalem College of Technology, Israel; Erez Kantor, Technion Israel Institute of Technology, Israel; Zvi Lotker, Ben-Gurion University, Israel; Merav Parter and David Peleg, Weizmann Institute of Science, Israel



4:30-4:50 New Additive spannersShiri Chechik, Microsoft Research, USA

4:55-5:15 Fast constructions of Light-Weight spanners for General GraphsMichael Elkin, Ben-Gurion University, Israel;

Shay Solomon, Weizmann Institute of Science, Israel

5:20-5:40 Distance Oracles for stretch Less than 2Rachit Agarwal and Philip Godfrey, University of

Illinois, Urbana-Champaign, USA

5:45-6:05 Approximate Distance Oracles with Improved Query timeChristian Wulff-Nilsen, University of

Copenhagen, Denmark

6:10-6:30 More compact Oracles for Approximate Distances in Undirected Planar GraphsChristian Sommer, Massachusetts Institute of

Technology, USA; Ken-ichi Kawarabayashi, National Institute of Informatics, Japan; Mikkel Thorup, University of Copenhagen, Denmark and AT&T Labs – Research, USA

ANALcO: session 34:30 PM-6:10 PMRoom: St. Charles A - 1st Floor

Chair: Hosam M. Mahmoud, George Washington University, USA

4:30-4:50 Approximate counting of Matchings in sparse Uniform HypergraphsEdyta Szymanska, Adam Mickiewicz University,

Poland; Andrzej Rucinski, Adam Mickiewicz University, Poland, and Emory University, USA; Marek Karpinski, University of Bonn, Germany

4:55-5:15 When Is It Worthwhile to Propagate a constraint? A Probabilistic Analysis of AlldifferentDaniele Gardy and Jeremie Du Boisberranger,

University of Versailles Saint-Quentin, France; Xavier Lorca, Ecole des Mines de Nancy, France; Charlotte Truchet, University of Nantes, France

5:20-5:40 On Delta-Method of Moments and Probabilistic sumsJacek Cichon, Zbigniew Golebiewski, Marcin

Kardas, and Marek Klonowski, Wroclaw University of Technology, Poland

5:45-6:05 Unlabeled equivalence for Matroids Representable over Finite FieldsSandra R. Kingan, City University of New York,

Brooklyn, USA


Sunday, January 6


Chair: Aaron Roth, University of Pennsylvania, USA

4:30-4:50 simple and Nearly Optimal Multi-Item AuctionsYang Cai, Massachusetts Institute of

Technology, USA; Zhiyi Huang, University of Pennsylvania, USA

4:55-5:15 Reducing Revenue to Welfare Maximization: Approximation Algorithms and Other Generalizations

Yang Cai, Constantinos Daskalakis, and Matt Weinberg, Massachusetts Institute of Technology, USA

5:20-5:40 Optimal and efficient Parametric AuctionsPablo Azar, Constantinos Daskalakis, Silvio

Micali, and Matt Weinberg, Massachusetts Institute of Technology, USA

5:45-6:05 clinching Auctions with Online supplyRenato Paes Leme, Cornell University, USA;

Gagan Goel, Google Research, USA; Vahab Mirrokni, Google, Inc., USA

6:10-6:30 Ironing in Dynamic Revenue Management: Posted Prices & biased AuctionsMallesh M. Pai, University of Pennsylvania,

USA; Rahul Deb, University of Toronto, Canada

Dinner break6:35 PM-8:00 PMAttendees on their own

ALeNeX & ANALcO business Meeting8:00 PM-9:00 PMRoom: St. Charles A - 1st Floor

Monday, January 7



Monday, January 7

ALeNeX: session 19:00 AM-11:05 AMRoom: St. Charles A - 1st Floor


9:00-9:20 efficient Algorithms for Dualizing Large-scale HypergraphsTakeaki Uno, National Institute of

Informatics, Japan; Keisuke Murakami, Aoyama Gakuin University, Japan

9:25-9:45 A Min-edge cost Flow Framework for capacitated covering ProblemsJessica Chang, University of Washington, USA;

Samir Khuller, University of Maryland, College Park, USA

9:50-10:10 short and simple cycle separators in Planar GraphsEli Fox-Epstein, Tufts University, USA; Shay

Mozes and Phitchaya Phothilimthana, Massachusetts Institute of Technology, USA; Christian Sommer, Unaffiliated

10:15-10:35 Polynomial-time construction of contraction Hierarchies for Multi-criteria ObjectivesStefan Funke, University of Stuttgart,

Germany; Sabine Storandt, University of Freiburg, Germany

10:40-11:00 engineering Java 7’s Dual Pivot Quicksort Using MalijanSebastian Wild, Markus Nebel, Raphael

Reitzig, and Ulrich Laube, University of Kaiserslautern, Germany


Monday, January 7 Monday, January 7 Monday, January 7



9:00-9:20 simple, Fast and Deterministic Gossip and Rumor spreadingBernhard Haeupler, Massachusetts Institute of

Technology, USA

9:25-9:45 On the complexity of Information spreading in Dynamic NetworksChinmoy Dutta, Northeastern University, USA;

Gopal Pandurangan, Nanyang Technological University, Singapore; Rajmohan Rajaraman, Zhifeng Sun, and Emanuele Viola, Northeastern University, USA

9:50-10:10 Anonymous Meeting in NetworksYoann Dieudonne and Andrzej Pelc, Université

du Québec en Outaouais, Canada

10:15-10:35 Near Optimal Leader election in Multi-Hop Radio NetworksMohsen Ghaffari and Bernhard Haeupler,

Massachusetts Institute of Technology, USA

10:40-11:00 Finding endogenously Formed communitiesMaria-Florina Balcan, Georgia Institute

of Technology, USA; Christian Borgs, Microsoft Research, USA; Mark Braverman, Princeton University, USA; Jennifer Chayes, Microsoft Research, USA; Shang-Hua Teng, University of Southern California, USA



9:00-9:20 Reporting Neighbors in High-Dimensional euclidean spaceDror Aiger, Google, Inc., USA; Haim Kaplan

and Micha Sharir, Tel Aviv University, Israel

9:25-9:45 euclidean spanners in High DimensionsSariel Har-Peled, University of Illinois, Urbana-

Champaign, USA; Piotr Indyk, Massachusetts Institute of Technology, USA; Anastasios Sidiropoulos, University of Illinois, Urbana-Champaign, USA

9:50-10:10 clustering Affine subspaces: Hardness and AlgorithmsEuiwoong Lee, Carnegie Mellon University, USA;

Leonard Schulman, California Institute of Technology, USA

10:15-10:35 the traveling salesman Problem for Lines, balls and PlanesAdrian Dumitrescu, University of Wisconsin,

Milwaukee, USA; Csaba D. Toth, University of Calgary, Canada

10:40-11:00 Approximating Watchman RoutesJoseph Mitchell, Stony Brook University, USA




9:00-9:20 the communication complexity of AdditionEmanuele Viola, Northeastern University, USA

9:25-9:45 Lower bounds for Adaptive sparse RecoveryEric Price, Massachusetts Institute of

Technology, USA; David Woodruff, IBM Almaden Research Center, USA

9:50-10:10 tight cell-Probe bounds for Online Hamming Distance computationMarkus Jalsenius and Raphael Clifford,

University of Bristol, United Kingdom; Benjamin Sach, University of Warwick, United Kingdom

10:15-10:35 On the Number of MatroidsNikhil Bansal, Eindhoven University of

Technology, Netherlands; Rudi Pendavingh and Jorn van Der Pol, Technische Universiteit Eindhoven, The Netherlands

10:40-11:00 Playing Mastermind with Many colorsBenjamin Doerr and Reto Spöhel, Max

Planck Institute for Informatics, Germany; Henning Thomas, ETH Zürich, Switzerland; Carola Winzen, Max Planck Institute for Informatics, Germany


Monday, January 7

IP2submodular Functions and their Applications11:30 AM-12:30 PMRoom: Grand Ballroom A/B/C - 2nd Floor


Submodular functions, a discrete analogue of convex functions, have played a fundamental role in combinatorial optimization since the 1970s. In the last decade, there has been renewed interest in submodular functions due to their interpretation as valuation functions of self-interested agents in algorithmic game theory. These developments have led to new questions as well as new algorithmic techniques. In this talk, we will discuss the concept of submodularity, its motivation and its unifying role in combinatorial optimization, as well as the evolution of the relevant algorithmic techniques. We will survey the state of the ar t in optimization of submodular functions, as well as selected applications in algorithmic game theory, social networks and machine learning, and some future challenges.

Jan VondrakIBM Almaden Research Center, USA

Lunch break12:30 PM-2:00 PMAttendees on their own

Monday, January 7

ALeNeX: session 22:00 PM-4:05 PMRoom: St. Charles A - 1st Floor

Chair: Gonzalo Navarro, University of Chile, Chile

2:00-2:20 3D Kinetic Alpha complexes and their ImplementationMichael Kerber, Stanford University, USA;

Herbert Edelsbrunner, Institute of Science and Technology, Austria

2:25-2:45 computational topology and Normal surfaces: theoretical and experimental complexity boundsBenjamin A. Burton and Jonathan Spreer,

University of Queensland, Australia; João Paixão, Pontíficia Universidade Católica Do Rio de Janeiro, Brazil

2:50-3:10 Inducing suffix and Lcp Arrays in external MemoryJohannes Fischer, Timo Bingmann, and Vitaly

Osipov, Karlsruhe Institute of Technology, Germany

3:15-3:35 Lempel-Ziv Factorization: simple, Fast, PracticalSimon J. Puglisi and Dominik Kempa,

University of Helsinki, Finland

3:40-4:00 Fast Packed string Matching for short PatternsSimone Faro, Universita’ di Catania, Italy;

Oguzhan Kulekci, National Research Institute of Electronics and Cryptology, Turkey

Monday, January 7



2:00-2:20 Windows into Relational events: Data structures for contiguous subsequences of edgesMichael J. Bannister, Christopher DuBois, David

Eppstein, and Padhraic Smyth, University of California, Irvine, USA

2:25-2:45 Optimal Dynamic sequence RepresentationsGonzalo Navarro and Yakov Nekrich, University

of Chile, Chile

2:50-3:10 towards More Realistic Probabilistic Models for Data structures: the external Path Length in tries under the Markov ModelRalph Neininger and Kevin Leckey, J.W. Goethe-

Universität, Germany; Wojciech Szpankowski, Purdue University, USA

3:15-3:35 Output-sensitive skyline Algorithms in external MemoryYufei Tao, Xiaocheng Hu, and Cheng Sheng,

Chinese University of Hong Kong, Hong Kong; Yi Yang and Shuigeng Zhou, Fudan University, China

3:40-4:00 Multiway simple cycle separators and I/O-efficient Algorithms for Planar GraphsLars Arge and Freek Van Walderveen, Aarhus

University, Denmark; Norbert Zeh, Dalhousie University, Canada


Monday, January 7



2:00-2:20 New Approximability Results for two-Dimensional bin PackingLars Praedel and Klaus Jansen, University of

Kiel, Germany

2:25-2:45 Minimum Makespan scheduling with Low Rank Processing timesAditya Bhaskara, EPFL, Switzerland;

Ravishankar Krishnaswami, Princeton University, USA; Kunal Talwar, Microsoft

Research Silicon Valley, USA; Udi Wieder, Microsoft Research, USA

2:50-3:10 energy efficient scheduling of Parallelizable JobsKyle Fox, University of Illinois at Urbana-

Champaign, USA; Sungjin Im, Duke University, USA; Benjamin Moseley, Toyota Technological Institute at Chicago, USA

3:15-3:35 Lyndon Words and short superstringsMarcin Mucha, University of Warsaw, Poland

3:40-4:00 A constant Factor Approximation Algorithm for Reordering buffer ManagementNoa Avigdor-Elgrabli, Technion Israel Institute

of Technology, Israel; Yuval Rabani, Hebrew University of Jerusalem, Israel

Monday, January 7

ALeNeX: session 34:30 PM-6:35 PMRoom: St. Charles A - 1st Floor

Chair: Norbert Zeh, Dalhousie University, Canada

4:30-4:50 On Parallelizing Matrix Multiplication by the column-Row MethodKonstantin Kutzkov, Andrea Campagna,

Rasmus Pagh, and Morten Stöckel, University of Copenhagen, Denmark

4:55-5:15 Robust Gossip-based Aggregation: A Practical Point of ViewGerhard Niederbrucker and Wilfried N.

Gansterer, University of Vienna, Austria

5:20-5:40 the cost of Address translationTomasz Jurkiewicz and Kurt Mehlhorn, Max

Planck Institute for Informatics, Germany

5:45-6:05 On Unifying the space of - sampling AlgorithmsGraham Cormode, AT&T Labs - Research,

USA; Donatella Firmani, University of Rome La Sapienza, Italy

6:10-6:30 Practical batch-Updatable external Hashing with sortingHyeontaek Lim and David Andersen,

Carnegie Mellon University, USA; Michael Kaminsky, Intel Labs, USA

Monday, January 7



2:00-2:20 5-coloring K3,k-Minor-Free GraphsKen-ichi Kawarabayashi, National Institute of

Informatics, Japan

2:25-2:45 List-coloring embedded GraphsZdenek Dvorak, Charles University, Czech

Republic; Ken-ichi Kawarabayashi, National Institute of Informatics, Japan

2:50-3:10 totally Odd subdivisions and Parity subdivisions: structures and coloringKen-ichi Kawarabayashi, National Institute of

Informatics, Japan

3:15-3:35 simultaneous PQ-Ordering with Applications to constrained embedding ProblemsIgnaz Rutter and Thomas Bläsius, Karlsruhe

Institute of Technology, Germany

3:40-4:00 Known Algorithms for edge clique cover are Probably OptimalMarek Cygan, University of Lugano,

Switzerland; Marcin Pilipczuk, University of Warsaw, Poland; Michal Pilipczuk, University of Bergen, Norway



Monday, January 7



4:30-4:50 breaking the nlog n barrier for solvable-Group IsomorphismDavid J. Rosenbaum, University of Washington,

USA

4:55-5:15 Fast Matrix Multiplication Using coherent configurationsHenry Cohn, Microsoft Research, USA;

Christopher Umans, California Institute of Technology, USA

5:20-5:40 Lattice sparsification and the Approximate closest Vector ProblemDaniel Dadush, New York University,

USA; Gabor Kun, Courant Institute of Mathematical Sciences, New York University, USA

5:45-6:05 Algorithms for the Densest sub-Lattice ProblemDaniel Dadush, New York University, USA;

Daniele Micciancio, University of California, San Diego, USA

6:10-6:30 Minimizing the Number of Lattice Points in a translated PolygonNicolai Hähnle, Technische Universität

Berlin, Germany; Friedrich Eisenbrand, École Polytechnique Fédérale de Lausanne, Switzerland

Monday, January 7



4:30-4:50 Dynamic Graph connectivity in Polylogarithmic Worst case timeBruce M. Kapron, Valerie King, and Ben

Mountjoy, University of Victoria, Canada

4:55-5:15 Decremental Maintenance of strongly connected componentsLiam Roditty, Bar-Ilan University, Israel

5:20-5:40 Approximate Maximum Flow on separable Undirected GraphsGary Miller and Richard Peng, Carnegie

Mellon University, USA

5:45-6:05 breaking the O(n2.5) Deterministic time barrier for Undirected Unit-capacity Maximum FlowRan Duan, Max Planck Institute for

Informatics, Germany

6:10-6:30 smoothed Analysis of the successive shortest Path AlgorithmTobias Brunsch, University of Bonn, Germany;

Kamiel Cornelissen and Bodo Manthey, University of Twente, Netherlands; Heiko Röglin, University of Bonn, Germany

Monday, January 7



4:30-4:50 Regret Minimization for Reserve Prices in second-Price AuctionsYishay Mansour, Tel Aviv University, Israel;

Nicolo’ Cesa-Bianchi, Universita degli Studi di Milano, Italy; Claudio Gentile, University of Insubria, Italy

4:55-5:15 communication complexity of combinatorial Auctions with submodular ValuationsShahar Dobzinski, Weizmann Institute of

Science, Israel; Jan Vondrak, IBM Almaden Research Center, USA

5:20-5:40 Online submodular Welfare Maximization: Greedy Is OptimalIan Post and Michael Kapralov, Stanford

University, USA; Jan Vondrak, IBM Almaden Research Center, USA

5:45-6:05 towards Polynomial simplex-Like Algorithms for Market equilibriaJugal Garg and Ruta Mehta, Georgia Institute

of Technology, USA; Milind Sohoni, Indian Institute of Technology-Bombay, India; Nisheeth K. Vishnoi, Microsoft Research, India

6:10-6:30 A Unified Approach to truthful scheduling on Related MachinesLeah Epstein, University of Haifa, Israel;

Asaf Levin, Technion Israel Institute of Technology, Israel; Rob van Stee, Max Planck Institute for Informatics, Germany

Dinner break6:35 PM-8:00 PMAttendees on their own


Tuesday, January 8



Tuesday, January 8



9:00-9:20 exponential Lower bounds for the PPsZ k-sat AlgorithmDominik A. Scheder, Aarhus University,

Denmark; Shiteng Chen and Bangsheng Tang, Tsinghua University, P. R. China; Navid Talebanfard, Aarhus University, Denmark

9:25-9:45 complexity of sat Problems, clone theory and the exponential time HypothesisVictor Lagerkvist, Peter Jonsson, and Gustav

Nordh, Linköping University, Sweden; Bruno Zanuttini, Université de Caen Basse Normandie, France

9:50-10:10 Dichotomy for Holant * Problems with a Function on Domain size 3Jin-Yi Cai, University of Wisconsin, Madison,

USA; Pinyan Lu, Microsoft Research Asia; Mingji Xia, Max Planck Institute for Informatics, Germany

10:15-10:35 skew bisubmodularity and Valued csPsAndrei Krokhin, Anna Huber, and Robert

Powell, Durham University, United Kingdom

10:40-11:00 Frozen Variables in Random boolean constraint satisfaction ProblemsMichael Molloy, University of Toronto,

Canada; Ricardo Restrepo, Universidad di Antioquia, Colombia

Monday, January 7sODA business Meeting and Awards Presentation8:00 PM-9:00 PMRoom: Grand Ballroom C - 2nd Floor

Complimentary wine and beer will be served.

congratulations to the following sODA award

winners!Best Papers

(awarded jointly to two papers)A Simple Algorithm for the Graph

Minor Decomposition - Logic meets Structural Graph Theory

Martin Grohe, Ken-Ichi Kawarabayashi and Bruce Reed

Dynamic Graph Connectivity in Polylogarithmic Worst Case Time

Bruce Kapron, Valerie King and Ben Mountjoy

Best Student Papers (awarded jointly to two papers)

New Additive Spanners

Shiri Chechik

Simple, Fast and Deterministic Gossip and Rumor Spreading

Bernhard Haeupler


Tuesday, January 8



9:00-9:20 exponentially Improved Algorithms and Lower bounds for testing signed MajoritiesRocco A. Servedio, Columbia University, USA;

Dana Ron, Tel Aviv University, Israel

9:25-9:45 testing Low-complexity Affine-Invariant PropertiesArnab Bhattacharyya, Rutgers University, USA;

Eldar Fischer, Technion Israel Institute of Technology, Israel; Shachar Lovett, Massachusetts Institute of Technology, USA

9:50-10:10 Learning and testing Integer submodular FunctionsGrigory Yaroslavtsev and Sofya Raskdhonikova,

Pennsylvania State University, USA

10:15-10:35 Learning Disjunctions: Near-Optimal trade-off between Mistakes and "I Don’t Know’s"Morteza Zadimoghaddam and Erik Demaine,

Massachusetts Institute of Technology, USA

10:40-11:00 Learning Mixtures of structured Distributions over Discrete DomainsIlias Diakonikolas, University of Edinburgh,

United Kingdom; Siu-On Chan, University of California, Berkeley, USA; Rocco A. Servedio and Xiaorui Sun, Columbia University, USA

Tuesday, January 8



9:00-9:20 On Differentially Private Low Rank ApproximationMichael Kapralov, Stanford University, USA;

Kunal Talwar, Microsoft Research Silicon Valley, USA

9:25-9:45 the Power of Linear Reconstruction AttacksShiva Prasad Kasiviswanathan, GE Global

Research, USA; Mark Rudelson, University of Michigan, USA; Adam Smith, Pennsylvania State University, USA

9:50-10:10 turning big Data into tiny Data: constant-size coresets for K-Means, PcA and Projective clusteringMelanie Schmidt, TU Dortmund, Germany;

Dan Feldman, Massachusetts Institute of Technology, USA; Christian Sohler, Technische Universität Dortmund, Germany

10:15-10:35 An Almost Optimal Algorithm for computing Nonnegative RankAnkur Moitra, Massachusetts Institute of

Technology, USA

10:40-11:00 the simplex Method Is strongly Polynomial for Deterministic Markov Decision ProcessesIan Post and Yinyu Ye, Stanford University,

USA


Tuesday, January 8

IP3On Graph Property testing, Arithmetic Progressions and communication11:30 AM-12:30 PMRoom: Grand Ballroom A/B/C - 2nd Floor

Chair: Sanjeev Khanna, University of Pennsylvania, USA

Tools from Extremal Graph Theory are helpful in the study of problems in Additive Number Theory, Theoretical Computer Science, and Information Theory. I will illustrate this fact by several closely related examples focusing on a recent one in a joint work with Moitra and Sudakov. The main combinatorial question addressed in this work is that of estimating the maximum possible density of a graph consisting of a union of pairwise edge disjoint large induced matchings. It turns out that variants of this problem appear in the investigation of several seemingly unrelated topics. The lecture will include a brief general introduction, a description of the methods used and results obtained, as well as a discussion of the r emaining challenges.

Noga AlonTel Aviv University, Israel

Lunch break12:30 PM-2:00 PMAttendees on their own


Tuesday, January 8



2:00-2:20 Nested Quantum Walks with Quantum Data structuresStacey Jeffery and Robin Kothari, University

of Waterloo, Canada; Frederic Magniez, Université Paris-Diderot, France

2:25-2:45 Improved Quantum Query Algorithms for triangle Finding and Associativity testingFrederic Magniez, Université Paris-Diderot,

France; Troy Lee, National University of Singapore, Singapore; Miklos Santha, Université Paris-Diderot, France

2:50-3:10 efficient Protocols for Generating bipartite classical Distributions and Quantum statesShengyu Zhang, Chinese University of Hong

Kong, Hong Kong; Rahul Jain, National University of Singapore, Singapore; Yaoyun Shi, University of Michigan, USA; Zhaohui Wei, National University of Singapore, Singapore

3:15-3:35 Fuel efficient computation in Passive self-AssemblyRobert T. Schweller and Michael Sherman,

University of Texas - Pan American, USA

3:40-4:00 Active self-Assembly of simple Units Using An Insertion PrimitiveNadine Dabby, California Institute of

Technology, USA; Ho-Lin Chen, National Taiwan University, Taiwan

Tuesday, January 8



2:00-2:20 Approximability and Proof complexityYuan Zhou and Ryan O’Donnell, Carnegie

Mellon University, USA

2:25-2:45 Graph Products Revisited: tight Approximation Hardness of Induced Matching, Poset Dimension and MoreParinya Chalermsook, IDSIA, Switzerland;

Danupon Nanongkai, University of Vienna, Austria; Bundit Laekhanukit, McGill University, Canada

2:50-3:10 the Diffusion of Networking technologiesZhenming Liu, Princeton University, USA;

Sharon Goldberg, Boston University, USA

3:15-3:35 the Power of Non-Uniform Wireless PowerMagnus M. Halldorsson, Reykjavik University,

Iceland; Stephan Holzer, ETH Zürich, Switzerland; Pradipta Mitra, Reykjavik University, Iceland; Roger Wattenhofer, ETH Zürich, Switzerland

3:40-4:00 Local-search based Approximation Algorithms for Mobile Facility Location ProblemsChaitanya Swamy, Sara Ahmadian, and Zachary

Friggstad, University of Waterloo, Canada

Tuesday, January 8



2:00-2:20 eps-samples for KernelsJeff Phillips, University of Utah, USA

2:25-2:45 Higher-Order Geodesic Voronoi Diagrams in a Polygonal Domain with HolesChih-Hung Liu, Academia Sinica, Taiwan; D.

T. Lee, National Chung-Hsing University, Taiwan

2:50-3:10 transforming curves on surfaces ReduxJeff Erickson, University of Illinois, USA; Kim

Whittlesey, University of Illinois at Urbana-Champaign, USA

3:15-3:35 Morphing Planar Graph Drawings with a Polynomial Number of stepsAnna Lubiw and Soroush Alamdari, University

of Waterloo, Canada; Patrizio Angelini, Universita di Roma Tre, Italy; Timothy M. Chan, University of Waterloo, Canada; Giuseppe Di Battista, Universita di Roma Tre, Italy; Fabrizio Frati, University of Sydney, Australia; Maurizio Patrignani and Vincenzo Roselli, Universita di Roma Tre, Italy; Sahil Singla, University of Waterloo, Canada; Bryan T. Wilkinson, Aarhus University, Denmark

3:40-4:00 combinatorial and Geometric Properties of Planar Laman Graphs

Stephen Kobourov, University of Arizona, USA; Torsten Ueckerdt, Charles University, Czech Republic; Kevin Verbeek, TU Eindhoven, The Netherlands



Tuesday, January 8



4:30-4:50 better bounds for Matchings in the streaming ModelMichael Kapralov, Stanford University, USA

4:55-5:15 space efficient streaming Algorithms for the Distance to Monotonicity and Asymmetric edit DistanceC. Seshadhri, Sandia National Laboratories,

USA; Michael Saks, Rutgers University, USA

5:20-5:40 (1+ε)-Approximation for Facility Location in Data streamsMorteza Monemizadeh, Karlsruhe Institute

of Technology, Germany; Artur Czumaj, University of Warwick, United Kingdom; Christiane Lammersen, Simon Fraser University, Canada; Christian Sohler, Technische Universität Dortmund, Germany

5:45-6:05 eigenvalues of a Matrix in the streaming ModelHuy Nguyen, Princeton University, USA;

Alexandr Andoni, Microsoft Research Silicon Valley, USA

6:10-6:30 beating the Direct sum theorem in communication complexity with Implications for sketchingMarco Molinaro, Carnegie Mellon

University, USA; David Woodruff, IBM Almaden Research Center, USA; Grigory Yaroslavtsev, Pennsylvania State University, USA

Tuesday, January 8



4:30-4:50 Faster Deterministic Fully-Dynamic Graph connectivityChristian Wulff-Nilsen, University of

Copenhagen, Denmark

4:55-5:15 Discrete convexity and Polynomial solvability in Minimum 0-extension ProblemsHiroshi Hirai, University of Tokyo, Japan

5:20-5:40 Mimicking Networks and succinct Representations of terminal cutsInbal Rika and Robert Krauthgamer,

Weizmann Institute of Science, Israel

5:45-6:05 Improved Algorithms for constructing consensus treesJesper Jansson, Kyoto University, Japan;

Chuanqi Shen, Stanford University, USA, and Raffles Institution, Singapore; Wing-Kin Sung, National University of Singapore, Singapore

6:10-6:30 efficient Algorithms for computing the triplet and Quartet Distance between trees of Arbitrary DegreeGerth S. Brodal, Aarhus University,

Denmark; Rolf Fagerberg, University of Southern Denmark, Denmark; Thomas Mailund, Christian N.S. Pedersen, and Andreas Sand, Aarhus University, Denmark

Tuesday, January 8



4:30-4:50 testing K-Modal Distributions: Optimal Algorithms Via ReductionsConstantinos Daskalakis, Massachusetts

Institute of Technology, USA; Ilias Diakonikolas, University of Edinburgh, United Kingdom; Rocco Servedio, Columbia University, USA; Gregory Valiant, Microsoft Research, USA; Paul Valiant, University of California, Berkeley, USA

4:55-5:15 Low-Distortion Inference of Latent similarities from a Multiplex social NetworkIttai Abraham, Microsoft Corporation, USA;

Shiri Chechik, Weizmann Institute of Science, Israel; David Kempe, University of Southern California, USA; Aleksandrs Slivkins, Microsoft Research, USA

5:20-5:40 Faster Walks in Graphs: A õ(n2) time-space trade-off for Undirected s-t connectivityAdrian Kosowski, INRIA Bordeaux Sud-Ouest,

France

5:45-6:05 Optimal Listing of cycles and st-Paths in Undirected GraphsEtienne Birmele, Universite d’Evry Val

d’Essone, Evry, France; Rui Ferreira, University of Pisa, Italy; Roberto Grossi, Universita di Pisa, Italy; Andrea Marino, Universita’ di Firenze, Italy; Nadia Pisanti, Universita di Pisa, Italy; Romeo Rizzi, Universita di Verona, Italy; Gustavo Sacomoto, INRIA Grenoble Rhône-Alpes, France

6:10-6:30 segmentation of trajectories on Non-Monotone criteriaBoris Aronov, Polytechnic Institute of New

York University, USA; Anne Driemel, Marc van Kreveld, Maarten Löffler, and Frank Staals, Utrecht University, The Netherlands


Abstracts

Abstracts for ALENEX and ANALCO appear in the beginning of the section, under "CP0",and are alphabetized by the speaker's last name.

Abstracts are printed as submitted by the authors.

24 DA13 Abstracts

IP1

”If You Can Specify It, You Can Analyze It” —TheLasting Legacy of Philippe Flajolet

The ”Flajolet School” of the analysis of algorithms andcombinatorial structures is centered on an effective calcu-lus, known as analytic combinatorics, for the developmentof mathematical models that are sufficiently accurate andprecise that they can be validated through scientific ex-perimentation. It is based on the generating function asthe central object of study, first as a formal object thatcan translate a specification into mathematical equations,then as an analytic object whose properties as a functionin the complex plane yield the desired quantitative results.Universal laws of sweeping generality can be proven withinthe framework, and easily applied. Standing on the shoul-ders of Cauchy, Polya, de Bruijn, Knuth, and many others,Philippe Flajolet and scores of collaborators developed thistheory and demonstrated its effectiveness in a broad rangeof scientific applications. Flajolet’s legacy is a vibrant fieldof research that holds the key not just to understandingthe properties of algorithms and data structures, but alsoto understanding the properties of discrete structures thatarise as models in all fields of science. This talk will surveyFlajolet’s story and its implications for future research.

Robert SedgewickPrinceton [email protected]

IP2

Submodular Functions and Their Applications

Submodular functions, a discrete analogue of convex func-tions, have played a fundamental role in combinatorial opti-mization since the 1970s. In the last decade, there has beenrenewed interest in submodular functions due to their in-terpretation as valuation functions of self-interested agentsin algorithmic game theory. These developments have ledto new questions as well as new algorithmic techniques.In this talk, we will discuss the concept of submodular-ity, its motivation and its unifying role in combinatorialoptimization, as well as the evolution of the relevant algo-rithmic techniques. We will survey the state of the art inoptimization of submodular functions, as well as selectedapplications in algorithmic game theory, social networksand machine learning, and some future challenges.

Jan VondrakIBM Almaden Research [email protected]

IP3

On Graph Property Testing, Arithmetic Progres-sions and Communication

Tools from Extremal Graph Theory are helpful in the studyof problems in Additive Number Theory, Theoretical Com-puter Science, and Information Theory. I will illustrate thisfact by several closely related examples focusing on a recentone in a joint work with Moitra and Sudakov. The maincombinatorial question addressed in this work is that ofestimating the maximum possible density of a graph con-sisting of a union of pairwise edge disjoint large inducedmatchings. It turns out that variants of this problem ap-pear in the investigation of several seemingly unrelated top-ics. The lecture will include a brief general introduction,a description of the methods used and results obtained, as

well as a discussion of the remaining challenges.

Noga AlonTel-Aviv [email protected]

CP0

Inducing Suffix and Lcp Arrays in External Mem-ory

We consider text index construction in external memory(EM). Our first contribution is an inducing algorithm forsuffix arrays in external memory. Practical tests show thatthis outperforms the previous best EM suffix sorter [De-mentiev et al., ALENEX 2005] by a factor of 2–3 in timeand I/O-volume. Our second contribution is to augmentthe first algorithm to also construct the array of longestcommon prefixes (LCPs). This yields the first EM con-struction algorithm for LCP arrays. The overhead in timeand I/O volume for this extended algorithm over plain suf-fix array construction is roughly two. Our algorithms scalefar beyond problem sizes previously considered in the lit-erature (text size of 80 GiB using only 4 GiB of RAM inour experiments).

Johannes FischerKIT76131 Karlsruhe, [email protected]

Timo Bingmann, Vitaly [email protected], [email protected]

CP0

The Number of Ways to Assemble a Graph

Motivated by the question of how macromolecules assem-ble, the notion of an assembly tree of a graph is introduced.Given a graph G, the paper is concerned with enumerat-ing the number of assembly trees of G. Explicit formulasor generating functions are provided for the number of as-sembly trees of several families of graphs. In some specialcases, we apply results of Zeilberger and Apagodu on mul-tivariate generating functions, and results of Wimp andZeilberger to deduce recurrence relations and very preciseasymptotic formulas for the number of assembly trees ofcomplete bipartite and tripartite graphs.

Miklos BonaUniversity of [email protected]

Andrew VinceDepartment of MathematicsUniversity of [email protected]

CP0

Bootstrap Percolation on Random GeometricGraphs

Bootstrap percolation has been used effectively to modelphenomena as diverse as emergence of magnetism in ma-terials, spread of infection, diffusion of software virusesin computer networks, adoption of new technologies, andemergence of collective action and cultural fads in humansocieties. It is defined on an (arbitrary) network of inter-

DA13 Abstracts 25

acting agents whose state is determined by the state oftheir neighbors according to a threshold rule. In a typicalsetting, bootstrap percolation starts by random and inde-pendent “activation’ of nodes with a fixed probability p, fol-lowed by a deterministic process for additional activationsbased on the density of active nodes in each neighborhood(θ activated nodes). Here, we study bootstrap percolationon random geometric graphs in the regime when the latterare (almost surely) connected. Random geometric graphsprovide an appropriate model in settings where the neigh-borhood structure of each node is determined by geograph-ical distance, as in wireless ad hoc and sensor networks aswell as in contagion. We derive bounds on the criticalthresholds p′c, p

′c such that for all p > p′c(θ) full percolation

takes place, whereas for p < p′c(θ) it does not. We concludewith simulations that compare numerical thresholds withthose obtained analytically.

Milan BradonjicMathematics of Networks and CommunicationsBell Labs, [email protected]

Iraj SanieeBell Labs, [email protected]

CP0

A Min-Edge Cost Flow Framework for CapacitatedCovering Problems

We present the Cov-MECF framework, a special case ofminimum-edge cost flow capturing several important cov-ering problems. We also introduce a heuristic LPO, whichharnesses the fractional relaxation to greedily build its so-lution. We empirically establish that LPO returns higherquality solutions than Wolsey’s classical greedy algorithm.We also observe that Leskovec et. al.’s optimization bene-fits LPO more than Wolsey’s algorithm, and that a freez-ing step heavily reduces the number of subroutine calls forboth.

Jessica ChangUniversity of [email protected]

Samir KhullerUniversity of Maryland, College [email protected]

CP0

Extremal Parameters in Sub-Critical GraphClasses

We analyze several extremal parameters like the diame-ter or the maximum degree in sub-critial graph classes.Sub-critical graph classes cover several well-known classesof graphs like trees, outerplanar graph or series-parallelgraphs which have been intensively studied during the lastfew years. However, this paper is the first one, where thesekinds of parameters are studied from a general point ofview.

Michael DrmotaTU [email protected]

Marc NoyUPC Barcelona

[email protected]

CP0

Fast Packed String Matching for Short Patterns

Searching for all occurrences of a pattern in a text is a fun-damental problem in computer science with applicationsin many other fields, like natural language processing, in-formation retrieval and computational biology. In the lasttwo decades a general trend has appeared trying to exploitthe power of the word RAM model to speed-up the per-formances of classical string matching algorithms. In thismodel an algorithm operates on words of length w, group-ing blocks of characters, and arithmetic and logic opera-tions on the words take one unit of time. In this paper weuse specialized word-size packed string matching instruc-tions, based on the Intel streaming SIMD extensions (SSE)technology, to design very fast string matching algorithmsin the case of short patterns. From our experimental resultsit turns out that, despite their quadratic worst case timecomplexity, the new presented algorithms become the clearwinners on the average for short patterns, when comparedagainst the most effective algorithms known in literature.

Simone FaroDipartimento di Matematica e InformaticaUniversita di [email protected]

Oguzhan KulekciNation. Res. Institute of Electronics and [email protected]

CP0

On Unifying the Space of �0-Sampling Algorithms

The problem of building an �0-sampler is to sample near-uniformly from the support set of a dynamic multiset. Thisproblem has a variety of applications within data analysis,computational geometry and graph algorithms. In this pa-per, we abstract a set of steps for building an �0-sampler,based on sampling, recovery and selection. We analyze theimplementation of an �0-sampler within this framework,and show how prior constructions of �0-samplers can all beexpressed in terms of these steps. Our experimental con-tribution is to provide a first detailed study of the accuracyand computational cost of �0-samplers.

Graham CormodeAT&T [email protected]

Donatella FirmaniSapienza University of [email protected]

CP0

Short and Simple Cycle Separators in PlanarGraphs

We provide an implementation of an algorithm that, givena triangulated planar graph with m edges, returns a simplecycle that is a 2/3–balanced separator consisting of at most√

8m edges. Cycle separators are essential in numerous pla-nar graph algorithms, e.g., for computing shortest paths,minimum cuts, or maximum flows. We demonstrate thatour algorithm is competitive with, and sometimes outper-forms other planar separator algorithms. Our implemen-

26 DA13 Abstracts

tation scales well to large graphs.

Eli Fox-EpsteinTufts [email protected]

Shay Mozes, Phitchaya PhothilimthanaMassachusetts Institute of [email protected], [email protected]

Christian [email protected]

CP0

Approximating Fault-Tolerant Domination in Gen-eral Graphs

We study the NP-complete problem of finding small k-dominating sets in general graphs, which allow k−1 nodesto fail and still dominate the graph. We show that theapproach of having at least k dominating nodes in theneighborhood of every node can be give solutions off bya multiplicative factor of nearly k and also show that agreedy algorithm achieves an almost tight approximationratio of ln(Δ + k) + 1.

Klaus-Tycho ForsterETH Zurich, Distributed Computing [email protected]

CP0

The Variance of the Number of 2-Protected Nodesin a Trie

We derive an asymptotic expression for the variance of thenumber of 2-protected nodes in a binary trie. In an unbi-ased trie on n leaves we find that the variance is approxi-mately .934n plus small fluctuations (of order n); but ourresult covers the general (biased) case as well. Our proofrelies on the asymptotic similarities between a trie and itsPoissonized counterpart, using Mellin transform and sin-gularity analysis.

Jeffrey B. GaitherPurdue UniversityDepartment of [email protected]

Mark Daniel WardPurdue UniversityDepartment of [email protected]

CP0

When Is It Worthwhile to Propagate a Constraint?A Probabilistic Analysis of Alldifferent

We present new work on analyzing the behaviour of a con-straint solver, with a view towards optimization. We pro-pose to quantify useless calls to propagator for the ALLD-IFFERENT constraint (bound consistency propagator).Our first contribution is the definition of a probabilisticmodel for the constraint and its variables. This model isused to compute the probability that a call to the propa-gation algorithm for ALLDIFFERENT modifies its input,

then give an asymptotic approximation of it.

Daniele Gardy

University of Versailles Saint-Quentin (France)[email protected]

Jeremie Du BoisberrangerPRiSM, University of Versailles [email protected]

Xavier LorcaEcole des Mines, [email protected]

Charlotte TruchetLINA, University of [email protected]

CP0

Exact-Size Sampling for Motzkin Trees in Lin-ear Time Via Boltzmann Samplers and HolonomicSpecification

Given a combinatorial class and an algebraic combinato-rial specification, one can automatically build a so-calledBoltzmann random sampler for that class. We introducea Boltzmann sampler for Motzkin trees built from a holo-nomic specification, that is, a specification that uses thepointing operator. This sampler is inspired by Rmy’s algo-rithm on binary trees. We show that our algorithm gives anexact size sampler with a linear time and space complexityin average.

Axel Bacher, Olivier Bodini, Alice JacquotLIPN, Universite Paris [email protected], [email protected], [email protected]

CP0

The Cost of Address Translation

Modern computers with memory hierarchy, multiple cores,and virtual memory are not random access machines(RAMs). Hence, the analysis of some algorithms (ran-dom scan of an array, binary search, heapsort) in the RAMmodel does not correctly predict growth rates of actual run-ning times. We propose the VAT model (virtual addresstranslation) that accounts for the cost of address transla-tions. Then, we analyze the VAT-cost of the algorithmsmentioned above, cache-oblivious algorithms, and others.

Tomasz JurkiewiczMax Planck Institute for InformaticsThe Saarbrucken Graduate School of Computer [email protected]

Kurt MehlhornMPI for InformaticsSaarbruecken, [email protected]

CP0

3D Kinetic Alpha Complexes and Their Implemen-tation

Motivated by an application in cell biology, we describean extension of the kinetic data structures framework fromDelaunay triangulations to fixed-radius alpha complexes.

DA13 Abstracts 27

Our algorithm is implemented using Cgal, following theexact geometric computation paradigm. We report on sev-eral techniques to accelerate the computation that turnour implementation applicable to the underlying biologicalproblem.

Michael Kerber, Herbert EdelsbrunnerInstitute of Science and Technology (Austria)[email protected], [email protected]

CP0

Unlabeled Equivalence for Matroids Representableover Finite Fields

Two matrices representing the same matroid over a finitefield are projective equivalent if one can be obtained fromthe other by elementary row operations and column scal-ing. Bounds for the number of projective inequivalent rep-resentations are known for only a few classes of matroids.We define two matrices to be geometric equivalent if col-umn permutations are also allowed, and show that froman algorithmic perspective this is the better definition ofequivalence for matroids.

Sandra R. KinganBrooklyn [email protected]

CP0

On Delta-Method of Moments and ProbabilisticSums

Abstract not available at time of publication.

Jacek Cichon, Zbigniew Golebiewski, Marcin Kardas,Marek KlonowskiWroclaw University of [email protected],[email protected], n/a,[email protected]

CP0

Analysis of Parameters of Trees Corresponding toHuffman Codes and Sums of Unit Fractions

For fixed t ≥ 2, we consider the class of representations of 1as sum of unit fractions whose denominators are powers oft or equivalently the class of canonical compact t-ary Huff-man codes or equivalently rooted t-ary plane “canonical’trees. We study the probabilistic behaviour of the height(limit distribution is shown to be normal), the number ofdistinct summands (normal distribution), the path length(normal distribution), the width (main term of the expecta-tion and concentration property) and the number of leavesat maximum distance from the root (discrete distribution).

Clemens HeubergerAlpen-Adria-Universitat Klagenfurt, [email protected]

Daniel KrennTU [email protected]

Stephan Wagner

Stellenbosch University, South [email protected]

CP0

Practical Batch-Updatable External Hashing withSorting

This paper presents a practical external hashing schemethat supports fast lookup (7 microseconds) for largedatasets (millions to billions of items) with a small mem-ory footprint (2.5 bits/item) and fast index construction(151 K items/s for 1-KiB key-value pairs). Our schemecombines three key techniques: (1) a new index data struc-ture (Entropy-Coded Tries); (2) the use of sorting as themain data manipulation method; and (3) support for in-cremental index construction for dynamic datasets. Weevaluate our scheme by building an external dictionary onflash-based drives and demonstrate our scheme’s high per-formance, compactness, and practicality.

Hyeontaek Lim, David AndersenCarnegie Mellon [email protected], [email protected]

Michael KaminskyIntel [email protected]

CP0

Perpetuities in Fair Leader Election Algorithms

We consider a broad class of fair leader election algo-rithms, and study the duration of contestants (the numberof rounds a randomly selected contestant stays in the com-petition) and the overall cost of the algorithm. We givesufficient conditions for the duration to have a geometriclimit distribution (a perpetuity built from Bernoulli ran-dom variables), and for the limiting distribution of the totalcost (after suitable normalization) to be a perpetuity. Forthe duration, the proof is established via convergence (to 0)of the first-order Wasserstein distance from the geometriclimit. For the normalized overall cost, the method of proofis also convergence of the first-order Wasserstein distance,augmented with an argument based on a contraction map-ping in the first-order Wasserstein metric space to showthat the limit approaches a unique fixed-point solution ofa perpetuity distributional equation. (The two steps arecommonly referred to as the contraction method.)

Hosam M. Mahmoud, Ravi KalpathyThe George Washington [email protected], [email protected]

CP0

Robust Gossip-Based Aggregation: A PracticalPoint of View

Several existing gossip-based aggregation algorithms havethe potential for providing resilience in failure-prone dis-tributed systems. In this paper, we evaluate the algorithmswith the most promising theoretical fault tolerance prop-erties experimentally in faulty environments. Our exper-iments reveal several shortcomings of existing approachesin terms of performance as well as in terms of numerical ac-curacy. We explain this behavior observed in practice andmotivate further theoretical research to efficiently meet the

28 DA13 Abstracts

desired robustness requirements.

Gerhard NiederbruckerUniversity of [email protected]

Wilfried N. GanstererUniversity of ViennaDepartment of Computer Science and [email protected]

CP0

Lempel-Ziv Factorization: Simple, Fast, Practical

For decades the Lempel-Ziv (LZ77) factorization has been acornerstone of data compression and string processing, anduses for it are still being uncovered. For example, LZ77 iscentral to recent index data structures for highly repetitivecollections. However, in many applications factorizationremains a computational bottleneck. We describe severalfast factorization algorithms which consistently outperformall previous approaches by almost a factor of two, use lessmemory, and still offer strong worstcase performance guar-antees.

Simon J. Puglisi, Dominik KempaDepartment of Computer ScienceUniversity of [email protected], [email protected]

CP0

Approximate Counting of Matchings in Sparse Uni-form Hypergraphs

In this paper we give a fully polynomial randomized ap-proximation scheme (FPRAS) for the number of matchingsin k-uniform hypergraphs whose intersection graphs con-tain few claws. Our method gives a generalization of thecanonical path method of Jerrum and Sinclair to hyper-graphs satisfying a local restriction. The proof depends onan application of the Euler tour technique for the canonicalpaths of the underlying Markov chains. On the other hand,we prove that it is NP-hard to approximate the number ofmatchings even for the class of 2-regular, linear, k-uniformhypergraphs, for all k ≥ 6, without the above restriction.

Edyta SzymanskaAdam Mickiewicz University, Poznan, [email protected]

Andrzej RucinskiAdam Mickiewicz University, Poznan, PolandEmory University, [email protected]

Marek KarpinskiDepartment of Computer ScienceUniversity of [email protected]

CP0

Computational Topology and Normal Surfaces:Theoretical and Experimental Complexity Bounds

In three-dimensional computational topology, the theory ofnormal surfaces is a tool of great theoretical and practicalsignificance. Although this theory typically leads to ex-

ponential time algorithms, very little is known about howthese algorithms perform in “typical’ scenarios, or how farthe best known theoretical bounds are from the real worst-case scenarios. Here we study the combinatorial and alge-braic complexity of normal surfaces from both the theoret-ical and experimental viewpoints. Theoretically, we obtainnew exponential lower bounds on the worst-case complex-ities in a variety of settings that are important for practi-cal computation. Experimentally, we study the worst-caseand average-case complexities over a comprehensive bodyof roughly three billion input triangulations. Many of ourlower bounds are the first known exponential lower boundsin these settings, and experimental evidence suggests thatmany of our theoretical lower bounds on worst-case growthrates may indeed be asymptotically tight.

Benjamin A. Burton, Jonathan SpreerThe University of [email protected], [email protected]

Joo PaixoPontifcia Universidade Catolica, Rio de [email protected]

CP0

Polynomial-Time Construction of Contraction Hi-erarchies for Multi-Criteria Objectives

We consider multi-criteria shortest path problems andshow that contraction hierarchies – a speed-up techniqueoriginally developed for standard shortest path queries– can be constructed efficiently for the case of linearlyweighted combinations of the edge costs. This extends pre-vious results which considered only the bicriteria case. Onthe theory side we prove a polynomial time bound via somepolyhedral considerations; experiments complement theseresults by showing the practicability of our approach.

Stefan FunkeFMI, University of Stuttgart, [email protected]

Sabine StorandtUniversity of [email protected]

CP0

On Parallelizing Matrix Multiplication by theColumn-Row Method

We consider the problem of sparse matrix multiplicationby the column row method in a distributed setting wherethe matrix product is not necessarily sparse. We presenta surprisingly simple method for “consistent’ parallel pro-cessing of sparse outer products (column-row vector prod-ucts) over several processors, in a communication-avoidingsetting where each processor has a copy of the input. Themethod is consistent in the sense that a given output entryis always assigned to the same processor independently ofthe specific structure of the outer product. We show guar-antees on the work done by each processor, and achievelinear speedup down to the point where the cost is domi-nated by reading the input. Our method gives a way of dis-tributing (or parallelizing) matrix product computations insettings where the main bottlenecks are storing the resultmatrix, and inter-processor communication. Motivated byobservations on real data that often the absolute valuesof the entries in the product adhere to a power law, wecombine our approach with frequent items mining algo-

DA13 Abstracts 29

rithms and show how to obtain a tight approximation ofthe weight of the heaviest entries in the product matrix. Asa case study we present the application of our approach tofrequent pair mining in transactional data streams, a prob-lem that can be phrased in terms of sparse {0, 1}-integermatrix multiplication by the column-row method. Experi-mental evaluation of the proposed method on real-life datasupports the theoretical findings.

Konstantin Kutzkov, Andrea Campagna, Rasmus Pagh,Morten StockelIT University of [email protected], [email protected], [email protected], [email protected]

CP0

Efficient Algorithms for Dualizing Large-Scale Hy-pergraphs

A hypergraph F is a set family defined on vertex set V .The dual of F is the set of minimal subsets H of V suchthat F ∩H �= ∅ for any F ∈ F . We propose efficient dual-ization algorithms using a new set system induced by theminimality condition, and efficient methods for the mini-mality check. The computational experiments show thatour algorithms outperform to existing algorithms drasti-cally.

Takeaki UnoNational Institute of [email protected]

Keisuke MurakamiAoyama Gakuin [email protected]

CP0

Engineering Java 7’s Dual Pivot Quicksort UsingMalijan

Recent results on Java 7’s dual pivot Quicksort have re-vealed its highly asymmetric nature. These insights sug-gest that asymmetric pivot choices are preferable to sym-metric ones for this Quicksort variant. From a theoreticalpoint of view, this should allow us to improve on the cur-rent implementation in Oracle’s Java 7 runtime library. Inthis paper, we use our new tool MaLiJAn to confirm this forasymptotic combinatorial cost measures such as the totalnumber of executed instructions. However, the observedrunning times show converse behavior. With the supportof data provided by MaLiJAn we are able to identify theprofiling capabilities of Oracle’s just-in-time compiler to beresponsible for this unexpected outcome.

Sebastian Wild, Markus Nebel, Raphael Reitzig, UlrichLaubeComputer Science DepartmentUniversity of Kaiserslauterns [email protected], [email protected], r [email protected], [email protected]

CP1

Convergence of Multivariate Belief Propagation,with Applications to Cuckoo Hashing and LoadBalancing

This paper is motivated by two applications, namely i) gen-eralizations of cuckoo hashing, a computationally simpleapproach to assigning keys to objects, and ii) load balanc-

ing in content distribution networks, where one is inter-ested in determining the impact of content replication onperformance. These two problems admit a common ab-straction: in both scenarios, performance is characterizedby the maximum weight of a generalization of a matchingin a bipartite graph, featuring node and edge capacities.Our main result is a law of large numbers characterizingthe asymptotic maximum weight matching in the limit oflarge bipartite random graphs, when the graphs admit alocal weak limit that is a tree. This result specializes tothe two application scenarios, yielding new results in bothcontexts. In contrast with previous results, the key noveltyis the ability to handle edge capacities with arbitrary in-teger values. An analysis of belief propagation algorithms(BP) with multivariate belief vectors underlies the proof.In particular, we show convergence of the correspondingBP by exploiting monotonicity of the belief vectors withrespect to the so-called upshifted likelihood ratio stochasticorder. This auxiliary result can be of independent interest,providing a new set of structural conditions which ensureconvergence of BP.

Mathieu LeconteTechnicolor - [email protected]

Marc [email protected]

Laurent MassoulieInria, Microsoft research - Inria Joint [email protected]

CP1

Mixing Times of Markov Chains for Self-Organizing Lists and Biased Permutations

We study the mixing time of a Markov chain M\\ on per-mutations in the non-uniform setting. We are given “pos-itively biased” probabilities {pi,j ≥ 1/2} for all i < j andlet pj,i = 1 − pi,j . In each step M\\ chooses two adjacentelements k, and � and exchanges their positions with prob-ability p�,k. Here we define two general classes and givethe first proofs that the chain is rapidly mixing for both.We also demonstrate that the chain can mix slowly.

Prateek Bhakta, Sarah Miracle, Dana RandallGeorgia Institute of [email protected], [email protected],[email protected]

Amanda StreibNational Institute of Standards and [email protected]

CP1

Balls into Bins Via Local Search

We propose a natural process for allocating n balls inton bins that are organized as the vertices of an undirectedgraph G. Each ball first chooses a vertex u in G uniformlyat random. Then the ball performs a local search in Gstarting from u until it reaches a vertex with local min-imum load, where the ball is finally placed on. In ourmain result, we prove that this process yields a maximumload of only Θ(log log n) on expander graphs. In addition,we show that for d-dimensional grids the maximum load

30 DA13 Abstracts

is Θ((

log nlog log n

) 1d+1

). Finally, for almost regular graphs

with minimum degree Ω(log n), we prove that the maxi-mum load is constant and also reveal a fundamental dif-ference between random and arbitrary tie-breaking rules.

Paul BogdanCarnegie Mellon [email protected]

Thomas SauerwaldMax Planck Institute for Computer [email protected]

Alexandre StaufferMicrosoft [email protected]

He SunMax Planck Institute for [email protected]

CP1

Approximate Counting Via Correlation Decay onPlanar Graphs

We show for a broad class of counting problems, correlationdecay (strong spatial mixing) implies FPTAS on planargraphs. The framework for the counting problems consid-ered by us is the Holant problems with arbitrary constant-size domain and symmetric constraint functions. We definea notion of regularity on the constraint functions, whichcovers a wide range of natural and important countingproblems, including all multi-state spin systems, count-ing graph homomorphisms, counting weighted matchingsor perfect matchings, the subgraphs world problem trans-formed from the ferromagnetic Ising model, and all count-ing CSPs and Holant problems with symmetric constraintfunctions of constant arity. The core of our algorithm isa fixed-parameter tractable algorithm which computes theexact values of the Holant problems with regular constraintfunctions on graphs of bounded treewidth. By utilizingthe locally tree-like property of apex-minor-free families ofgraphs, the parameterized exact algorithm implies an FP-TAS for the Holant problem on these graph families when-ever the Gibbs measure defined by the problem exhibitsstrong spatial mixing. We further extend the recursive cou-pling technique to Holant problems and establish strongspatial mixing for the ferromagnetic Potts model and thesubgraphs world problem. As consequences, we have newdeterministic approximation algorithms on planar graphsand all apex-minor-free graphs for several counting prob-lems.

Yitong YinNanjing [email protected]

Chihao ZhangShanghai Jiaotong [email protected]

CP1

Correlation Decay Up to Uniqueness in Spin Sys-tems

We give a complete characterization of the two-state anti-ferromagnetic spin systems which exhibit strong spatial

mixing on general graphs. We show that a two-stateanti-ferromagnetic spin system exhibits strong spatial mix-ing on all graphs of maximum degree at most Δ if andonly if the system has a unique Gibbs measure on infiniteregular trees of degree up to Δ, where Δ can be eitherbounded or unbounded. As a consequence, there existsan FPTAS for the partition function of a two-state anti-ferromagnetic spin system on graphs of maximum degreeat most Δ when the uniqueness condition is satisfied oninfinite regular trees of degree up to Δ. In particular, anFPTAS exists for arbitrary graphs if the uniqueness is sat-isfied on all infinite regular trees. This covers as specialcases all previous algorithmic results for two-state anti-ferromagnetic systems on general-structure graphs. Com-bining with the FPRAS for two-state ferromagnetic spinsystems of Jerrum-Sinclair and Goldberg-Jerrum-Paterson,and the very recent hardness results of Sly-Sun and inde-pendently of Galanis-Stefankovic-Vigoda, this gives a com-plete classification, except at the phase transition bound-ary, of the approximability of all two-state spin systems,on either degree-bounded families of graphs or family ofall graphs.

Liang LiPeking [email protected]

Pinyan LuMicrosoft Research [email protected]

Yitong YinNanjing [email protected]

CP2

Online Mixed Packing and Covering

Recent advances in online optimization have heavily reliedon algorithmic techniques for obtaining near-optimal so-lutions to LPs that are revealed online. We give the firstpolylogarithmic-competitive algorithm for obtaining a frac-tional solution to mixed packing and covering LPs online,where packing constraints are given offline and coveringconstraints arrive online. Using this generic technique, wegive near-optimal algorithms for two canonical problems inthis class: unrelated machine scheduling with startup costsand capacity-constrained facility location.

Yossi AzarTel-Aviv [email protected]

Umang BhaskarCalifornia Institute of [email protected]

Lisa FleischerDartmouth [email protected]

Debmalya PanigrahiMicrosoft [email protected]

CP2

Randomized Primal-Dual Analysis of Ranking for

DA13 Abstracts 31

Online Bipartite Matching

We give a simple proof that the algorithm of Karp, Vazi-rani and Vazirani is 1-1/e competitive for the online bi-partite matching problem. The proof is via a random-ized primal-dual argument. Primal-dual algorithms havebeen successfully used for many online algorithm problems,but the dual constraints are always satisfied deterministi-cally. This is the first instance of a non-trivial randomizedprimal-dual algorithm in which the dual constraints onlyhold in expectation. The approach also generalizes eas-ily to the vertex-weighted version considered by Agarwalet al. Further we show that the proof is very similar tothe deterministic primal-dual argument for the online bud-geted allocation problem with small bids (also called theAdWords problem) of Mehta et al.

Nikhil R. DevanurMicrosoft Research, [email protected]

Kamal JainEbay [email protected].

Robert KleinbergCornell [email protected]

CP2

Matroid Secretary for Regular and DecomposableMatroids

In the matroid secretary problem we are given a Matroidand a stream of elements. The goal is to choose an indep-ndent set with maximum profit. If we choose to acceptan element when it is presented by the stream then we cannever get rid of it, and if we choose not to accept it then wecannot later add it. We give O(1)-competitive algorithmsfor regular matroids and Min-cut max flow matroids

Michael DinitzComputer Science,[email protected]

Guy KortsarzComputer [email protected]

CP2

Weighted Flowtime on Capacitated Machines

Motivated by modern systems which run multiplejobs/VMs at a time subject to overall CPU, memory,network, and disk capacity constraints, we give the firstscheduling algorithms to minimize weighted flow time oncapacitated machines. We give several positive and nega-tive results, including a near-optimal (1+ε)-speed, (1+ε)-capacity O(1/ε3)-competitive algorithm using a novel com-bination of knapsacks, densities, job classification into cat-egories, and potential function methods. Our results alsoextend to the multiple unrelated machines setting.

Kyle FoxUniversity of Illinois, [email protected]

Madhukar KorupoluGoogle Research

[email protected]

CP2

A New Approach to Online Scheduling: Approxi-mating the Optimal Competitive Ratio

We propose a new approach to competitive analysis inonline scheduling by introducing the novel concept ofcompetitive-ratio approximation schemes. Such a schemealgorithmically constructs an online algorithm with a com-petitive ratio arbitrarily close to the best possible compet-itive ratio for any online algorithm. We study the problemof scheduling jobs online to minimize the weighted sum ofcompletion times on parallel, related, and unrelated ma-chines, and we derive both deterministic and randomizedalgorithms which are almost best possible among all onlinealgorithms of the respective settings.

Andreas WieseMPII [email protected]

Elisabeth Gunther, Olaf Maurer, Nicole MegowTU [email protected], [email protected],[email protected]

CP3

Computing the Discrete Frechet Distance in Sub-quadratic Time

The discrete Frechet distance is a similarity measure be-tween two sequences of points A and B, that takes intoaccount the location and ordering of the points along thesequences. We present a subquadratic algorithm for com-puting the discrete Frechet distance between two sequencesof points in the plane, of respective lengths m and n. Thealgorithm runs in O

(mn log logn

log n

)time and uses O(n+ m)

storage.

Pankaj AgarwalDuke [email protected]

Rinat Ben AvrahamTel Aviv [email protected]

Haim KaplanTel-Aviv [email protected]

Micha SharirTel Aviv University & New York [email protected]

CP3

The Complexity of Detecting Taut Angle Struc-tures on Triangulations

There are many fundamental algorithmic problems on tri-angulated 3-manifolds whose complexities are unknown.Here we study the problem of finding a taut angle structureon a 3-manifold triangulation, whose existence has impli-cations for both the geometry and combinatorics of thetriangulation. We prove that detecting taut angle struc-tures is NP-complete, but also fixed-parameter tractable

32 DA13 Abstracts

in the treewidth of the face pairing graph of the triangu-lation. These results have deeper implications: the coretechniques can serve as a launching point for approachingdecision problems such as unknot recognition and primedecomposition of 3-manifolds.

Benjamin A. Burton, Jonathan SpreerThe University of [email protected], [email protected]

CP3

Approximate Shortest Descending Paths

We present an approximate algorithm for the shortest de-scending path (SDP) problem. Given a source s and adestination t in a polygonal terrain T , an SDP from s tot is a path in T of minimum Euclidean length subject tothe constraint that the height decreases monotonically aswe traverse that path from s to t. Given any ε ∈ (0, 1),our algorithm returns in O(n4 log(n/ε)) time a descendingpath of length at most 1 + ε times the optimum. This isthe first algorithm whose running time is polynomial in nand log(1/ε) and independent of the terrain geometry.

Siu-Wing ChengThe Hong Kong University of Science and [email protected]

Jiongxin JinThe Hong Kong University of Science and [email protected]

CP3

An Infinite Class of Sparse-Yao Spanners

Given a finite set P of points in the plane and a fixed integerk > 1, the Yao graph Yk for P is obtained by extending kequally spaced rays from each point u ∈ P , and connectingu to a closest neighbor in each sector formed by these rays.It has been shown that, Yk is a t-spanner, for any k ≥ 6 andsome real constant t > 1 depending on k, meaning that ithas a path between each pair of points in P no longer thant times the Euclidean distance between the two points. Toreduce the potential high in-degree of Yk, the Sparse-Yaograph Y Yk eliminates from Yk all but a shortest incomingedge in each sector. In this paper we establish that theSparse-Yao graph defined by integer parameter 6k, withk ≥ 6, is an 11.67-spanner.

Mirela Damian, Matthew BauerVillanova [email protected], [email protected]

CP3

Weighted Graph Laplace Operator under Topolog-ical Noise

Recently, various applications have motivated the study ofspectral structures (eigenvalues and eigenfunctions) of theso-called Laplace-Beltrami operator of a manifold and theirdiscrete versions. A popular choice for the discrete versionis the so-called Gaussian weighted graph Laplacian whichcan be applied to point cloud data that samples a mani-fold. Naturally, the question of stability of the spectrumof this discrete Laplacian under the perturbation of thesampled manifold becomes important for its practical us-age. Previous results showed that the spectra of both themanifold Laplacian and discrete Laplacian are stable when

the perturbation is “nice’ in the sense that it is restrictedto a diffeomorphism with minor area distortion. However,this forbids, for example, small topological changes. Westudy the stability of the spectrum of the weighted graphLaplacian under more general perturbations. In particu-lar, we allow arbitrary, including topological, changes tothe hidden manifold as long as they are localized in theambient space and the area distortion is small. ManifoldLaplacians may change dramatically in this case. Never-theless, we show that the weighted graph Laplacians com-puted from two sets of points, uniformly randomly sampledfrom a manifold and a perturbed version of it, have simi-lar spectra. The distance between the two spectra can bebounded in terms of the size of the perturbation and someintrinsic properties of the original manifold.

Tamal Dey, Pawas RanjanThe Ohio State [email protected], [email protected]

Yusu WangThe Ohio State University, Columbus, [email protected]

CP4

Compressed Static Functions with Applications

Given a set of integer keys from a bounded universe alongwith associated data, the dictionary problem asks to an-swer two queries: membership and retrieval. Membershiphas to tell whether a given element is in the dictionary ornot; Retrieval has to return the data associated with thesearched key. In this paper we provide time and space op-timal solutions for three well-established relaxations of thisbasic problem: (Compressed) Static functions, Approxi-mate membership and Relative membership.

Djamal BelazzouguiLIAFA, Univ. Paris Diderot - Paris [email protected]

Rossano VenturiniUniversity of [email protected]

CP4

Near-Optimal Range Reporting Structures for Cat-egorical Data

Range reporting on categorical (or colored) data is a well-studied generalization of the classical range reporting prob-lem in which each of the N input points has an associatedcolor (category). A query then asks to report the set ofcolors of the points in a given rectangular query range,which may be far smaller than the set of all points in thequery range. We study two-dimensional categorical rangereporting in both the word-RAM and I/O-model. For theI/O-model, we present two alternative data structures forthree-sided queries. The first answers queries in optimalO(lgB N +K/B) I/Os using O(N lg∗N) space, where K isthe number of distinct colors in the output, B is the diskblock size, and lg∗N is the iterated logarithm of N . Oursecond data structure uses linear space and answers queriesin O(lgB N +lg(h)N +K/B) I/Os for any constant integer

h ≥ 1. Here lg(1)N = lgN and lg(h)N = lg(lg(h−1)N)when h > 1. Both solutions use only comparisons on thecoordinates. We also show that the lgB N terms in thequery costs can be reduced to optimal lg lgB U when the

DA13 Abstracts 33

input points lie on a U × U grid and we allow word-levelmanipulations of the coordinates. We further reduce thequery time to just O(1) if the points are given on an N×Ngrid. Both solutions also lead to improved data structuresfor four-sided queries. For the word-RAM, we obtain opti-mal data structures for three-sided range reporting, as wellas improved upper bounds for four-sided range reporting.Finally, we show a tight lower bound on one-dimensionalcategorical range counting using an elegant reduction from(standard) two-dimensional range counting.

Kasper G. LarsenMADALGO, Aarhus [email protected]

Freek Van [email protected]

CP4

Twisted Tabulation Hashing

We introduce a new tabulation-based hashing schemecalled “twisted tabulation’. It is essentially as simple andfast as simple tabulation, but has some powerful distribu-tional properties illustrating its promise:

1. If we sample keys with arbitrary probabilities, thenwith high probability, the number of samples in-side any subset is concentrated exponentially. Withbounded independence we only get polynomial con-centration, and with simple tabulation, we have nogood bound even in the basic case of tossing an (un-biased) coin for each key.

2. With classic hash tables such as linear probing andcollision-chaining, a window of B operations takesO(B) time with high probability, for B = Ω(lg n).Good amortized performance over any window of sizeB is equivalent to guaranteed throughput for an on-line system processing a stream via a buffer of size B(e.g., Internet routers).

Mihai PatrascuAT&T Labs - ResearchDied June 5, 2012mip@heaven

Mikkel ThorupAT&T Labs - Research and University of [email protected]

CP4

The Space Complexity of 2-Dimensional Approxi-mate Range Counting

We study the problem of 2-dimensional orthogonal rangecounting with additive error. Given a set P of n pointsdrawn from an n × n grid and an error parameter ε, thegoal is to build a data structure, such that for any orthog-onal range R, the data structure can return the number ofpoints in P ∩ R with additive error εn. A well-known so-lution for this problem is the ε-approximation. Informallyspeaking, an ε-approximation of P is a subset A ⊆ P thatallows us to estimate the number of points in P ∩ R bycounting the number of points in A∩R. It is known that anε-approximation of size O( 1

εlog2.5 1

ε) exists for any P with

respect to orthogonal ranges, and the best lower bound isΩ( 1

εlog 1

ε). The ε-approximation is a rather restricted data

structure, as we are not allowed to store any information

other than the coordinates of a subset of points in P . Inthis paper, we explore what can be achieved without anyrestriction on the data structure. We first describe a datastructure that uses O( 1

εlog 1

εlog log 1

εlog n) bits that an-

swers queries with error εn. We then prove a lower boundthat any data structure that answers queries with errorO(log n) must use Ω(n log n) bits. This lower bound hastwo consequences: 1) answering queries with error O(log n)is as hard as answering the queries exactly; and 2) our up-per bound cannot be improved in general by more than anO(log log 1

ε) factor.

Zhewei WeiCenter for Massive Data Algorithmics (MADALGO)Aarhus [email protected]

Ke YiThe Hong Kong University of Science and [email protected]

CP4

Adaptive and Approximate Orthogonal RangeCounting

We consider 2-D orthogonal range counting, a basic prob-lem in geometric data structures, under the w-bit wordRAM model. There are linear-space data structures for thisproblem with optimal worst-case query timeO(logw n). Wegive an O(n log log n)-space adaptive data structure withquery time O(log log n + logw k), where k is the outputcount. We also give an efficient data structure for approxi-mate counting and an optimal adaptive data structure forthe closely related 1-D range selection problem.

Bryan T. WilkinsonMADALGO, Aarhus [email protected]

Timothy ChanDavid R. Cheriton School of Computer ScienceUniversity of [email protected]

CP5

Better Balance by Being Biased: A 0.8776-Approximation for Max Bisection

We present an approximation algorithm for Max-Bisectionthat achieves ratio 0.8766. As Max-Bisection is hard toapproximate within αGW + ε ≈ 0.8786 under the UniqueGames Conjecture (UGC), our algorithm is nearly opti-mal. We also obtain an optimal algorithm (assuming theUGC) for the analogous variant of Max-2-Sat. Our approx-imation ratio for this problem exactly matches the optimalapproximation for Max-2-Sat, i.e. αLLZ + ε ≈ 0.9401.

Konstantinos GeorgiouUniversity of [email protected]

Per AustrinKTH Royal Institute of [email protected]

Siavosh BenabbasUniversity of [email protected]

34 DA13 Abstracts

CP5

Local Distribution and the Symmetry Gap: Ap-proximability of Multiway Partitioning Problems

We study the approximability of multiway partitioningproblems, examples of which include Multiway Cut, Node-weighted Multiway Cut, and Hypergraph Multiway Cut.We investigate these problems from the point of view of twopossible generalizations: as Min-CSPs, and as SubmodularMultiway Partition problems. These two generalizationslead to two natural relaxations that we call respectivelythe Local Distribution LP, and the Lovasz relaxation. TheLocal Distribution LP is generally stronger than the Lovaszrelaxation, but applicable only to Min-CSP with predicatesof constant size. The relaxations coincide in some casessuch as Multiway Cut where they are both equivalent to theCKR relaxation. We show that the Lovasz relaxation givesa (2−2/k)-approximation for Submodular Multiway Parti-tion with k terminals, improving a recent 2-approximation.We prove that this factor is optimal in two senses: (1) A(2−2/k−ε)-approximation for Submodular Multiway Par-tition with k terminals would require exponentially manyvalue queries (in the oracle model), or imply NP = RP(for certain explicit submodular functions). (2) For Hy-pergraph Multiway Cut and Node-weighted Multiway Cutwith k terminals, both special cases of Submodular Multi-way Partition, we prove that a (2−2/k−ε)-approximationis NP-hard, assuming the Unique Games Conjecture. Bothour hardness results are more general: (1) We show thatthe notion of symmetry gap, previously used for submod-ular maximization problems, also implies hardness resultsfor submodular minimization problems. (2) Assuming theUnique Games Conjecture, we show that the Local Distri-bution LP gives an optimal approximation for every Min-CSP that includes the Not-Equal predicate.

Alina EneUniversity of Illinois at [email protected]


Yi WuPurdue [email protected]

CP5

Poly-Logarithmic Approximation for MaximumNode Disjoint Paths with Constant Congestion

We consider the Maximum Node Disjoint Paths (MNDP)problem in undirected graphs. The input consists ofan undirected graph G = (V,E) and a collection{(s1, t1), . . . , (sk, tk)} of k source-sink pairs. The goal isto select a maximum cardinality subset of pairs that canbe routed/connected via node-disjoint paths. A relaxedversion of MNDP allows up to c paths to use a node,where c is the congestion parameter. We give a polyno-mial time algorithm that routes Ω(OPT/polylog(k)) pairswith O(1) congestion, where OPT is the value of an opti-mum fractional solution to a natural multicommodity flowrelaxation. Our result builds on the recent breakthroughof Chuzhoy who gave the first poly-logarithmic approxi-mation with constant congestion for the Maximum EdgeDisjoint Paths (MEDP) problem.

Alina Ene, Chandra ChekuriUniversity of Illinois at Urbana-Champaign

[email protected], [email protected]

CP5

How to Sell Hyperedges: The Hypermatching As-signment Problem

In the Hypermatching Assignment Problem (HAP), we aregiven a set of clients with budget constraints and a set ofindivisible items. Each client is willing to buy one or morebundles of (at most) k items each (bundles can be seen ashyperedges in a k-hypergraph). If client i gets a bundlee, she pays bi,e and yields a net profit wi,e. The goal isto assign a set of pairwise disjoint bundles to clients so asto maximize the total profit while respecting the budgets.HAP generalizes both the weighted k-hypergraph matchingproblem and the Generalized Assignment Problem (GAP).We give a randomized (k+ 1 + ε) approximation algorithmfor HAP, which is based on rounding the 1-round Lasserrestrengthening of a novel LP. Furthermore we obtain betterapproximation ratios for special cases of the problem.

Fabrizio Grandoni, Marek Cygan, Monaldo MastrolilliIDSIA, University of [email protected], [email protected], [email protected]

CP5

Approximating Non-Uniform Sparsest Cut ViaGeneralized Spectra

We give an approximation algorithm for non-uniformsparsest cut with the following guarantee: For any ε, δ ∈(0, 1), given cost and demand graphs, we can find a set T ⊆V whose non-uniform sparsity is at most 1+ε

δtimes the op-

timal non-uniform sparsest cut value, in time 2r/(δε)nO(1)

provided the rth minimum generalized eigenvalue satisfiesλr ≥ Φ∗/(1 − δ). In words, we show that the non-uniformsparsest cut problem is easy when the generalized spectrumgrows moderately fast.

Venkatesan Guruswami, Ali K. SinopCarnegie Mellon [email protected], [email protected]

CP6

Shortest Non-Trivial Cycles in Directed and Undi-rected Surface Graphs

Let G be a graph embedded on a surface ofgenus g with b boundary cycles. If G is undirected,then we give algorithms to compute a shortest non-separating, non-contractible, or non-null-homologous cyclein 2O(g)n log log n time. In the more interesting case whereG is directed, we give an algorithm to compute a shortestnon-contractible cycle in O(g3n log n) time or a shortestnon-null-homologous cycle in O(g2n log n) time.


CP6

A Simple Algorithm for the Graph Minor Decom-position - Logic Meets Structural Graph Theory -

A key result of Robertson and Seymour’s graph minor the-ory is a structure theorem stating that all graphs excludingsome fixed graph as a minor have a tree decomposition into

DA13 Abstracts 35

pieces that are almost embeddable in a fixed surface. Mostalgorithmic applications of graph minor theory rely on analgorithmic version of this result. However, the known al-gorithms for computing such graph minor decompositionsheavily rely on the very long and complicated proofs of theexistence of such decompositions, essentially they retracethese proofs and show that all steps are algorithmic. Inthis paper, we give a simple quadratic time algorithm forcomputing graph minor decompositions. The best previ-ously known algorithm due to Kawarabayashi and Wollanruns in cubic time and is far more complicated. Our algo-rithm combines techniques from logic and structural graphtheory, or more precisely, a variant of Courcelle’s Theoremstating that monadic second-order logic formulas can beevaluated in linear time on graphs of bounded tree widthand Robertson and Seymour’s so called Weak StructureTheorem.

Martin GroheHumboldt-Universitat zu [email protected]

Ken-ichi KawarabayashiNational Institute of Informatics, Japank [email protected]

Bruce ReedMcGill [email protected]

CP6

Packing Directed Cycles Through a Specified Ver-tex Set

We show that the so-called Erdos-Posa property does holdfor half-integral packings of directed cycles each containinga vertex from a given set S, i.e. where every vertex of thegraph is contained in at most 2 cycles. On the other hand,an example shows that the Erdos-Posa property does nothold without this relaxation.


Daniel Kral, Marek KrcalCharles [email protected], [email protected]

Stephan KreutzerTechnical University [email protected]

CP6

4-Connected Projective-Planar Graphs AreHamiltonian-Connected

We prove that every 4-connected projective planar graphis hamiltonian-connected. Here, hamiltonian-connectedmeans that for any two vertices, there is a hamiltonianpath between them (and hence this generalizes “hamilto-nian’). This proves a conjecture of Dean in 1990. Our proofis constructive in the sense that there is a polynomial time(in fact, O(n2) time) algorithm to find, given two verticesin a 4-connected projective planar graph, a hamiltonianpath between these two vertices.

Kenta Ozeki, Ken-ichi KawarabayashiNational Institute of Informatics, Japan

[email protected], k [email protected]

CP6

Jungles, Bundles, and Fixed-Parameter Tractabil-ity

We give a fixed-parameter tractable (FPT) approximationalgorithm computing the pathwidth of a tournament, andmore generally, of a semi-complete digraph. Based on thisresult, we prove that the Topological Containment andRooted Immersion problems are FPT on semi-completedigraphs. Moreover, our results also imply that computingvertex deletion distance to every immersion-closed class ofsemi-complete digraphs is fixed-parameter tractable.

Fedor FominDep. of InformaticsUniversity of [email protected]

Michal PilipczukUniversity of [email protected]

CP7

Fast Algorithms for Interactive Coding

Consider two parties who wish to communicate in order toexecute some interactive protocol π. However, the com-munication channel between them is noisy: An adversarysees everything that is transmitted over the channel andcan change a constant fraction of the bits as he pleases,thus interrupting the execution of π (which was designedfor an errorless channel). If π only contained one message,then a good error correcting code would have overcamethe noise with only a constant overhead in communication,but this solution is not applicable to interactive protocolswith many short messages. Schulman (FOCS 92, STOC93) presented the notion of interactive coding : A simula-tor that, given any protocol π, is able to simulate it (i.e.produce its intended transcript) even with constant rateadversarial channel errors, and with only constant (mul-tiplicative) communication overhead. Until recently, how-ever, the running time of all known simulators was expo-nential (or sub-exponential) in the communication com-plexity of π (denoted N in this work). Brakerski and Kalai(FOCS 12) recently presented a simulator that runs in timepoly(N). Their simulator is randomized (each party flipsprivate coins) and has failure probability roughly 2−N . Inthis work, we improve the computational complexity of in-teractive coding. While at least N computational steps arerequired (even just to output the transcript of π), the BK

simulator runs in time Ω(N2). We present two efficientalgorithms for interactive coding: The first with computa-tional complexity O(N logN) and exponentially small fail-ure probability; and the second with computational com-plexity O(N), but failure probability 1/poly(N). (Compu-tational complexity is measured in the RAM model.)

Zvika BrakerskiStanford [email protected]

Moni NaorWeizmann Institute of [email protected]

36 DA13 Abstracts

CP7

Restricted Isometry of Fourier Matrices and ListDecodability of Random Linear Codes

We prove that a random linear code over GF(q), withprobability arbitrarily close to 1, is list decodable at ra-dius 1 − 1/q − ε with list size L = O(1/ε2) and rate R =Ωq(ε2/(log3(1/ε))). Up to the polylogarithmic factor in1/ε and constant factors depending on q, this matches thelower bound L = Ωq(1/ε2) for the list size and upper boundR = Oq(ε2) for the rate. Previously only existence (andnot abundance) of such codes was known for the specialcase q = 2 (Guruswami, Hastad, Sudan and Zuckerman,2002). In order to obtain our result, we employ a relaxedversion of the well known Johnson bound on list decod-ing that translates the average Hamming distance betweencodewords to list decoding guarantees. We furthermoreprove that the desired average-distance guarantees holdfor a code provided that a natural complex matrix encod-ing the codewords satisfies the Restricted Isometry Prop-erty with respect to the Euclidean norm (RIP-2). For thecase of random binary linear codes, this matrix coincideswith a random submatrix of the Hadamard-Walsh trans-form matrix that is well studied in the compressed sensingliterature. Finally, we improve the analysis of Rudelsonand Vershynin (2008) on the number of random frequencysamples required for exact reconstruction of k-sparse sig-nals of length N . Specifically, we improve the numberof samples from O(k log(N) log2(k)(log k + log logN)) toO(k log(N) log3(k)). The proof involves bounding the ex-pected supremum of a related Gaussian process by usingan improved analysis of the metric defined by the process.This improvement is crucial for our application in list de-coding.

Mahdi CheraghchiCarnegie Mellon UniversityComputer Science [email protected]

Venkatesan GuruswamiCarnegie Mellon [email protected]

Ameya VelingkerCarnegie Mellon UniversityComputer Science [email protected]

CP7

Shift Finding in Sub-Linear Time

We study the following basic pattern matching problem.Consider a “code’ sequence c consisting of n bits chosenuniformly at random, and a “signal’ sequence x obtained byshifting c (modulo n) and adding noise. The goal is to effi-ciently recover the shift with high probability. The problemmodels tasks of interest in several applications, includingGPS synchronization and motion estimation. We presentan algorithm that solves the problem in time O(n(f/(1+f)),

where O(Nf ) is the running time of the best algorithmfor finding the closest pair among N “random’ sequencesof length O(logN). A trivial bound of f = 2 leads to

a simple algorithm with a running time of O(n2/3). Theasymptotic running time can be further improved by plug-ging in recent more efficient algorithms for the closest pairproblem. Our results also yield a sub-linear time algorithmfor approximate pattern matching algorithm for a randomsignal (text), even for the case when the error between the

signal and the code (pattern) is asymptotically as large asthe code size. This is the first sublinear time algorithm forsuch error rates.

Haitham [email protected]

Alexandr AndoniMicrosoft [email protected]

Piotr IndykMassachusetts Institute of [email protected]

Dina [email protected]

CP7

The Fast Cauchy Transform and Faster Robust Lin-ear Regression

We provide fast algorithms for overconstrained Lp regres-sion and related problems that improve upon the best pre-vious algorithms. We also provide a suite of results forfinding well-conditioned bases via ellipsoidal rounding, il-lustrating various tradeoffs. To complement this theory, weprovide an empirical evaluation. Among other things, thisshow the theory is a good guide to practical performance.Our algorithms use a fast subspace embedding of indepen-dent interest that we call the Fast Cauchy Transform.

Michael MahoneyStanford UniversityApplied and Computational [email protected]

Kenneth [email protected]

Petros DrineasComputer Science DepartmentRennselaer Polytechnic [email protected]

Malik Magdon-IsmailRensselaer Polytechnic [email protected]

Xiangrui [email protected]

David WoodruffIBM Almaden Research Center, [email protected]

CP7

Generalized Perron-Frobenius Theorem for Multi-ple Choice Matrices, and Applications

The celebrated Perron–Frobenius (PF) theorem is statedfor irreducible nonnegative square matrices, and providesa simple characterization of their eigenvectors and eigen-values. The importance of this theorem stems from the

DA13 Abstracts 37

fact that eigenvalue problems on such matrices arise inmany fields of science and engineering, including dynami-cal systems theory, economics, statistics and optimization.However, many real-life scenarios give rise to nonsquarematrices. Despite the extensive development of spectraltheories for nonnegative matrices, the applicability of suchtheories to non-convex optimization problems is not clear.In particular, a natural question is whether the PF The-orem (along with its applications) can be generalized toa nonsquare setting. Our paper provides a generalizationof the PF Theorem to nonsquare multiple choice matrices.The extension can be interpreted as representing systemswith additional degrees of freedom, where each client en-tity may choose between multiple servers that can cooper-ate in serving it (while potentially interfering with otherclients). This formulation is motivated by applications topower control in wireless networks, economics and others,all of which extend known examples for the use of the orig-inal PF Theorem. We show that the option of cooperationdoes not improve the situation, in the sense that in theoptimum solution, no cooperation is needed, and only oneserver per client entity needs to work. Hence, the addi-tional power of having several potential servers per clienttranslates into choosing the “best” single server and notinto sharing the load between the servers in some way, asone might have expected. The two main contributions ofthe paper are (i) a generalized PF Theorem that character-izes the optimal solution for a non-convex problem, and (ii)an algorithm for finding the optimal solution in polynomialtime.

Chen AvinBen Gurion University, Beer-Sheva, [email protected]

Michael BorokhovichBen Gurion University, Beer-Sheva, [email protected]

Yoram HaddadJerusalem College of Technology,Jerusalem, [email protected]

Erez KantorThe Technion, Haifa, [email protected]

Zvi LotkerBen Gurion University, Beer-Sheva, [email protected]

Merav ParterThe Weizmann Institute of Science, Rehovot, [email protected]

David PelegWeizmann Institute of [email protected]

CP8

Distance Oracles for Stretch Less Than 2

We present distance oracles for weighted undirected graphsthat return distances of stretch less than 2. For the realisticcase of sparse graphs, our distance oracles exhibit a smooththree-way trade-off between space, query time and stretch— a phenomenon that does not occur in dense graphs. Inparticular, for any positive integer t and for any 1 ≤ α ≤ n,our distance oracle is of size O(m + n2/α) and returns

distances of stretch at most (1 + 2t+1

) in time O((αμ)t),

where μ = 2m/n is the average degree of the graph. Thequery time can be further reduced to O((α + μ)t) at theexpense of a small additive stretch.

Rachit Agarwal, Philip GodfreyUniversity of Illinois at [email protected], [email protected]

CP8

New Additive Spanners

This paper considers additive and purely additive spanners.We present a new purely additive spanner of size O(n7/5)with additive stretch 4. This construction fills in the gapbetween the two existing constructions for purely addi-tive spanners, one for 2-additive spanner of size O(n3/2)

and the other for 6-additive spanner of size O(n4/3), andthus answers a main open question in this area. In addi-tion, we present a construction for additive spanners withO(n1+δ) edges and additive stretch of O(n1/2−3δ/2) forany 3/17 ≤ δ < 1/3, improving the stretch of the exist-

ing constructions from O(n1−3δ) to O(√n1−3δ). Finally,

we show that our (1, n1/2−3δ/2)-spanner construction canbe tweaked to give a sublinear additive spanner of sizeO(n1+3/17) with additive stretch O(

√distance).

Shiri ChechikMicrosoft Research [email protected]

CP8

Fast Constructions of Light-Weight Spanners forGeneral Graphs

Since the pioneering works of Peleg and Schaffer [PS89],Althofer et al. [ADDJS93], and Chandra et al. [CDNS92],it is known that for every weighted undirected n-vertexm-edge graph G = (V,E), and every integer k ≥ 1,

there exists a ((2k − 1) · (1 + ε))-spanner with O(n1+1/k)

edges and weight O(k · n1/k) · ω(MST (G)). Nearly lineartime algorithms for constructing (2k − 1)-spanners with

nearly O(n1+1/k) edges were devised in [Baswana-Sen 2003,Roditty-Zwick 2004, Roditty-Thorup-Zwick 2005]. How-ever, these algorithms fail to guarantee any meaningful up-per bound on the weight of the constructed spanners. Toour knowledge, there are only two known algorithms forconstructing sparse and light spanners for general graphs.One of them is the greedy algorithm of [ADDJS93], ana-lyzed in [CDNS92]. The drawback of the greedy algorithm

is that it requires O(m · (n1+1/k + n · log n)) time. Theother algorithm is due to Awerbuch et al. [ABP91], from

1991. It constructs O(k)-spanners with O(k · n1+1/k · Λ)

edges, weight O(k2 · n1/k · Λ) · ω(MST (G)), within time

O(m·k·n1/k ·Λ), where Λ is the logarithm of the aspect ratioof the graph. In this paper we devise an efficient algorithmfor constructing sparse and light spanners. Specifically,our algorithm constructs ((2k − 1) · (1 + ε))-spanners with

O(k · n1+1/k) edges and weight O(k · n1/k) · ω(MST (G)),where ε > 0 is an arbitrarily small constant. The runningtime of our algorithm is O(k ·m+ min{n · log n,m ·α(n)}).

Michael ElkinBen-Gurion [email protected]

38 DA13 Abstracts

Shay SolomonWeizmann Institute of [email protected]

CP8

More Compact Oracles for Approximate Distancesin Undirected Planar Graphs

Distance oracles are data structures that provide fast (pos-sibly approximate) answers to shortest-path and distancequeries in graphs. The tradeoff between the space require-ments and the query time of distance oracles is of particularinterest and the main focus of this paper. Unless statedotherwise, we assume all graphs to be planar and undi-rected. In FOCS 2001 (J. ACM 2004), Thorup introducedapproximate distance oracles for planar graphs (concur-rent with Klein, SODA 2002). Thorup proved that, for anyε > 0 and for any undirected planar graph on n nodes, thereexists a (1 + ε)-approximate distance oracle using spaceO((n log n)/ε) such that approximate distance queries canbe answered in time O(1/ε). In this paper, we aim at reduc-ing the polynomial dependency on 1/ε and log n, gettingthe first improvement in the query time-space tradeoff. Tosimplify the statement of our bounds, we define Ô() tohide log log n and log(1/ε) factors. We provide the firstoracle with a time-space product that is subquadratic in1/ε. We obtain an oracle with space Ô(n log n) and querytime Ô(1/ε). For unweighted graphs we show how thelogarithmic dependency on n can be removed. We obtainan oracle with space Ô(n) and query time Ô(1/ε). Thisbound also holds for graphs with polylogarithmic averageedge length, which may be a quite reasonable assumption,e.g., for road networks.

Christian [email protected]


Mikkel ThorupAT&T Labs - Research and University of [email protected]

CP8

Approximate Distance Oracles with ImprovedQuery Time

For a graph with m edges and n vertices and for an integerk ≥ 2, we give a (2k − 1)-approximate distance oracle of

size O(kn1+1/k) and with O(log k) query time, improvingthe O(k) query time of Thorup and Zwick. Furthermore,

for any 0 < ε ≤ 1, we give an oracle of size O(kn1+1/k) thatanswers ((2 + ε)k)-approximate distance queries in O(1/ε)time. This improves a 128k approximation of Mendel andNaor and approaches the best possible tradeoff betweensize and stretch.

Christian Wulff-NilsenDepartment of Mathematics and Computer ScienceUniversity of Southern [email protected]

Christian Wulff-NilsenDepartment of Computer ScienceUniversity of Copenhagen

[email protected]

CP9

Optimal and Efficient Parametric Auctions

Consider a seller who seeks to provide service to a collec-tion of interested parties, subject to feasibility constraintson which parties may be simultaneously served. Assumingthat a distribution is known on the value of each party forservice—arguably a strong assumption— Myerson’s semi-nal work provides revenue optimizing auctions. We showinstead that, for very general feasibility constraints, onlyknowledge of the median of each party’s value distribution,or any other quantile of these distributions, or approxima-tions thereof, suffice for designing simple auctions that si-multaneously approximate both the optimal revenue andthe optimal welfare. Our results apply to all downward-closed feasibility constraints under the assumption thatthe underlying, unknown value distributions are monotonehazard rate, and to all matroid feasibility constraints un-der the weaker assumption of regularity of the underlyingdistributions. Our results jointly generalize the single-itemresults obtained by Azar and Micali on parametric auc-tions, and Daskalakis and Pierrakos for simultaneously ap-proximately optimal and efficient auctions.

Pablo Azar, Constantinos [email protected], [email protected]

Silvio MicaliMassachusetts Institute of [email protected]

Matt [email protected]

CP9

Simple and Nearly Optimal Multi-Item Auctions

We provide a Polynomial Time Approximation Scheme(PTAS) for the Bayesian optimal multi-item multi-bidderauction problem under two conditions. First, bidders areindependent, have additive valuations and are from thesame population. Second, every bidder’s value distribu-tions of items are independent but not necessarily identi-cal monotone hazard rate (MHR) distributions. For non-i.i.d. bidders, we also provide a PTAS when the number ofbidders is small. Prior to our work, even for a single bidder,only constant factor approximations are known. Anotherappealing feature of our mechanism is the simple allocationrule. Indeed, the mechanism we use is either the second-price auction with reserve price on every item individually,or VCG allocation with a few outlying items that requiresadditional treatments. It is surprising that such simpleallocation rules suffice to obtain nearly optimal revenue.

Yang [email protected]

Zhiyi HuangUniversity of [email protected]

DA13 Abstracts 39

CP9

Clinching Auctions with Online Supply

Auctions for perishable goods such as internet ad inven-tory need to make real-time allocation and pricing deci-sions as the supply of the good arrives in an online manner,without knowing the entire supply in advance. These allo-cation and pricing decisions get complicated when buyershave some global constraints. In this work, we considera multi-unit model where buyers have global budget con-straints, and the supply arrives in an online manner. Ourmain contribution is to show that for this setting there isan individually-rational, incentive-compatible and Pareto-optimal auction that allocates these units and calculatesprices on the fly, without knowledge of the total supply.We do so by showing that the Adaptive Clinching Auctionsatisfies a supply-monotonicity property. We also analyzeand discuss, using examples, how the insights gained by theallocation and payment rule can be applied to design betterad allocation heuristics in practice. Finally, while our maintechnical result concerns multi-unit supply, we propose aformal model of online supply that captures scenarios be-yond multi-unit supply and has applications to sponsoredsearch. We conjecture that our results for multi-unit auc-tions can be extended to these more general models.

Renato Paes LemeCornell [email protected]

Gagan GoelGoogle Research, New [email protected]

Vahab MirrokniGoogle [email protected]

CP9

Ironing in Dynamic Revenue Management: PostedPrices & Biased Auctions

We consider the design of the revenue maximizing mecha-nism for a seller with a fixed capacity selling to buyers whoarrive over time. The buyers have single unit demand andboth their value for the object and the deadline by whichthey must make a purchase are unknown to the seller.In previous work where buyers’ deadlines are publicly ob-served and only values are private, the optimal mechanismcan be computed by running a dynamic stochastic knap-sack algorithm. However, these mechanisms are only fea-sible with private deadlines when the calculated allocationrule is appropriately monotone. In the classic static envi-ronment of Myerson (1981) monotonicity is only violatedfor ‘irregular’ value distributions. Here, he characterizesthe optimal mechanism by a procedure he calls ‘ironing.’We characterize the optimal mechanism in our general dy-namic environment by providing the dynamic counterpartof ironing. We show that only a subset of the monotonicityconstraints can bind in a solution of the seller’s dynamicprogramming problem. The optimal mechanism can becharacterized by relaxing these constraints with their ap-propriate dual multiplier. Further, the optimal mechanismcan be implemented by a series of posted prices followedby a ‘biased’ auction in the final period where buyers havethe auction biased in their favor depending on their ar-rival time. Our theoretical characterization complementsthe existing computational approaches for ironing in these

settings (e.g. Parkes et al. (2007)).

Mallesh M. PaiDepartment of EconomicsUniversity of [email protected]

Rahul DebDepartment of EconomicsUniversity of [email protected]

CP9

Reducing Revenue to Welfare Maximization: Ap-proximation Algorithms and Other Generalizations

It was recently shown in [CDW12b] that revenue opti-mization can be computationally efficiently reduced to wel-fare optimization in all multi-dimensional Bayesian auctionproblems with arbitrary (possibly combinatorial) feasibil-ity constraints and independent additive bidders with ar-bitrary (possibly combinatorial) demand constraints. Thisreduction provides a poly-time solution to the optimalmechanism design problem in all auction settings wherewelfare optimization can be solved efficiently, but it is frag-ile to approximation and cannot provide solutions to set-tings where welfare maximization can only be tractablyapproximated. In this paper, we extend the reductionto accommodate approximation algorithms, providing anapproximation preserving reduction from (truthful) rev-enue maximization to (not necessarily truthful) welfaremaximization. The mechanisms output by our reductionchoose allocations via black-box calls to welfare approxima-tion on randomly selected inputs, thereby generalizing alsoour earlier structural results on optimal multi-dimensionalmechanisms to approximately optimal mechanisms. Un-like [CDW12b], our results here are obtained through noveluses of the Ellipsoid algorithm and other optimization tech-niques over non-convex regions.

Yang Cai, Constantinos Daskalakis, Matt [email protected], [email protected],[email protected]

CP10

On the Number of Matroids

We consider the problem of determining mn, the numberof matroids on n elements. Knuth (1974) showed thatlog logmn ≥ n − 3

2log n − O(1), while Piff (1973) showed

that log logmn ≤ n − log n + log log n + O(1), and it hasbeen conjectured that the right answer is perhaps closerto Knuth’s bound. We show that this is indeed the case,and prove an upper bound on log logmn that is within anadditive 1 + o(1) term of Knuth’s lower bound.

Nikhil BansalEindhoven University of [email protected]

Rudi Pendavingh, Jorn van Der PolTU [email protected], [email protected]

40 DA13 Abstracts

CP10

Playing Mastermind with Many Colors

We analyze the general version of the classic guessing gameMastermind with n positions and k colors. Since the casek ≤ n1−ε, ε > 0 constant, is well understood, we concen-trate on larger numbers of colors. For the most prominentcase k = n, our results imply that Codebreaker can find thesecret code with O(n log log n) guesses. This bound is validalso when only black answer-pegs are used. It improves theO(n log n) bound first proven by Chvatal (Combinatorica3 (1983), 325–329). We also show that if both black andwhite answer-pegs are used, then the O(n log log n) boundholds for up to n2 log log n colors. These bounds are almosttight as the known lower bound of Ω(n) shows. Unlikefor k ≤ n1−ε, simply guessing at random until the secretcode is determined is not sufficient. In fact, we show thatany non-adaptive strategy needs an expected number ofΩ(n log n) guesses.

Benjamin Doerr, Reto SpohelMax Planck Institute for [email protected], [email protected]

Henning ThomasInstitute of Theoretical Computer Science ETH Zurich8092 Zurich, [email protected]

Carola WinzenMax-Planck-Institut fur [email protected]

CP10

Tight Cell-Probe Bounds for Online Hamming Dis-tance Computation

We show tight bounds for online Hamming distance com-putation in the cell-probe model with word size w. Thetask is to output the Hamming distance between a fixedstring of length n and the last n symbols of a stream. Wegive a lower bound of Ω

(δw

log n)

time on average per out-put, where δ is the number of bits needed to represent aninput symbol. We argue that this bound is tight within themodel. The lower bound holds under randomisation andamortisation.

Markus Jalsenius, Raphael CliffordUniversity of [email protected], [email protected]

Benjamin SachUniversity of [email protected]

CP10

Lower Bounds for Adaptive Sparse Recovery

We give lower bounds for the problem of stable sparse re-covery from adaptive linear measurements. In this prob-lem, one would like to estimate a vector x ∈ Rn from mlinear measurements A1x, . . . , Amx. One may choose eachvector Ai based on A1x, . . . , Ai−1x, and must output xsatisfying

‖x− x‖p ≤ (1 + ε) mink-sparse x′

‖x− x′‖p

with probability at least 1 − δ > 2/3, for some p ∈ {1, 2}.For p = 2, it was recently shown that this is possible with

m = O( 1εk log log(n/k)), while nonadaptively it requires

Θ( 1εk log(n/k)). It is also known that even adaptively, it

takes m = Ω(k/ε) for p = 2. For p = 1, there is a non-

adaptive upper bound of O( 1√εk log n). We show:

• For p = 2, m = Ω(log log n). This is tight for k = O(1)and constant ε, and shows that the log log n depen-dence is correct.

• If the measurement vectors are chosen in R “rounds”,then m = Ω(R log1/R n). For constant ε, this matchesthe previously known upper bound up to an O(1) fac-tor in R.

• For p = 1, m = Ω(k/(√ε · log k/ε)). This shows that

adaptivity cannot improve more than logarithmic fac-tors, providing the analogue of the m = Ω(k/ε) boundfor p = 2.

Eric [email protected]


CP10

The Communication Complexity of Addition


Emanuele ViolaNortheastern [email protected]

CP11

Finding Endogenously Formed Communities

We define a natural notion of (potentially overlapping)communities, in which the members of a given communitycollectively prefer each other to anyone else outside thecommunity. We provide a tight polynomial bound on thenumber of such communities as a function of the robustnessof the community. Moreover, we present polynomial-timealgorithms for finding these communities and a local algo-rithm that can find a community in time nearly linear inits size.

Maria-Florina BalcanGeorgia Institute of [email protected]

Christian BorgsMicrosoft [email protected]

Mark BravermanPrinceton [email protected]

Jennifer ChayesMicrosoft [email protected]

Shang-Hua TengUniversity of Southern [email protected]

DA13 Abstracts 41

CP11

Anonymous Meeting in Networks

A team of an unknown number of mobile agents, in an un-known network, have to meet at the same node. Agentsare anonymous, execute the same deterministic algorithmand move synchronously. An initial configuration is gather-able if there exists a deterministic algorithm that achievesmeeting of all agents in one node. Which configurationsare gatherable and how to gather the agents deterministi-cally by the same algorithm? We give a complete solutionof this gathering problem in arbitrary networks.

Yoann DieudonneDepartement InformatiqueUniversite du Quebec en [email protected]

Andrzej PelcDepartement dinformatiqueUniversite du Quebec en [email protected]

CP11

On the Complexity of Information Spreading inDynamic Networks

We study how to spread k tokens of information to everynode on an n-node dynamic network, the edges of whichchange at each round. This basic gossip problem can becompleted in O(n + k) rounds in any static network, anddetermining its complexity in dynamic networks is centralto understanding the algorithmic limits and capabilities ofvarious dynamic network models. We provide several al-most tight upper and lower bounds in some of these models.

Chinmoy DuttaNortheastern [email protected]

Gopal PanduranganDivision of Mathematical SciencesNanyang Technological [email protected]

Rajmohan Rajaraman, Zhifeng SunNortheastern [email protected], [email protected]

Emanuele ViolaNortheastern [email protected]

CP11

Near Optimal Leader Election in Multi-Hop RadioNetworks

We design leader election protocols for multi-hop radionetworks that elect a leader in almost the same timeTBC that it takes for broadcasting one message (oneID). For the setting without collision detection our algo-rithm runs with high probability in O(D log n

D+ log3 n) ·

min{log log n, log nD} rounds on any n-node network with

diameter D. Since TBC = Θ(D log nD

+ log2 n) is a lowerbound, our upper bound is optimal up to a factor of atmost log log n and the extra log n factor on the additive

term. Our algorithm is furthermore the first O(n) timealgorithm for this setting. This algorithm improves overa 23 year old simulation approach of Bar-Yehuda, Gol-dreich and Itai with a O(TBC log n) running time. Wealso give an O(D+ log n log log n) · min{log log n, log n

D} =

O(D + log n) · O(log log n)2 leader election algorithm forthe setting with collision detection (even with single-bitmessages). This time bound — which is also in O(n) —is optimal up to log log n factors and improves over a de-terministic algorithm that requires Θ(n) rounds indepen-dently of D.

Mohsen Ghaffari, Bernhard [email protected], [email protected]

CP11

Simple, Fast and Deterministic Gossip and RumorSpreading


Bernhard [email protected]

CP12

Reporting Neighbors in High-Dimensional Eu-clidean Space

We consider the following problem, which arises in manydatabase and web-based applications: Given a set P of npoints in a high-dimensional space Rd and a distance r, wewant to report all pairs of points of P at Euclidean dis-tance at most r. We present two randomized algorithms,one based on randomly shifted grids, and the other on ran-domly shifted and rotated grids. The running time of bothalgorithms is of the form C(d)(n+ k) log n, where k is theoutput size and C(d) is a constant that depends on thedimension d. The log n factor is needed to guarantee, withhigh probability, that all neighbor pairs are reported, andcan be dropped if it suffices to report, in expectation, anarbitrarily large fraction of the pairs. When only transla-tions are used, C(d) is of the form (a

√d)d, for some (small)

absolute constant a ≈ 0.484; this bound is worst-case tight,up to an exponential factor of about 2d. When both ro-tations and translations are used, C(d) can be improved

to roughly 6.74d, getting rid of the super-exponential fac-

tor√dd. When the input set (lies in a subset of d-space

that) has low doubling dimension δ, the performance ofthe first algorithm improves to C(d, δ)(n + k) log n (or to

C(d, δ)(n + k)), where C(d, δ) = O((ed/δ)δ), for δ ≤√d.

Otherwise, C(d, δ) = O(e√d√dδ)

. We also present ex-

perimental results on several large datasets, demonstrat-ing that our algorithms run significantly faster than all theleading existing algorithms for reporting neighbors.

Dror AigerGoogle [email protected]

Haim Kaplan, Micha SharirTel Aviv [email protected], [email protected]

42 DA13 Abstracts

CP12

The Traveling Salesman Problem for Lines, Ballsand Planes

We revisit the traveling salesman problem with neigh-borhoods (TSPN) and obtain several approximation algo-rithms. These constitute either improvements over previ-ously best approximations achievable in comparable times(for unit disks in the plane), or first approximations ever(for planes, lines and unit balls in 3-space). (I) Given a

set of n planes in 3-space, a TSP tour that is at most 2.31times longer than the optimal can be computed in O(n)time. (II) Given a set of n lines in 3-space, a TSP tour that

is at most O(log3 n) times longer than the optimal can becomputed in polynomial time. (III) Given a set of n unit

disks in the plane (resp., unit balls in 3-space), we improvethe approximation ratio using a black box that computesan approximate tour for a set of points (the centers of asubset of the disks or the balls).

Adrian DumitrescuDepartment of Computer ScienceUniversity of [email protected]

Csaba D. TothUniversity of [email protected]

CP12

Clustering Affine Subspaces: Hardness and Algo-rithms

We study a generalization of the famous k-center prob-lem where each object is an affine subspace of dimensionΔ, and give either the first or significantly improved al-gorithms and hardness results for many combinations ofparameters. This generalization from points (Δ = 0) ismotivated by the analysis of incomplete data, a pervasivechallenge in statistics: incomplete data objects in Rd canbe modeled as affine subspaces. We give three algorithmicresults for different values of k, under the assumption thatall subspaces are axis-parallel, the main case of interestbecause of the correspondence to missing entries in datatables. 1) k = 1: Two polynomial time approximation

schemes which runs in poly(Δ, 1/ε)nd. 2) k = 2: O(Δ1/4)-approximation algorithm which runs in poly(n, d,Δ) 3)General k: Polynomial time approximation scheme which

runs in 2O(Δk log k(1+1/ε2))nd We also prove nearly matchinghardness results; in both the general (not necessarily axis-parallel) case (for k ≥ 2) and in the axis-parallel case (fork ≥ 3), the running time of an approximation algorithmwith any approximation ratio cannot be polynomial in evenone of k and Δ, unless P = NP. Furthermore, assuming thatthe 3-SAT problem cannot be solved subexponentially, thedependence on both k and Δ must be exponential in thegeneral case (in the axis-parallel case, only the dependence

on k drops to 2Ω(√

k)). The simplicity of the first and thethird algorithm suggests that they might be actually usedin statistical applications. The second algorithm, whichdemonstrates a theoretical gap between the axis-paralleland general case for k = 2, displays a strong connection be-tween geometric clustering and classical coloring problemson graphs and hypergraphs, via a new Helly-type theorem.

Euiwoong LeeCarnegie Mellon University

[email protected]

Leonard [email protected]

CP12

Approximating Watchman Routes

Given a connected polygonal domain P , the watchmanroute problem is to compute a shortest path or tour fora mobile guard (the “watchman’) that is required to seeevery point of P . While the watchman route problem ispolynomially solvable in simple polygons, it is known tobe NP-hard in polygons with holes. We present the firstpolynomial-time approximation algorithm for the watch-man route problem in polygonal domains. Our algorithmhas an approximation factor O(log2 n). Further, we provethat the problem cannot be approximated in polynomialtime to within a factor of c log n, for a constant c > 0,assuming that P�=NP.

Joseph MitchellStony Brook [email protected]

CP12

Euclidean Spanners in High Dimensions

A classical result in metric geometry asserts that anyn-point metric admits a linear-size spanner of dilationO(log n) [Peleg, and Schaffer, 1989]. More generally, forany c > 1, any metric space admits a spanner of sizeO(n1+1/c), and dilation at most c. This bound is tightassuming the well-known girth conjecture of Erdos [Erdos,1963]. We show that for a metric induced by a set of npoints in high-dimensional Euclidean space, it is possible toobtain improved dilation/size trade-offs. More specifically,we show that any n-point Euclidean metric admits a near-linear size spanner of dilation O(

√log n). Using the LSH

scheme of Andoni and Indyk [Andoni, and Indyk, 2006] wefurther show that for any c > 1, there exist spanners of

size roughly O(n1+1/c2 ) and dilation O(c). Finally, we alsoexhibit super-linear lower bounds on the size of spannerswith constant dilation.

Sariel Har-PeledDepartment of Computer ScienceUniversity of Illinios, [email protected]

Piotr [email protected]

Anastasios SidiropoulosToyota Technological Institute at Chicago, [email protected]

CP13

Windows into Relational Events: Data Structuresfor Contiguous Subsequences of Edges

We consider the problem of analyzing social network datasets in which the edges of the network have timestamps,and we wish to analyze the subgraphs formed from edgesin contiguous subintervals of these timestamps. We pro-

DA13 Abstracts 43

vide data structures for these problems that use near-linearpreprocessing time, linear space, and sublogarithmic querytime to handle queries that ask for the number of con-nected components, number of components that containcycles, number of vertices whose degree equals or is at mostsome predetermined value, number of vertices that can bereached from a starting set of vertices by time-increasingpaths, and related queries.

Michael J. BannisterUniversity of California, IrvineDepartment of Computer [email protected]

Christopher DuBoisUniversity of California, IrvineDepartment of Computer [email protected]

David EppsteinDepartment of Information and Computer ScienceUniversity of California, [email protected]

Padhraic SmythUC [email protected]

CP13

Towards More Realistic Probabilistic Models forData Structures: The External Path Length inTries under the Markov Model

Tries are among the most versatile and widely used datastructures on words. Its performance analysis, even forsimplest memoryless sources has proved difficult and rig-orous findings about inherently complex parameters undermore realistic models of string generations were rarely ana-lyzed. Here, we consider general Markov sources and provea central limit theorem for the external path length of atrie.

Ralph NeiningerJ.W. Goethe University, [email protected]

Kevin LeckeyGoethe University Frankfurt am Main, [email protected]

Wojciech SzpankowskiPurdue UniversityDepratment of Computer [email protected]

CP13

Optimal Dynamic Sequence Representations

We describe a data structure that supports access, rankand select queries, as well as symbol insertions and dele-tions, on a string S[1, n] over alphabet [1..σ] in timeO(lg n/ lg lg n), which is optimal. The time is worst-case for the queries and amortized for the updates. Thiscomplexity is better than the best previous ones by aΘ(1 + lg σ/ lg lg n) factor. Our structure uses nH0(S) +O(n + σ(lg σ + lg1+ε n)) bits, where H0(S) is the zero-order entropy of S and 0 < ε < 1 is any constant. Thisspace redundancy over nH0(S) is also better, almost al-

ways, than that of the best previous dynamic structures,o(n lg σ)+O(σ(lg σ+lgn)). We can also handle general al-phabets in optimal time, which has been an open problemin dynamic sequence representations.

Gonzalo Navarro, Yakov NekrichUniversity of [email protected], [email protected]

CP13

Output-Sensitive Skyline Algorithms in ExternalMemory

This paper presents new results in external memoryfor finding the skyline (a.k.a. maxima) of N pointsin d-dimensional space. The state of the art usesO((N/B) logd−2

M/B(N/B)) I/Os for fixed d ≥ 3, and

O((N/B) logM/B(N/B)) I/Os for d = 2, where M and

B are the sizes (in words) of memory and a disk block,respectively. We give algorithms whose running time de-pends on the number K of points in the skyline. Specif-ically, we achieve O((N/B) logd−2

M/B(K/B)) expected cost

for fixed d ≥ 3, and O((N/B) logM/B(K/B)) worst-casecost for d = 2. As a side product, we solve two problemsboth of independent interest. The first one, the M-skylineproblem, aims at reporting M arbitrary skyline points, orthe entire skyline if its size is at most M . We settle thisproblem in O(N/B) expected time in any fixed dimension-ality d. The second one, the M-pivot problem, is morefundamental: given a set S of N elements drawn from anordered domain, it outputs M evenly scattered elements(called pivots) from S, namely, S has asymptotically thesame number of elements between each pair of consecutivepivots. We give a deterministic algorithm for solving theproblem in O(N/B) I/Os.

Yufei Tao, Xiaocheng Hu, Cheng ShengChinese University of Hong [email protected], [email protected],[email protected]

Yi Yang, Shuigeng ZhouFudan [email protected], [email protected]

CP13

Multiway Simple Cycle Separators and I/O-Efficient Algorithms for Planar Graphs

We revisit I/O-efficient solutions to a number of funda-mental problems on planar graphs: single-source short-est paths, topological sorting, and computing stronglyconnected components. Existing I/O-efficient solutionsto these problems pay for I/O efficiency using excessivecomputation time in internal memory, thereby completelynegating the performance gain achieved by minimizing thenumber of disk accesses. In this paper, we show how tomake these algorithms simultaneously efficient in inter-nal and external memory so they achieve I/O complexityO(sort(N)) and take O(N logN) time in internal memory,where sort(N) is the number of I/Os needed to sort N itemsin external memory. The key, and the main technical con-tribution of this paper, is a multiway version of Miller’ssimple cycle separator theorem. We show how to computethese separators in linear time in internal memory, andusing O(sort(N)) I/Os and O(N logN) (internal-memorycomputation) time in external memory.

Lars Arge, Freek Van Walderveen

44 DA13 Abstracts

MADALGO, Department of Computer ScienceAarhus [email protected], freek [email protected]

Norbert ZehDalhousie [email protected]

CP14

A Constant Factor Approximation Algorithm forReordering Buffer Management

In the reordering buffer management problem (RBM) asequence of n colored items enters a buffer with limitedcapacity k. When the buffer is full, one item is removedto the output sequence, making room for the next inputitem. This step is repeated until the input sequence is ex-hausted and the buffer is empty. The objective is to finda sequence of removals that minimizes the total numberof color changes in the output sequence. The problem for-malizes numerous applications in computer and productionsystems, and is known to be NP-hard. We give the firstconstant factor approximation guarantee for RBM. Our al-gorithm is based on an intricate “rounding” of the solu-tion to an LP relaxation for RBM, so it also establishes aconstant upper bound on the integrality gap of this relax-ation. Our results improve upon the best previous boundof O(

√log k) of Adamaszek et al. (STOC 2011) that used

different methods and gave an online algorithm. Our con-stant factor approximation beats the super-constant lowerbounds on the competitive ratio given by Adamaszek et al.This is the first demonstration of a polynomial time offlinealgorithm for RBM that is provably better than any onlinealgorithm.

Noa [email protected]

Yuval RabaniThe Hebrew University of [email protected]

CP14

Energy Efficient Scheduling of Parallelizable Jobs

We consider scheduling parallelizable jobs in the non-clairvoyant speed scaling setting to minimize the objectiveof weighted flow time plus energy. Previously, strong lowerbounds for certain classes of algorithms were given evenassuming constant resource augmentation and unweightedjobs. We overcome these lower bounds and give a scalablealgorithm under the natural assumption that algorithmsknow the current parallelizability of alive jobs. Our algo-rithm is also the first positive result in this setting whenconsidering general power functions.


Sungjin ImDuke [email protected]

Benjamin MoseleyToyota Technological Institute at [email protected]

CP14

Lyndon Words and Short Superstrings

In the Shortest Superstring Problem (SSP), we are givena set of strings {s1, . . . , sk} and want to find a string thatcontains all si as substrings and has minimum length. Thisis a classical problem in approximation and the best knownapproximation factor is 2 1

2, given by Sweedyk in 1999.

Since then no improvement has been made, howerever twoother approaches yielding a 2 1

2-approximation algorithms

have been proposed by Kaplan et al. and recently by Paluchet al. — both based on a reduction to maximum asym-metric TSP path (MAX-ATSPP) and structural results ofBreslauer et al. In this paper we give an algorithm thatachieves an approximation ratio of 2 11

23, breaking through

the long-standing bound of 2 12. We use the standard re-

duction of SSP to MAX-ATSPP. The new, somewhat sur-prising, algorithmic idea is to take the better of the twosolutions obtained by using: (a) the currently best 2

3-

approximation algorithm for MAX-ATSPP, and (b) a naıvecycle-cover based 1

2-approximation algorithm. To prove

that this indeed results in an improvement, we further de-velop a theory of string overlaps, extending the results ofBreslauer et al. This theory is based on the novel use ofLyndon words, as a substitute for generic unbordered ro-tations and critical factorizations, as used by Breslauer etal.

Marcin MuchaUniversity of WarsawFaculty of Mathematics, Informatics and [email protected]

CP14

New Approximability Results for Two-DimensionalBin Packing

We study the two-dimensional bin packing problem: Givena list of n rectangles the objective is to find a feasible pack-ing of all rectangles into the minimum number of unit sizedbins. Rotation of the rectangles by 90◦ may be allowed orforbidden. We give a new asymptotic upper bound of ourproblem: For any fixed ε and any instance our algorithmcomputes a packing into (1.5 + ε) · OPT + 69 bins withpolynomial running time.

Lars PraedelUniversity of [email protected]

Klaus JansenUniversity of KielInstitute for Computer [email protected]

CP14

Minimum Makespan Scheduling with Low RankProcessing Times

We investigate approximation algorithms for the classicalminimum makespan scheduling problem, focusing on in-stances where the rank of the matrix describing the pro-cessing times of the jobs is bounded. A bounded rankmatrix arises naturally when the processing time of a jobon machine depends upon a bounded set of resources. Abounded rank matrix also shows up when jobs have vary-ing degrees of parallelizability, and the machines have mul-tiple cores. We are interested in studying the tractabil-ity of the problem as a function of the (positive) rank of

DA13 Abstracts 45

the processing-time matrix. At one extreme is the case ofunit rank, also known as related machines, which admits aPTAS, and at the other extreme is the full rank case (unre-lated machines), which is NP-hard to approximate withina factor better than 3/2. Our main technical contributionis in showing that the approximability of the problem isnot smooth with the rank of the matrix. From the in-approximability side, we show that the problem becomesAPX-hard, even for rank four matrices. For rank sevenmatrices, we prove that it is hard to approximate to a fac-tor 3/2, matching the inapproximability result for generalunrelated machines. From the algorithmic side, we obtaina quasi-polynomial approximation scheme (i.e., a (1+ε) ap-

proximation in time npoly(1/ε,logn)) for the rank two case.This implies that the problem is not APX-hard in this case,unless NP has quasi-polynomial algorithms. Our algorithmis a subtle dynamic program which runs in polynomial timein some interesting special cases. The classification of thethree dimensional problem remains open.

Aditya [email protected]

Ravishankar [email protected]

Kunal TalwarMicrosoft Research Silicon [email protected]

Udi WiederMicrosoft [email protected]

CP15

Simultaneous PQ-Ordering with Applications toConstrained Embedding Problems

In this paper, we define and study the new problem Simul-

taneous PQ-Ordering. Its input consists of a set of PQ-trees, which represent sets of circular orders of their leaves,together with a set of child-parent relations between thesePQ-trees, such that the leaves of the child form a subsetof the leaves of the parent. Simultaneous PQ-Ordering

asks whether orders of the leaves of each of the trees canbe chosen simultaneously, that is, for every child-parentrelation the order chosen for the parent is an extension ofthe order chosen for the child. We show that Simulta-

neous PQ-Ordering is NP-complete in general and weidentify a family of instances that can be solved efficiently,the 2-fixed instances. We show that this result serves as aframework for several other problems that can be formu-lated as instances of Simultaneous PQ-Ordering.

Ignaz RutterKarlsruhe Institute of [email protected]

Thomas BlasiusKarlsruhe Institute of Technology (KIT)[email protected]

CP15

List-Coloring Embedded Graphs

For any fixed surface Σ of genus g, we give an algorithm to

decide whether a graph G of girth at least five embeddedin Σ is colorable from an assignment of lists of size threein time O(|V (G)|). Furthermore, we can allow a subgraph(of any size) with at most s components to be precolored,at the expense of increasing the time complexity of the al-

gorithm to O(|V (G)|K(g+s)+1

)for some absolute constant

K; in both cases, the multiplicative constant hidden inthe O-notation depends on g and s. This also enables usto find such a coloring when it exists. The idea of thealgorithm can be applied to other similar problems, e.g.,5-list-coloring of graphs on surfaces.

Zdenek DvorakComputer Science InstituteCharles University, [email protected]


CP15

Totally Odd Subdivisions and Parity Subdivisions:Structures and Coloring

A totally odd H-subdivision means a subdivision of a graphH in which each edge of H corresponds to a path of oddlength. Thus this concept is a generalization of a sub-division of H . In this paper, we give a structure theo-rem for graphs without a fixed graph H as a totally oddsubdivision. Namely, every graph with no totally odd H-subdivision has a tree-decomposition such that each pieceis either

1. after deleting bounded number of vertices, an “al-most” embedded graph into a bounded-genus surface,or

2. after deleting bounded number of vertices, a bipartitegraph, or

3. after deleting bounded number of vertices, a graphwith maximum degree at most f |H | for some functionf of |H | or a 6|H |-degenerate graph.

We also prove that any graph with no totally odd Kk-subdivision is 79k2/4-colorable. The bound on the chro-matic number is essentially best possible since the correctorder of the magnitude of the chromatic number even forgraphs with no Kk-subdivision is Θk2.


CP15

5-Coloring K3,k-Minor-Free Graphs

We first show that if G is 6-color-critical 4-connected K3,k-minor-free graphs, then G has tree-width at most g(k) forsome function g of k. This allows us to show the followingalgorithmic result, which is of independent interest: Fora 4-connected graph G and any k, there is an O(n3) al-gorithm to test whether or not G is 5-colorable. We alsoshow the following algorithmic result. For a graph G andany k, there is an O(n3) algorithm to output one of thefollowing:

1. a 5-coloring of G, or

2. a K3,k-minor of G, or

3. a minor R of G of order at most f(k) (f(k) comes

46 DA13 Abstracts

from the above theorem) such that R does not have aK3,k-minor nor is 5-colorable.


CP15

Known Algorithms for Edge Clique Cover areProbably Optimal

In the Edge Clique Cover (ECC) problem, given a graphG and an integer k, we ask whether the edges of G canbe covered with k complete subgraphs of G. Gramm etal. [JEA 2008] have shown a set of simple rules that re-duce the number of vertices of G to 2k, and no algorithm isknown with significantly better running time bound than abrute-force search on this reduced instance. In this paperwe show that the approach of Gramm et al. is essentiallyoptimal: we present a polynomial time algorithm that re-duces an arbitrary 3-CNF-SAT formula with n variablesand m clauses to an equivalent ECC instance (G, k) withk = O(log n) and |V (G)| = O(n+m). Consequently, there

is no 22o(k)

poly(n) time algorithm for the ECC problem,unless the Exponential Time Hypothesis fails. To the bestof our knowledge, these are the first results for a natu-ral, fixed-parameter tractable problem, and proving that adoubly-exponential dependency on the parameter is essen-tially necessary.

Marek CyganIDSIA, University of [email protected]

Marcin PilipczukUniversity of [email protected]

Michal PilipczukUniversity of [email protected]

CP16

Fast Matrix Multiplication Using Coherent Config-urations

We introduce a relaxation of the notion of tensor rank,called s-rank, and show that upper bounds on the s-rankof the matrix multiplication tensor imply upper boundson the ordinary rank. In particular, if the “s-rank expo-nent of matrix multiplication’ equals 2, then ω = 2. Usingthis machinery, we generalize the group-theoretic approachof Cohn and Umans from group algebras to general alge-bras. We identify adjacency algebras of coherent configura-tions as a promising family of algebras in the generalizedframework. Finally, we prove a closure property involvingsymmetric powers of adjacency algebras, which enables usto prove nontrivial bounds on ω using commutative coher-ent configurations and suggests that commutative coherentconfigurations may be sufficient to prove ω = 2.

Henry CohnMicrosoft Research New [email protected]

Christopher UmansComputing + Mathematical SciencesCalifornia Institute of Technology

[email protected]

CP16

Lattice Sparsification and the Approximate ClosestVector Problem

We give a deterministic algorithm for solving the (1 + ε)approximate Closest Vector Problem (CVP) on any n di-

mensional lattice and any norm in 2O(n)(1 + 1/ε)n timeand 2npoly(n) space. Our algorithm builds on the latticepoint enumeration techniques of Micciancio and Voulgaris(STOC 2010) and Dadush, Peikert and Vempala (FOCS2011), and gives an elegant, deterministic alternative to the“AKS Sieve’ based algorithms for (1 + ε)-CVP (Ajtai, Ku-mar, and Sivakumar; STOC 2001 and CCC 2002). Further-

more, assuming the existence of a poly(n)-space and 2O(n)

time algorithm for exact CVP in the l2 norm, the spacecomplexity of our algorithm can be reduced to polynomial.Our main technical contribution is a method for “sparsi-fying’ any input lattice while approximately maintainingits metric structure. To this end, we employ the idea ofrandom sublattice restrictions, which was first employedby Khot (FOCS 2003) for the purpose of proving hardnessfor Shortest Vector Problem (SVP) under lp norms.

Daniel DadushGeorgia Institute of [email protected]

Gabor KunCourant Institute for Mathematical [email protected]

CP16

Minimizing the Number of Lattice Points in aTranslated Polygon

The parametric lattice-point counting problem is as fol-lows: Given an integer matrix A ∈ Zm×n, compute anexplicit formula parameterized by b ∈ Rm that deter-mines the number of integer points in the polyhedron{x ∈ Rn : Ax ≤ b}. In the last decade, this countingproblem has received considerable attention in the litera-ture. Several variants of Barvinok’s algorithm have beenshown to solve this problem in polynomial time if the num-ber n of columns of A is fixed. Central to our investigationis the following question:

Can one also efficiently determine a parameter bsuch that the number of integer points in {x ∈Rn : Ax ≤ b} is minimized?

Here, the parameter b can be chosen from a given polyhe-dron Q ⊆ Rm. Our main result is a proof that finding sucha minimizing parameter is NP -hard, even in dimension 2and even if the parametrization reflects a translation of a2-dimensional convex polygon. This result is establishedvia a relationship of this problem to arithmetic progres-sions and simultaneous Diophantine approximation. Onthe positive side we show that in dimension 2 there existsa polynomial time algorithm for each fixed k that eitherdetermines a minimizing translation or asserts that anytranslation contains at most 1 + 1/k times the minimalnumber of lattice points.

Nicolai HahnleTechnische Universitat [email protected]

Friedrich Eisenbrand

DA13 Abstracts 47

Ecole Polytechnique Federale de [email protected]

CP16

Algorithms for the Densest Sub-Lattice Problem

We give algorithms for computing the densest k-dimensional sublattice of an arbitrary lattice, and relatedproblems. This is an important problem in the algorith-mic geometry of numbers that includes as special casesRankin’s problem (which corresponds to the densest sub-lattice problem with respect to the Euclidean norm, andhas applications to the design of lattice reduction algo-rithms), and the shortest vector problem for arbitrarynorms (which corresponds to setting k = 1) and its dual(k = n − 1). Our algorithm works for any norm and has

running time kO(k·n) and uses 2npoly(n) space. In partic-

ular, the algorithm runs in single exponential time 2O(n)

for any constant k = O(1).

Daniel DadushGeorgia Institute of [email protected]

Daniele MicciancioUniversity of California - San [email protected]

CP16

Breaking the nlog n Barrier for Solvable-Group Iso-morphism

The nlog n barrier for group isomorphism has withstood allattacks ever since the nlog n+O(1) generator-enumeration al-gorithm. We present the first significant improvement overnlog n by showing an n(1/2) log n+O(1) algorithm for p-groupisomorphism. We then generalize our techniques from p-groups and derive an n(1/2) logn+O(logn/ log log n) algorithmfor solvable-group isomorphism. Finally, we relate groupisomorphism to the collision problem which allows us re-place the 1/2 in the exponents with 1/4 using randomizedalgorithms and 1/6 using quantum algorithms.

David J. RosenbaumUniversity of [email protected]

CP17

Smoothed Analysis of the Successive Shortest PathAlgorithm

Several algorithms for the minimum-cost flow problem havebeen developed, and it seems that both the problem andthe algorithms are well understood. However, some of thealgorithms’ running times observed in empirical studiescontrast the running times obtained by worst-case analysis.We explain the good performance of the Successive Short-est Path (SSP) algorithm in empirical studies by studyingthe SSP algorithm in the framework of smoothed analysis.

Tobias BrunschUniversity of [email protected]

Kamiel Cornelissen, Bodo MantheyUniversity of Twente

[email protected], [email protected]

Heiko RoglinUniversity of [email protected]

CP17

Breaking the O(n2.5) Deterministic Time Barrierfor Undirected Unit-Capacity Maximum Flow

This paper gives the first o(n2.5) deterministic algorithmfor the maximum flow problem in any undirected unit-capacity graph with no parallel edges. In an n-vertex,m-edge graph with maximum flow value v, our runningtime is O(n9/4v1/8) = O(n2.375). Note that v ≤ n forsimple unit-capacity graphs. The previous deterministicalgorithms [Karger and Levine 1998] achieve O(m+nv3/2)

and O(nm2/3v1/6) time bound, which are both O(n2.5) fordense simple graphs and v = Θ(n).

Ran DuanUniversity of MichiganEECS [email protected]

CP17

Dynamic Graph Connectivity in PolylogarithmicWorst Case Time

The dynamic graph connectivity problem is the following:given a graph on a fixed set of n nodes which is under-going a sequence of edge insertions and deletions, answerqueries of the form q(a, b): “Is there a path between nodesa and b?’ While data structures for this problem with poly-logarithmic amortized time per operation have been knownsince the mid-1990’s, these data structures have Θ(n) worstcase time. In fact, no previously known solution has worstcase time per operation which is o(

√n). We present a

solution with worst case times O(log4 n) per edge inser-tion, O(log5 n) per edge deletion, and O(log n/ log log n)per query. The answer to each query is correct if the an-swer is “yes” and is correct with high probability if theanswer is “no”. The data structure is based on a simplenovel idea which can be used to quickly identify an edge ina cutset. Our technique can be used to simplify and signif-icantly speed up the preprocessing time for the emergencyplanning problem while matching previous bounds for anupdate, and to approximate the sizes of cutsets of dynamicgraphs in time O(min{|S|, |V \S|}) for an oblivious adver-sary.

Bruce M. KapronUniversity of VictoriaComputer Science [email protected]

Valerie King, Ben MountjoyUniversity of [email protected], [email protected]

CP17

Approximate Maximum Flow on Separable Undi-rected Graphs

We present faster algorithms for approximate maximumflow in undirected graphs with good separator struc-tures, such as bounded genus, minor free, and geometric

48 DA13 Abstracts

graphs. Given such a graph with n vertices, m edges alongwith a recursive

√n-vertex separator structure, our algo-

rithm finds an 1 − ε approximate maximum flow in timeO(m6/5poly(ε−1)), ignoring poly-logarithmic terms. Sim-ilar speedups are also achieved for separable graphs withlarger size separators albeit with larger run times. Thesebounds also apply to image problems in two and three di-mensions. Key to our algorithm is an intermediate problemthat we term grouped L2 flow, which exists between maxi-mum flows and electrical flows. Our algorithm also makesuse of spectral vertex sparsifiers in order to remove verticeswhile preserving the energy dissipation of electrical flows.We also give faster spectral vertex sparsification algorithmson well separated graphs, which may be of independent in-terest.

Gary Miller, Richard PengCarnegie Mellon [email protected], [email protected]

CP17

Decremental Maintenance of Strongly ConnectedComponents

We consider the problem of maintaining the strongly con-nected components (SCCs) of an n-nodes and m-edges di-rected graph that undergoes a sequence of edge deletions.Recently, in SODA 2011, �Lacki presented a deterministicalgorithm that preprocess the graph in O(mn) time andcreates a data structure that maintains the SCCs of a graphunder edge deletions with a total update time of O(mn).The data structure answers strong connectivity queries inO(1) time. The worst case update time after a single edgedeletion might be as large as O(mn). In this paper we re-duce the preprocessing time and the worst case update timeof �Lacki’s data structure from O(mn) to O(m log n). Thequery time and the total update time remain unchanged.

Liam RodittyBar-Ilan [email protected]

CP18

Regret Minimization for Reserve Prices in Second-Price Auctions

We show a regret minimization algorithm for setting thereserve price in second-price auctions. We make the as-sumption that all bidders draw their bids from the same un-known and arbitrary distribution. Our algorithm is compu-tationally efficient, and achieves a regret of O(T 1/2), evenwhen the number of bidders is stochastic with a knowndistribution.

Yishay MansourTel Aviv [email protected]

Nicolo’ Cesa-BianchiUniversita’ degli Studi di Milano, [email protected]

Claudio GentileUniversity of [email protected]

CP18

A Unified Approach to Truthful Scheduling on Re-

lated Machines

We present a unified framework for designing determin-istic monotone polynomial time approximation schemes(PTAS’s) for a wide class of scheduling problems on uni-formly related machines. This class includes (among oth-ers) minimizing the makespan, maximizing the minimumload, and minimizing the �p norm of the machine loads vec-tor. Previously, this kind of result was only known for themakespan objective.

Leah EpsteinUniversity of Haifa, [email protected]

Asaf LevinThe [email protected]

Rob van SteeMax Planck Institute for [email protected]

CP18

Online Submodular Welfare Maximization:Greedy Is Optimal

We prove that no online algorithm (even randomized,against an oblivious adversary) is better than 1/2-competitive for welfare maximization with coverage val-uations, unless NP = RP . Since the Greedy algorithmis known to be 1/2-competitive for monotone submodu-lar valuations, of which coverage is a special case, thisproves that Greedy provides the optimal competitive ratio.On the other hand, we prove that Greedy in a stochasticsetting with i.i.d. items and valuations satisfying dimin-ishing returns is (1 − 1/e)-competitive, which is optimaleven for coverage valuations, unless NP = RP . For onlinebudget-additive allocation, we prove that no algorithm canbe 0.612-competitive with respect to a natural LP whichhas been used previously for this problem.

Ian Post, Michael KapralovStanford [email protected], [email protected]


CP18

Towards Polynomial Simplex-Like Algorithms forMarket Equilibria

In this paper we consider the problem of computing mar-ket equilibria in the Fisher setting for utility models suchas spending constraint and perfect, price-discrimination.These models were inspired from modern e-commerce set-tings and attempt to bridge the gap between the computa-tionally hard but realistic separable, piecewise-linear andconcave utility model and, the tractable but less relevantlinear utility case. While there are polynomial time algo-rithms known for these problems, the question of whetherthere exist polynomial time Simplex-like algorithms has re-mained elusive, even for linear markets. Such algorithmsare desirable due to their conceptual simplicity, ease of im-plementation and practicality. This paper takes a signifi-cant step towards this goal by presenting the first Simplex-like algorithms for these markets assuming a positive res-

DA13 Abstracts 49

olution of an algebraic problem of Cucker, Koiran andSmale. Unconditionally, our algorithms are FPTASs; theycompute prices and allocations such that each buyer de-rives at least a 1

1+ε-fraction of the utility at a true mar-

ket equilibrium, and their running times are polynomial inthe input length and 1

ε. We start with convex programs

which capture market equilibria in each setting and, in asystematic way, convert them into linear complementarityproblem (LCP) formulations. Then, departing from previ-ous approaches which try to pivot on a single polyhedronassociated to the LCP obtained, we carefully construct apolynomial-length sequence of polyhedra, one containingthe other, such that starting from an optimal solution toone allows us to obtain an optimal solution to the nextin the sequence in a polynomial number of complementarypivot steps. Our framework to convert a convex programinto an LCP and then come up with a Simplex-like algo-rithm that moves on a sequence of connected polyhedramay be of independent interest.

Jugal Garg, Ruta MehtaGeorgia [email protected], [email protected]

Milind [email protected]

Nisheeth K. VishnoiMicrosoft Research [email protected]

CP18

Communication Complexity of Combinatorial Auc-tions with Submodular Valuations

We prove the first communication complexity lower boundfor constant-factor approximation of the submodular wel-fare problem. More precisely, we show that a (1− 1

2e+ ε)-

approximation (� 0.816) for welfare maximization in com-binatorial auctions with submodular valuations would re-quire exponential communication. We also show NP-hardness of (1− 1

2e+ ε)-approximation in a computational

model where each valuation is given explicitly by a table ofconstant size. Both results rule out better than (1 − 1

2e)-

approximations in every oracle model with a separate ora-cle for each player, such as the demand oracle model. Ourmain tool is a new construction of monotone submodu-lar functions that we call multi-peak submodular functions.Roughly speaking, given a family of sets , we construct amonotone submodular function f with a high value f(S)for every set S ∈ (a “peak”), and a low value on every setthat does not intersect significantly any set in .

Shahar DobzinskiThe Weizmann Institute of Science, [email protected]


CP19

Dichotomy for Holant∗ Problems with a Functionon Domain Size 3

Holant problems are a general framework to study the al-gorithmic complexity of counting problems. Both count-

ing constraint satisfaction problems and graph homomor-phisms are special cases. All previous results of Holantproblems are over the Boolean domain. In this paper, wegive the first dichotomy theorem for Holant problems fordomain size > 2. We discover unexpected tractable familiesof counting problems, by giving new polynomial time al-gorithms. This paper also initiates holographic reductionsin domains of size > 2. This is our main algorithmic tech-nique, and is used for both tractable families and hardnessreductions. The dichotomy theorem is the following: Forany complex-valued symmetric function F with arity 3 ondomain size 3, we give an explicit criterion on F, such thatif F satisfies the criterion then the problem Holant∗(F) iscomputable in polynomial time, otherwise Holant∗(F) is#P-hard.

Jin-Yi CaiUniversity of Wisconsin, [email protected]

Pinyan LuMicrosoft Research [email protected]

Mingji XiaMax Planck Institute for [email protected]

CP19

Skew Bisubmodularity and Valued CSPs

The Valued Constraint Satisfaction Problem (VCSP)amounts to minimising a rational-valued function (of manyvariables) on a finite set, given as a sum of bounded-arity“basic’ functions depending on subsets of the variables. Weshow that, when the set is small, any VCSP with a fixedset of “basic’ functions either satisfies a submodularity-likecondition and can be solved by LP, or else can express MaxCut in a specific way, and so is NP-hard.

Andrei Krokhin, Anna Huber, Robert PowellDurham [email protected],[email protected], [email protected]

CP19

Complexity of Sat Problems, Clone Theory and theExponential Time Hypothesis

The construction of exact exponential-time algorithms forNP-complete problems has for some time been a very ac-tive research area. Unfortunately, there is a lack of generalmethods for studying and comparing the time complex-ity of algorithms for such problems. We propose such amethod based on clone theory and demonstrate it on theSAT problem. Schaefer has completely classified the com-plexity of SAT with respect to the set of allowed relationsand proved that this parameterized problem exhibits a di-chotomy: it is either in P or is NP-complete. We showthat there is a certain partial order on the NP-completeSAT problems with a close connection to their worst-casetime complexities; if a problem SAT(S) is below a prob-lem SAT(S′) in this partial order, then SAT(S′) cannot besolved strictly faster than SAT(S). By using this order,we identify a relation R such that SAT({R}) is the com-putationally easiest NP-complete SAT(S) problem. Thisresult may be interesting when investigating the border-line between P and NP since one appealing way of study-ing this borderline is to identify problems that, in some

50 DA13 Abstracts

sense, are situated close to it. We strengthen the resultby showing that SAT({R})-2 (i.e. SAT({R}) restrictedto instances where no variable appears more than twice)is NP-complete, too. This is in contrast to, for example,1-in-3-SAT, which is in P under the same restriction. Wethen relate SAT({R})-2 to the exponential-time hypothesis(ETH) and show that ETH holds if and only if SAT({R})-2 is not sub-exponential. This constitutes a strong con-nection between ETH and the SAT problem under bothsevere relational and severe structural restrictions. In theprocess, we also prove a stronger version of Impagliazzo etal.’s sparsification lemma for k-SAT; namely that all finiteBoolean constraint languages S and S′ such that SAT(·) isNP-complete can be sparsified into each other.

Victor Lagerkvist, Peter Jonsson, Gustav NordhLinkoping [email protected], [email protected],[email protected]

Bruno ZanuttiniUniversite de Caen [email protected]

CP19

Frozen Variables in Random Boolean ConstraintSatisfaction Problems

We determine the exact freezing threshold, rf , for a familyof models of random boolean constraint satisfaction prob-lems, including NAE-SAT and hypergraph 2-colouring,when the constraint size is sufficiently large. If theconstraint-density of a random CSP, F , in our family isgreater than rf then for almost every solution of F , a linearnumber of variables are frozen, meaning that their colourscannot be changed by a sequence of alterations in whichwe change o(n) variables at a time, always switching toanother solution. If the constraint-density is less than rf ,then almost every solution has o(n) frozen variables.

Michael MolloyUniversity of [email protected]

Ricardo RestrepoUniversidad di [email protected]

CP19

Exponential Lower Bounds for the PPSZ k-Sat Al-gorithm

In 1998, Paturi, Pudlak, Saks, and Zane presented PPSZ,an elegant randomized algorithm for k-SAT. Fourteen yearson, this algorithm is still the fastest known worst-case al-gorithm. They proved that its expected running time onk-CNF formulas with n variables is at most 2(1−εk)n, whereεk ∈ Ω(1/k). So far, no exponential lower bounds at allhave been known. In this paper, we construct hard in-stances for PPSZ. That is, we construct satisfiable k-CNFformulas over n variables on which the expected runningtime is at least 2(1−εk)n, for εk ∈ O(log2 k/k).

Dominik A. SchederAarhus [email protected]

Shiteng Chen, Bangsheng Tang

Tsinghua [email protected], [email protected]

Navid TalebanfardAarhus [email protected]

CP20

Testing Low-Complexity Affine-Invariant Proper-ties

Invariance with respect to linear/affine transformations ofthe domain is arguably the most common symmetry ex-hibited by natural algebraic properties. In this work, weshow that any low complexity affine-invariant property ofmultivariate functions over finite fields is testable with aconstant number of queries. This immediately reproves,for instance, that the Reed-Muller code over Fp of degreed < p is testable, via an argument that uses no detailedalgebraic information about polynomials.

Arnab BhattacharyyaCenter for Computational Intractability, Princeton, NJ,[email protected]

Eldar FischerTechnion Israel Institute of Technology,Haifa, [email protected]

Shachar [email protected]

CP20

Learning Mixtures of Structured Distributions overDiscrete Domains

Let C be a class of probability distributions over the dis-crete domain [n] = {1, . . . , n}. We show that if C satisfiesa rather general condition – essentially, that each distri-bution in C can be well-approximated by a variable-widthhistogram with few bins – then there is a highly efficient(both in terms of running time and sample complexity) al-gorithm that can learn any mixture of k unknown distribu-tions from C. We analyze several natural types of distribu-tions over [n], including log-concave, monotone hazard rateand unimodal distributions, and show that they have therequired structural property of being well-approximated bya histogram with few bins. Applying our general algorithm,we obtain near-optimally efficient algorithms for all thesemixture learning problems.

Ilias DiakonikolasUniversity of [email protected]

Siu-On ChanUC [email protected]

Rocco A. Servedio, Xiaorui SunColumbia [email protected], [email protected]

CP20

Exponentially Improved Algorithms and Lower

DA13 Abstracts 51

Bounds for Testing Signed Majorities

A signed majority function is a linear threshold functionf : {−1, 1}n → {−1, 1} of the form f(x) = sign(

∑n

i=1σixi)

where each σi ∈ {−1, 1}. Signed majority functions are ahighly symmetrical subclass of the class of all linear thresh-old functions. We study the query complexity of testingwhether an unknown f : {−1, 1}n → {−1, 1} is a signedmajority function versus ε-far from every signed majorityfunction. As our main results we exponentially improveprior upper and lower bounds on the number of queriesneeded to test signed majority functions.

Rocco A. ServedioColumbia [email protected]

Dana RonTel Aviv [email protected]

CP20

Learning and Testing Integer Submodular Func-tions

We prove that any submodular function f : {0, 1}n →{0, 1, ..., k} can be represented as a pseudo-Boolean 2k-DNF formula. Pseudo-Boolean DNFs are a natural gen-eralization of DNF representation for functions with inte-ger range. Each term in such a formula has an associatedintegral constant. We show that an analog of switchinglemma holds for pseudo-Boolean k-DNFs if all constantsassociated with the terms of the formula are bounded. Ourcharacterization implies efficient learning and testing algo-rithms for integer submodular functions.

Grigory YaroslavtsevPenn State [email protected]

Sofya RaskdhonikovaPennsylvania State [email protected]

CP20

Learning Disjunctions: Near-Optimal Trade-offBetween Mistakes and “I Don’t Know’s

We develop polynomial-time online algorithms for learningdisjunctions. In this model, we are given an online adver-sarial sequence of inputs for an unknown function of theform f(x1, x2, . . . , xn) =

∨i∈S

xi, and for each such input,we must guess “true’, “false’, or “I don’t know’. On thealgorithm side, we show how to make at most εn mistakes

while answering “I don’t know’ at most (1/ε)2O(1/ε)

n times.

Furthermore, we show how to make O(n log log n

logn

)mistakes

while answering “I don’t know’ O(n2 log log n) times. Wealso show that any algorithm making o(n/ log n) mistakesmust answer “I don’t know’ a superpolynomial number oftimes.

Morteza [email protected]

Erik DemaineMIT

[email protected]

CP21

On Differentially Private Low Rank Approximation

Low rank approximation is a fundamental computationalprimitive widely used in data analysis. In many applica-tions the dataset that the algorithm operates on may con-tain sensitive information about contributing individuals(e.g. user/movie ratings in the Netflix challenge), motivat-ing the need to design low rank approximation algorithmsthat preserve privacy of individual entries of the input ma-trix. In this paper, we give a polynomial time algorithmthat, given a privacy parameter ε > 0, for a symmetricmatrix A, outputs an ε-differentially approximation to theprincipal eigenvector of A, and then show how this algo-rithm can be used to obtain a differentially private rank-kapproximation. We also provide lower bounds showing thatour utility/privacy tradeoff is close to best possible. Whilethere has been significant progress on this problem recentlyfor a weaker notion of privacy, namely (ε, δ)-differentialprivacy [HR12, BBDS12], our result is the first to achieve(ε, 0)-differential privacy guarantees with a near-optimalutility/privacy tradeoff in polynomial time.

Michael KapralovStanford [email protected]

Kunal TalwarMicrosoft Research Silicon [email protected]

CP21

The Power of Linear Reconstruction Attacks

We consider the power of linear reconstruction attacks instatistical data privacy by mounting attacks on any releasewhich gives:

1. the fraction of records that satisfy a given non-degenerate boolean function. Such releases includecontingency tables as well as more complex outputslike the error rate of certain classifiers such as deci-sion trees;

2. any one of a large class of M-estimators including thestandard estimators for linear and logistic regression.

Shiva Prasad KasiviswanathanGE Global [email protected]

Mark RudelsonUniversity of [email protected]

Adam SmithPennsylvania State [email protected]

CP21

An Almost Optimal Algorithm for ComputingNonnegative Rank

Here, we give an algorithm for deciding if the nonnegativerank of a matrix M of dimension m×n is at most r which

runs in time (nm)O(r2). This is the first exact algorithmthat runs in time singly-exponential in r. This algorithm

52 DA13 Abstracts

(and earlier algorithms) are built on methods for finding asolution to a system of polynomial inequalities (if one ex-ists). Notably, the best algorithms for this task run in timeexponential in the number of variables but polynomial inall of the other parameters (the number of inequalities andthe maximum degree). Hence these algorithms motivatenatural algebraic questions whose solution have immediatealgorithmic implications: How many variables do we needto represent the decision problem, does M have nonneg-ative rank at most r? A naive formulation uses nr + mrvariables and yields an algorithm that is exponential in nand m even for constant r. (Arora, Ge, Kannan, Moitra,STOC 2012) recently reduced the number of variables to2r22r, and here we exponentially reduce the number of vari-ables to 2r2 and this yields our main algorithm. In fact, thealgorithm that we obtain is nearly-optimal (under the Ex-ponential Time Hypothesis) since an algorithm that runs

in time (nm)o(r) would yield a subexponential algorithmfor 3-SAT. Our main result is based on establishing a nor-mal form for nonnegative matrix factorization – which inturn allows us to exploit algebraic dependence among alarge collection of linear transformations with variable en-tries. Additionally, we also demonstrate that nonnegativerank cannot be certified by even a very large submatrix ofM , and this property also follows from the intuition gainedfrom viewing nonnegative rank through the lens of systemsof polynomial inequalities.

Ankur [email protected]

CP21

The Simplex Method Is Strongly Polynomial forDeterministic Markov Decision Processes

We prove that the simplex method with the highestgain/most-negative-reduced cost pivoting rule converges instrongly polynomial time for deterministic Markov decisionprocesses (MDPs) regardless of the discount factor. For adeterministic MDP with n states and m actions, we provethe simplex method runs in O(n3m2 log2 n) iterations if thediscount factor is uniform and O(n5m3 log2 n) iterationsif each action has a distinct discount factor. Previouslythe simplex method was known to run in polynomial timeonly for discounted MDPs where the discount was boundedaway from 1 [Ye 2011].

Ian Post, Yinyu YeStanford [email protected], [email protected]

CP21

Turning Big Data into Tiny Data: Constant-SizeCoresets for K-Means, PCA and Projective Clus-tering

Projective Clustering is the task to find k subspaces of di-mension j minimizing the sum of the squared distances toa set of points. It includes popular problems as the k-means problem and principal component analysis. We ob-serve that projecting points to a O(‖/ε)-dimensional bestfit subspace preserves the cost function up to a factor of(1 + ε). Based on this, we provide coresets for projectiveclustering problems containing a number of coreset pointsindependent of the number of input points and their di-mension, leading to fast streaming and parallel algorithms.

Melanie SchmidtTU [email protected]

Dan [email protected]

Christian SohlerDepartment of Computer ScienceTechnische Universitat [email protected]

CP22

Active Self-Assembly of Simple Units Using An In-sertion Primitive

We describe a formal model for studying the complexityof self-assembled structures with active molecular compo-nents that utilize an insertion primitive. We can directlymap our model onto a molecular implementation usingDNA. The expressive power of this language is strongerthan regular languages, but at most as strong as contextfree grammars. We explore the trade-off between the com-plexity of the system and the behavior of the system andspeed of its assembly.

Nadine DabbyCalifornia Institute of [email protected]

Ho-Lin ChenNational Taiwan [email protected]

CP22

Nested Quantum Walks with Quantum Data Struc-tures

We develop a new framework that extends the quantumwalk framework of Magniez, Nayak, Roland, and Santha,by utilizing the idea of quantum data structures to con-struct an efficient method of nesting quantum walks. Sur-prisingly, only classical data structures were considered be-fore for searching via quantum walks. The recently pro-posed learning graph framework of Belovs has yielded im-proved upper bounds for several problems, including tri-angle finding and more general subgraph detection. Weexhibit the power of our framework by giving a simple ex-plicit constructions that reproduce both the O(n35/27) and

O(n9/7) learning graph upper bounds (up to logarithmicfactors) for triangle finding, and discuss how other knownupper bounds in the original learning graph framework canbe converted to algorithms in our framework. We hope thatthe ease of use of this framework will lead to the discoveryof new upper bounds.

Stacey Jeffery, Robin KothariInstitute for Quantum ComputingUniversity of [email protected], [email protected]

Frederic MagniezCNRS, LIAFA, Univ Paris [email protected]

DA13 Abstracts 53

CP22

Improved Quantum Query Algorithms for TriangleFinding and Associativity Testing

We show that the quantum query complexity of detect-ing if an n-vertex graph contains a triangle is O(n9/7).This improves the previous best algorithm of Belovs mak-ing O(n35/27) queries. For the problem of determining ifan operation ◦ : S × S → S is associative, we give an algo-rithm making O(|S|10/7) queries, the first improvement to

the trivial O(|S|3/2) application of Grover search. Our al-gorithms are designed using the learning graph frameworkof Belovs. We give a family of algorithms for detectingconstant-sized subgraphs, which can possibly be directedand colored. These algorithms are designed in a simplehigh-level language; our main theorem shows how this high-level language can be compiled as a learning graph andgives the resulting complexity. The key idea to our im-provements is to allow more freedom in the parametersof the database kept by the algorithm. As in our previ-ous work, the edge slots maintained in the database arespecified by a graph whose edges are the union of regularbipartite graphs, the overall structure of which mimics thatof the graph of the certificate. By allowing these bipartitegraphs to be unbalanced and of variable degree we obtainbetter algorithms.

Frederic MagniezCNRS, LIAFA, Univ Paris [email protected]

Troy LeeCentre for Quantum TechnologiesNational University of [email protected]

Miklos SanthaCNRS, LIAFA, Univ Paris DiderotCQT, National University of [email protected]

CP22

Fuel Efficient Computation in Passive Self-Assembly

In this paper we show that passive self-assembly in the con-text of the tile self-assembly model is capable of performingfuel efficient, universal computation. A computationallyuniversal tile system is said to be fuel efficient if the num-ber of tiles used up per computation step is bounded by aconstant.

Robert T. Schweller, Michael ShermanUniversity of Texas - Pan [email protected], [email protected]

CP22

Efficient Protocols for Generating Bipartite Classi-cal Distributions and Quantum States

We investigate the fundamental problem of generating bi-partite classical distributions or quantum states. By de-signing efficient communication protocols and proving theiroptimality, we establish a number of intriguing connectionsto fundamental measures in optimization, convex geome-try, and information theory.

1. To generate a classical distribution P (x, y), we tightlycharacterize the minimum amount of quantum com-

munication needed by the psd-rank of P (as a ma-trix), a measure recently proposed by Fiorini, Massar,Pokutta, Tiwary and de Wolf (Proceedings of the 44thACM Symposium on Theory of Computing, pages 95-106, 2012) in studies of the minimum size of extendedformulations of optimization problems such as TSP.This echos the previous characterization for the opti-mal classical communication cost by the nonnegativerank of P . The result is obtained via investigating themore general case of bipartite quantum state genera-tion and designing an optimal protocol for it.

2. When an approximation of ε is allowed to generatea distribution (X,Y ) ∼ P , we present a classicalprotocol of the communication cost O((C(X, Y ) +1)/ε), where C(X,Y ) is common information, a well-studied measure in information theory introduced byWyner (IEEE Transactions on Information Theory,21(2):163-179, 1975). This also links nonnegative rankand common information, two seemingly unrelatedquantities in different fields.

3. For approximately generating a quantum pure state|ψ〉, we completely characterize the minimum cost bya corresponding approximate rank, closing a possi-bly exponential gap left in Ambainis, Schulman, Ta-Shma, Vazirani and Wigderson (SIAM Journal onComputing, 32(6):1570-1585, 2003).

Shengyu ZhangThe Chinese University of Hong [email protected]

Rahul JainNational University of [email protected]

Yaoyun ShiUniversity of [email protected]

Zhaohui WeiNational University of [email protected]

CP23

Local-Search Based Approximation Algorithms forMobile Facility Location Problems

We consider the mobile facility location (MFL) problem.We are given a set of facilities and clients located in acommon metric space G = (V, c). The goal is to moveeach facility from its initial location to a destination (inV ) and assign each client to the destination of some fa-cility so as to minimize the sum of the movement-costsof the facilities and the client-assignment costs. We givethe first local-search based approximation algorithm for thisproblem and achieve the best-known approximation guar-antee. Our main result is (3 + ε)-approximation for thisproblem for any constant ε > 0 using local search. Theprevious best guarantee for MFL was an 8-approximationalgorithm due to Friggstad and Salavatipour based on LP-rounding. Our guarantee matches the best-known ap-proximation guarantee for the k-median problem. Sincethere is an approximation-preserving reduction from thek-median problem to MFL, any improvement of our re-sult would imply an analogous improvement for the k-median problem. Furthermore, our analysis is tight (upto o(1) factors) since the tight example for the local-searchbased 3-approximation algorithm for k-median can be eas-ily adapted to show that our local-search algorithm has a

54 DA13 Abstracts

tight approximation ratio of 3. One of the chief noveltiesin the analysis is that in order to generate a suitable collec-tion of local-search moves whose resulting inequalities yieldthe desired bound on the cost of a local-optimum, we de-fine a tree-like structure that (loosely speaking) functionsas a “recursion tree’, using which we spawn off local-searchmoves by exploring this tree to a constant depth. Our re-sults extend to the weighted generalization wherein eachfacility i has a non-negative weight wi and the movementcost for i is wi times the distance traveled by i.

Chaitanya Swamy, Sara Ahmadian, Zachary FriggstadUniversity of [email protected], [email protected], [email protected]

CP23

The Power of Non-Uniform Wireless Power

We study a fundamental measure for wireless interferencein the SINR model. known as (weighted) inductive inde-pendence. This measure characterizes the effectiveness ofusing oblivious power — when the power used by a trans-mitter only depends on the distance to the receiver — as amechanism for improving wireless capacity. We prove opti-mal bounds for inductive independence, implying a numberof algorithmic applications. An algorithm is provided thatachieves — due to existing lower bounds — capacity that isasymptotically best possible using oblivious power assign-ments. Improved approximation algorithms are providedfor a number of problems for oblivious power and for powercontrol, including distributed scheduling, connectivity, sec-ondary spectrum auctions, and dynamic packet scheduling.

Magnus M. HalldorssonReykjavik [email protected]

Stephan HolzerETH [email protected]

Pradipta MitraReykjavik [email protected]

Roger WattenhoferETH [email protected]

CP23

The Diffusion of Networking Technologies

There has been significant interest in the networking com-munity on the impact of cascade effects on the diffusion ofnetworking technology upgrades in the Internet. Thinkingof the global Internet as a graph, where each node repre-sents an economically-motivated Internet Service Provider(ISP), a key problem is to determine the smallest set ofnodes that can trigger a cascade that causes every othernode in the graph to adopt the protocol. We design thefirst approximation algorithm with a provable performanceguarantee for this problem, in a model that captures thefollowing key issue: a node’s decision to upgrade should beinfluenced by the decisions of the remote nodes it wishesto communicate with. Given an internetwork G(V,E) andthreshold function θ, we assume that node u activates (up-grades to the new technology) when it is adjacent to a

connected component of active nodes in G of size exceed-ing node u’s threshold θ(u). Our objective is to choose thesmallest set of nodes that can cause the rest of the graph toactivate. Our main contribution is an approximation algo-rithm based on linear programming, which we complementwith computational hardness results and a near-optimumintegrality gap. Our algorithm, which does not rely on sub-modular optimization techniques, also highlights the sub-stantial algorithmic difference between our problem andsimilar questions studied in the context of social networks.

Zhenming LiuHarvard [email protected]

Sharon GoldbergBoston [email protected]

CP23

Graph Products Revisited: Tight ApproximationHardness of Induced Matching, Poset Dimensionand More

Graph product is a fundamental tool with rich applicationsin both graph theory and theoretical computer science. Itis usually studied in the form f(G ∗ H) where G and Hare graphs, ∗ is a graph product and f is a graph prop-erty. For example, if f is the independence number and ∗is the disjunctive product, then the product is known to bemultiplicative: f(G ∗ H) = f(G)f(H). In this paper, westudy graph products in the following non-standard form:f((G⊕H) ∗ J) where G, H and J are graphs, ⊕ and ∗ aretwo different graph products and f is a graph property. Weshow that if f is the induced and semi-induced matchingnumber, then for some products ⊕ and ∗, it is subadditivein the sense that f((G ⊕ H) ∗ J) ≤ f(G ∗ J) + f(H ∗ J).Moreover, when f is the poset dimension number, it is al-most subadditive. As applications of this result (we onlyneed J = K2 here), we obtain tight hardness of approxima-tion for various problems in discrete mathematics and com-puter science: bipartite induced and semi-induced match-ing (a.k.a. maximum expanding sequences), poset dimen-sion, maximum feasible subsystem with 0/1 coefficients,unit-demand min-buying and single-minded pricing, dona-tion center location, boxicity, cubicity, threshold dimensionand independent packing.

Parinya [email protected]

Danupon NanongkaiUniversity of [email protected]

Bundit LaekhanukitSchool of Computer ScienceMcGill [email protected]

CP23

Approximability and Proof Complexity

This work is concerned with the proof-complexity of cer-tifying that optimization problems do not have good so-lutions. Specifically we consider bounded-degree “Sum ofSquares’ (SOS) proofs, a powerful algebraic proof system

DA13 Abstracts 55

introduced in 1999 by Grigoriev and Vorobjov. Work ofShor, Lasserre, and Parrilo shows that this proof is au-tomatizable using semidefinite programming (SDP), mean-ing that any n-variable degree-d proof can be found intime nO(d). Furthermore, the SDP is dual to the well-known Lasserre SDP hierarchy, meaning that the “d/2-round Lasserre value’ of an optimization problem is equalto the best bound provable using a degree-d SOS proof.These ideas were exploited in a recent paper by Baraket al. (STOC 2012) which shows that the known “hardinstances’ for the Unique-Games problem are in fact opti-mally solved a constant level of the Lasserre SDP hierarchy.We continue the study of the power of SOS proofs in thecontext of difficult optimization problems. In particular,we show that the Balanced-Separator integrality gap in-stances proposed by Devanur et al. can have their optimalvalue certified by a degree-4 SOS proof. The key ingredientis an SOS proof of the KKL Theorem. We also investigatethe extent to which the Khot–Vishnoi Max-Cut integralitygap instances can have their optimum value certified byan SOS proof. We show they can be certified to within afactor .952 (> .878) using a constant-degree proof. Theseinvestigations also raise an interesting mathematical ques-tion: is there a constant-degree SOS proof of the CentralLimit Theorem?

Yuan ZhouCarnegie Mellon [email protected]

Ryan O’DonnellCarnegie Mellon University, [email protected]

CP24

Transforming Curves on Surfaces Redux

Almost exactly 100 years ago, Max Dehn described the firstcombinatorial algorithm to determine whether two givencycles on a compact surface are homotopic, meaning onecycle can be continuously deformed into the other withoutleaving the surface. We describe a simple variant of Dehn’salgorithm that runs in linear time, with no hidden depen-dence on the genus of the surface. Specifically, given twoclosed vertex-edge walks of length at most � in a combi-natorial surface of complexity n, our algorithm determineswhether the walks are homotopic in O(n+ �) time. Our al-gorithm simplifies and corrects a similar algorithm of Deyand Guha [JCSS 1999] and simplifies the more recent algo-rithm of Lazarus and Rivaud [FOCS 2012], who identifieda subtle flaw in Dey and Guha’s results. Our algorithmcombines components of these earlier algorithms, classicalresults in small cancellation theory by Gersten and Short[Inventiones 1990], and simple run-length encoding.

Jeff Erickson, Kim WhittleseyUniversity of Illinois, [email protected], [email protected]

CP24

Higher-Order Geodesic Voronoi Diagrams in aPolygonal Domain with Holes

Given a set of n point sites in a simple polygon withh polygonal holes and c corners, the kth-order geodesicVoronoi diagram partitions the free space into several re-gions such that all points in a region share the same knearest sites. We prove that the total complexity of sucha diagram is Θ(k(n− k) + kc) and the number of faces is

Θ(k(n− k) + kh), and also show that it can be computedin O(k2(n+ c) log(n+ c)) time.

Chih-Hung LiuResearch Center for Information Technology InnovationAcademia Sinicachliu [email protected]

D. T. LeeNational Chung Hsing [email protected]

CP24

Eps-Samples for Kernels

We study the worst case error of kernel density estimatesvia subset approximation. A kernel density estimate of adistribution is the convolution of that distribution with afixed kernel (e.g. Gaussian kernel). Given a subset (i.e.a point set) of the input distribution, we can compare thekernel density estimates of the input distribution with thatof the subset and bound the worst case error. If the max-imum error is ε, then this subset can be thought of as anε-sample (aka an ε-approximation) of the range space de-fined with the input distribution as the ground set andthe fixed kernel representing the family of ranges. Inter-estingly, in this case the ranges are not binary, but have acontinuous range (for simplicity we focus on kernels withrange of [0, 1]); these allow for smoother notions of rangespaces. It turns out, the use of this smoother family ofrange spaces has an added benefit of greatly decreasingthe size required for ε-samples. For instance, in the planethe size is O((1/ε4/3) log2/3(1/ε)) for disks (based on VC-

dimension arguments) but is only O((1/ε)√

log(1/ε)) forGaussian kernels and for kernels with bounded slope thatonly affect a bounded domain. These bounds are accom-plished by studying the discrepancy of these “kernel’ rangespaces, and here the improvement in bounds are even morepronounced. In the plane, we show the discrepancy isO(

√log n) for these kernels, whereas for balls there is a

lower bound of Ω(n1/4).

Jeff PhillipsUniversity of [email protected]

CP24

Morphing Planar Graph Drawings with a Polyno-mial Number of Steps

In 1944, Cairns proved the following theorem: given anytwo straight-line planar drawings of a triangulation withthe same outer face, there exists a morph (i.e., a continu-ous transformation) between the two drawings so that thedrawing remains straight-line planar at all times. Cairns’soriginal proof required exponentially many morphing steps.We prove that there is a morph that consists of O(n2) steps,where each step is a linear morph that moves each vertex atconstant speed along a straight line. Using a known resulton compatible triangulations this implies that for a generalplanar graph G and any two straight-line planar drawingsof G with the same embedding, there is a morph betweenthe two drawings that preserves straight-line planarity andconsists of O(n4) steps.

Anna LubiwUniversity of Waterloo, [email protected]

56 DA13 Abstracts

Soroush AlamdariDavid R. Cheriton School of Computer Science,University of [email protected]

Patrizio AngeliniDipartimento di Informatica e Automazione,Roma Tre [email protected]

Timothy M. ChanDavid R. Cheriton School of Computer Science,University of [email protected]

Giuseppe Di BattistaDipartimento di Informatica e Automazione,Roma Tre [email protected]

Fabrizio FratiSchool of Information Technologies,The University of [email protected]

Maurizio Patrignani, Vincenzo RoselliDipartimento di Informatica e Automazione,Roma Tre [email protected], [email protected]

Sahil SinglaDavid R. Cheriton School of Computer Science,University of [email protected]

Bryan T. WilkinsonMADALGO, Aarhus [email protected]

CP24

Combinatorial and Geometric Properties of PlanarLaman Graphs

Laman graphs naturally arise in structural mechanics andrigidity theory. Specifically, they characterize minimallyrigid planar bar-and-joint systems which are frequentlyneeded in robotics, as well as in molecular chemistry andpolymer physics. We introduce three new combinatorialstructures for planar Laman graphs: angular structures,angle labelings, and edge labelings. The latter two struc-tures are related to Schnyder realizers for maximally planargraphs. We prove that planar Laman graphs are exactlythe class of graphs that have an angular structure that isa tree, called angular tree, and that every angular tree hasa corresponding angle labeling and edge labeling. Usinga combination of these powerful combinatorial structures,we show that every planar Laman graph has an L-contactrepresentation, that is, planar Laman graphs are contactgraphs of axis-aligned L-shapes. Moreover, we show thatplanar Laman graphs and their subgraphs are the onlygraphs that can be represented this way. We present ef-ficient algorithms that compute, for every planar Lamangraph G, an angular tree, angle labeling, edge labeling,and finally an L-contact representation of G. The overallrunning time is O(n2), where n is the number of verticesof G, and the L-contact representation is realized on the

n× n grid.

Stephen KobourovDepartment of Computer ScienceUniversity of [email protected]

Torsten UeckerdtCharles [email protected]

Kevin VerbeekTU [email protected]

CP25

Better Bounds for Matchings in the StreamingModel

In this paper we present improved bounds for approximat-ing maximum matchings in bipartite graphs in the stream-ing model. First, we consider the question of how wellmaximum matching can be approximated in a single passover the input when O(n) space is allowed, where n is thenumber of vertices in the input graph. Two natural vari-ants of this problem have been considered in the litera-ture: (1) the edge arrival setting, where edges arrive inthe stream and (2) the vertex arrival setting, where ver-tices on one side of the graph arrive in the stream to-gether with all their incident edges. The latter settinghas also been studied extensively in the context of on-line algorithms, where each arriving vertex has to eitherbe matched irrevocably or discarded upon arrival. In theonline setting, the celebrated algorithm of Karp-Vazirani-Vazirani achieves a 1 − 1/e approximation by crucially us-

ing randomization (and using O(n) space). Despite thefact that the streaming model is less restrictive in that thealgorithm is not constrained to match vertices irrevocablyupon arrival, the best known approximation in the stream-ing model with vertex arrivals and O(n) space is the samefactor of 1 − 1/e. We show that no single pass stream-

ing algorithm constrained to use O(n) space can achieve abetter than 1− 1/e approximation to maximum matching,even in the vertex arrival setting. This leads to the strik-ing conclusion that no single pass streaming algorithm canget any advantage over online algorithms unless it uses sig-nificantly more than O(n) space. Second, we consider theproblem of approximating matchings in multiple passes inthe vertex arrival setting. We show that a simple frac-tional load balancing approach achieves approximation ra-tio 1 − e−kkk−1/(k − 1)! = 1 − 1√

2πk+ o(1/k) in k passes

using linear space, improving upon the state of the art forthis setting.

Michael KapralovStanford [email protected]

CP25

Beating the Direct Sum Theorem in Communica-tion Complexity with Implications for Sketching

A direct sum theorem for two parties and a function fstates that the communication cost of solving k copiesof f simultaneously with error probability 1/3 is at leastk·R1/3(f), where R1/3(f) is the communication required tosolve a single copy of f with error probability 1/3. We im-prove this for a natural family of functions f , showing that

DA13 Abstracts 57

the 1-way communication required to solve k copies of f si-multaneously with probability 2/3 is Ω(k ·R1/k(f)). SinceR1/k(f) may be as large as Ω(R1/3(f)·log k), we asymptot-ically beat the direct sum bound for such functions, show-ing that the trivial upper bound of solving each of the kcopies of f with probability 1−O(1/k) and taking a unionbound is optimal! In order to achieve this, our direct suminvolves a novel measure of information cost which allowsa protocol to abort with constant probability, and other-wise must be correct with very high probability. Moreover,for the functions considered, we show strong lower boundson the communication cost of protocols with these relaxedguarantees; indeed, our lower bounds match those for pro-tocols that are not allowed to abort. In the distributedand streaming models, where one wants to be correct notonly on a single query, but simultaneously on a sequenceof n queries, we obtain optimal lower bounds on the com-munication or space complexity. Lower bounds obtainedfrom our direct sum result show that a number of tech-niques in the sketching literature are optimal, including thefollowing: dimension of (oblivious) Johnson-Lindenstrausstransforms, multiple �1 or �2-norm estimation, sketch sizefor approximate matrix multiplication and database joinsize estimation.

Marco MolinaroCarnegie Mellon [email protected]


Grigory YaroslavtsevPenn State [email protected]

CP25

(1 + ε)-Approximation for Facility Location in DataStreams

We consider the Euclidean facility location problem withuniform opening cost. We obtain two main results:

• A (1 + ε)-approximation algorithm with running timeO(n log2 n log log n) for constant ε,

• The first (1 + ε)-approximation algorithm for the costof the facility location problem for dynamic geomet-ric data streams, i.e., when the stream consists ofinsert and delete operations of points from a dis-crete space {1, . . . ,Δ}2. The streaming algorithm

uses(log Δ

ε

)O(1)space.

Morteza MonemizadehInstitute of Theoretical InformaticsKarlsruhe Institute of [email protected]

Artur CzumajDepartment of Computer Science and DIMAPUniversity of [email protected]

Christiane LammersenSchool of Computing ScienceSimon Fraser [email protected]

Christian Sohler

Department of Computer ScienceTechnische Universitat [email protected]

CP25

Eigenvalues of a Matrix in the Streaming Model

We study the question of estimating the eigenvalues ofa matrix in the streaming model, addressing a questionposed in [Muthukrishnan ’05]. We show that the eigen-value “heavy hitters’ of a matrix can be computed in a sin-gle pass. In particular, we show that the φ-heavy hitters(in the �1 or �2 norms) can be estimated in space propor-tional to 1/φ2. Such a dependence on φ is optimal. Wealso show how the same techniques may give an estimateof the residual error tail of a rank-k approximation of thematrix (in the Frobenius norm), in space proportional tok2. All our algorithms are linear and hence can support ar-bitrary updates to the matrix in the stream. In fact, whatwe show can be seen as a form of a bi-linear dimensionalityreduction: if we multiply an input matrix with projectionmatrices on both sides, the resulting matrix preserves thetop eigenvalues and the residual Frobenius norm.

Huy NguyenPrinceton [email protected]

Alexandr AndoniMSR Silicon [email protected]

CP25

Space Efficient Streaming Algorithms for the Dis-tance to Monotonicity and Asymmetric Edit Dis-tance

We study streaming algorithms for approximating thelength of the longest increasing sequence (LIS) of an ar-ray. We present a 1-pass algorithm that, for any δ > 0,provides a (1 + δ)-multiplicative approximation to the dis-tance to monotonicity of a data stream of length n (nminusthe length of the LIS), and uses O((log2 n)/δ) space. Ouralgorithm is simple and just 3 lines of pseudocode. Theprevious best known approximation using polylogarithmicspace was a multiplicative 2-factor. Our algorithm esti-mates the length of the LIS to within an additive δn forany δ > 0 while previous algorithms could only achieveadditive error n(1/2 − o(1)).

C. SeshadhriSandia National [email protected]

Michael SaksRutgers [email protected]

CP26

Efficient Algorithms for Computing the Triplet andQuartet Distance Between Trees of Arbitrary De-gree

The triplet and quartet distances are distance measures tocompare two rooted and two unrooted trees, respectively.The leaves of the two trees should have the same set of la-bels. The distances are defined by enumerating all subsetsof three labels (triplets) and four labels (quartets), respec-

58 DA13 Abstracts

tively, and counting how often the induced topologies inthe two input trees are different. In this paper we presentefficient algorithms for computing these distances.

Gerth S. BrodalMADALGO, Aarhus [email protected]

Rolf FagerbergUniversity Of Southern [email protected]

Thomas MailundAarhus UniversityBioinformatics Research [email protected]

Christian N.S. PedersenAarhus UniversityDepartment of Computer [email protected]

Andreas SandAarhus UniversityBioinformatics Research [email protected]

CP26

Discrete Convexity and Polynomial Solvability inMinimum 0-Extension Problems

The minimum 0-extension problem 0-Ext[Γ ] on a graphΓ includes a number of basic combinatorial optimizationproblems, such as minimum (s, t)-cut problem and multi-way cut problem. Karzanov proved that 0-Ext[Γ ] is NP-hard if Γ is not modular or not orientable (in a certainsense). In this paper, we prove the converse: if Γ is ori-entable and modular, then 0-Ext[Γ ] can be solved in poly-nomial time.

Hiroshi HiraiGraduate School of Information Science and TechnologyUniversity of [email protected]

CP26

Improved Algorithms for Constructing ConsensusTrees

A consensus tree is a single phylogenetic tree that sum-marizes the branching structure in a given set of conflict-ing phylogenetic trees. Many different types of consensustrees have been proposed in the literature; three of themost well-known and widely used ones are the majorityrule consensus tree, the loose consensus tree, and the greedyconsensus tree. This paper presents new deterministic al-gorithms for constructing them that are faster than all thepreviously known ones. Given k phylogenetic trees withn leaves each and with identical leaf label sets, our algo-rithms run in O(nk log k) time (majority rule consensustree), O(nk) time (loose consensus tree), and O(n2k) time(greedy consensus tree).

Jesper JanssonKyoto [email protected]

Chuanqi Shen

Stanford University & Raffles [email protected]

Wing-Kin SungNational University of [email protected]

CP26

Mimicking Networks and Succinct Representationsof Terminal Cuts

Given a large edge-weighted network G with k terminalvertices, we wish to compress and store all the minimumterminal-cut values. One appealing methodology is to con-struct a mimicking network : a small network G′ with thesame k terminals, where the minimum terminal-cut val-ues are the same as in G. We provide several new boundson the mimicking network’s size, which together narrowthe previously known gap from doubly-exponential to onlysingly-exponential.

Inbal Rika, Robert KrauthgamerWeizmann Institute of [email protected],[email protected]

CP26

Faster Deterministic Fully-Dynamic Graph Con-nectivity

We give new deterministic bounds for fully-dynamic graphconnectivity. Our data structure supports updates (edgeinsertions/deletions) in O(log2 n/ log log n) amortized timeand connectivity queries in O(log n/ log log n) worst-casetime, where n is the number of vertices of the graph. Thisimproves the deterministic data structures of Holm, deLichtenberg, and Thorup (STOC 1998, J.ACM 2001) andThorup (STOC 2000) which both haveO(log2 n) amortizedupdate time and O(log n/ log log n) worst-case query time.

Christian Wulff-NilsenDepartment of Mathematics and Computer ScienceUniversity of Southern [email protected]

Christian Wulff-NilsenDepartment of Computer ScienceUniversity of [email protected]

CP27

Testing K-Modal Distributions: Optimal Algo-rithms Via Reductions

We give highly efficient algorithms, and almost matchinglower bounds, for a range of basic statistical problems thatinvolve testing and estimating the L1 (total variation) dis-tance between two k-modal distributions p and q over thediscrete domain {1, . . . , n}. As our main conceptual con-tribution, we introduce a new reduction-based approachfor distribution-testing problems that lets us obtain ourresults in a unified way. Roughly speaking, this approachenables us to transform various distribution testing prob-lems for k-modal distributions over {1, . . . , n} to the cor-responding distribution testing problems for unrestricteddistributions over a much smaller domain {1, . . . , �} where

DA13 Abstracts 59

� = O(k log n).

Constantinos [email protected]

Ilias DiakonikolasUniversity of [email protected]

Rocco ServedioColumbia [email protected]

Gregory [email protected]

Paul ValiantUC [email protected]

CP27

Optimal Listing of Cycles and st-Paths in Undi-rected Graphs

We present the first optimal algorithm for the classicalproblem of listing all the cycles in an undirected graph.The time complexity is proportional to the time takento read the input graph plus the time to list the output,namely, the edges in each of the cycles. The algorithm usesa reduction to the problem of listing all the simple pathsfrom a vertex s to a vertex t which we also solve optimally.

Etienne BirmeleUniversite [email protected]

Rui FerreiraUniversity of [email protected]

Roberto GrossiUniversita‘ di [email protected]

Andrea MarinoUniversita di [email protected]

Nadia PisantiUniversita di [email protected]

Romeo RizziUniversita di [email protected]

Gustavo SacomotoINRIA Grenoble Rhone-AlpesUniversite Lyon [email protected]

CP27

Faster Walks in Graphs: A O(n2) Time-Space

Trade-off for Undirected s-t Connectivity

In this paper, we make use of the Metropolis-type walksdue to Nonaka et al. (2010) to provide a faster solution tothe s-t-connectivity problem in undirected graphs (UST-CON). As our main result, we propose a family of random-ized algorithms for USTCON which achieves a time-spaceproduct of S · T = O(n2) in graphs with n nodes and m

edges (where the O-notation disregards poly-logarithmicterms). This improves the previously best trade-off of

O(nm), due to Feige (1995). Our algorithm consists in de-ploying several short Metropolis-type walks, starting fromlandmark nodes distributed using the scheme of Broder etal. (1994) on a modified input graph. In particular, we ob-

tain an algorithm running in time O(n + m) which is, ingeneral, more space-efficient than both BFS and DFS. Weclose the paper by showing how to fine-tune the Metropolis-type walk so as to match the performance parameters (e.g.,average hitting time) of the unbiased random walk for any

graph, while preserving a worst-case bound of O(n2) oncover time.

Adrian KosowskiINRIA Bordeaux [email protected]

CP27

Low-Distortion Inference of Latent Similaritiesfrom a Multiplex Social Network

It is commonly assumed that individuals tend to be moresimilar to their friends than to strangers. Thus, we canview an observed social network as a noisy signal aboutthe latent underlying ”social space”: the way in whichindividuals are (dis)similar. We present near-linear timealgorithms which - under reasonably standard models ofsocial network generation - can infer the similarities fromthe observed network with provable guarantees.

Ittai [email protected]

Shiri ChechikWeizmann Institute of [email protected]

David [email protected]

Aleksandrs SlivkinsMicrosoft [email protected]

CP27

Segmentation of Trajectories on Non-MonotoneCriteria

We study the trajectory segmentation problem, and sug-gest a general framework for solving it, based on the start-stop diagram: a 2-dimensional diagram that represents allvalid and invalid segments of a given trajectory. This yieldstwo subproblems: (i) computing the start-stop diagram,and (ii) finding the optimal segmentation for a given dia-gram. We show that (ii) is NP-hard in general. However,we identify properties of the start-stop diagram that makethe problem tractable, and give polynomial-time algorithm

60 DA13 Abstracts

for this case.

Boris AronovDepartment of Computer Science and Engineering,Polytechnic Institute of [email protected]

Anne Driemel, Marc van Kreveld, Maarten Loffler,Frank StaalsUtrecht [email protected], [email protected], [email protected],[email protected]


speaker Index

61


AAgarwal, Rachit, CP8, 5:20 SunAiger, Dror, CP12, 9:00 MonAlon, Noga, IP3, 11:30 TueAustrin, Per, CP5, 2:00 SunAvigdor-Elgrabli, Noa, CP14, 3:40 MonAzar, Pablo, CP9, 5:20 Sun

bBalcan, Maria-Florina, CP11, 10:40 MonBannister, Michael J., CP13, 2:00 MonBansal, Nikhil, CP10, 10:15 MonBelazzougui, Djamal, CP4, 2:25 SunBen Avraham, Rinat, CP3, 9:25 SunBhaskar, Umang, CP2, 9:00 SunBhattacharyya, Arnab, CP20, 9:25 TueBingmann, Timo, ALENEX , 2:50 MonBläsius, Thomas, CP15, 3:15 MonBona, Miklos, ANALCO , 9:25 SunBradonjic, Milan, ANALCO , 9:50 SunBrakerski, Zvika, CP7, 4:55 SunBrodal, Gerth S., CP26, 6:10 TueBrunsch, Tobias, CP17, 6:10 MonBurton, Benjamin A., CP3, 9:50 Sun

cCai, Jin-Yi, CP19, 9:50 TueCai, Yang, CP9, 4:30 SunChang, Jessica, ALENEX, 9:25 MonChechik, Shiri, CP8, 4:30 SunCheng, Siu-Wing, CP3, 9:00 SunCheraghchi, Mahdi, CP7, 4:30 SunCohn, Henry, CP16, 4:55 Mon

DDabby, Nadine, CP22, 3:40 TueDadush, Daniel, CP16, 5:20 MonDamian, Mirela, CP3, 10:15 SunDevanur, Nikhil R., CP2, 9:25 SunDiakonikolas, Ilias, CP20, 10:40 TueDiakonikolas, Ilias, CP27, 4:30 TueDieudonne, Yoann, CP11, 9:50 MonDinitz, Michael, CP2, 9:50 Sun

Doerr, Benjamin, CP10, 10:40 MonDrmota, Michael, ANALCO, 9:00 SunDuan, Ran, CP17, 5:45 MonDumitrescu, Adrian, CP12, 10:15 MonDutta, Chinmoy, CP11, 9:25 MonDvorak, Zdenek, CP15, 2:25 Mon

eEne, Alina, CP5, 2:50 SunEne, Alina, CP5, 3:15 SunErickson, Jeff, CP24, 2:50 Tue

FFaro, Simone, ALENEX, 3:40 MonFerreira, Rui, CP27, 5:45 TueFirmani, Donatella, ALENEX, 5:45 MonFörster, Klaus-Tycho, ANALCO, 10:15 SunFox, Kyle, CP2, 10:40 SunFox, Kyle, CP6, 2:00 SunFox-Epstein, Eli, ALENEX, 9:50 MonFriggstad, Zachary, CP23, 3:40 Tue

GGaither, Jeffrey B., ANALCO, 2:25 SunGardy, Daniele, ANALCO, 4:55 SunGentile, Claudio, CP18, 4:30 MonGhaffari, Mohsen, CP11, 10:15 MonGrohe, Martin, CP6, 3:40 Sun

HHaeupler, Bernhard, CP11, 9:00 MonHähnle, Nicolai, CP16, 6:10 MonHassanieh, Haitham, CP7, 5:20 SunHirai, Hiroshi, CP26, 4:55 TueHolzer, Stephan, CP23, 3:15 Tue

IIm, Sungjin, CP14, 2:50 Mon

JJacquot, Alice, ANALCO, 2:50 SunJalsenius, Markus, CP10, 9:50 MonJansson, Jesper, CP26, 5:45 TueJeffery, Stacey, CP22, 2:00 TueJurkiewicz, Tomasz, ALENEX, 5:20 Mon

KKapralov, Michael, CP21, 9:00 TueKapralov, Michael, CP25, 4:30 TueKasiviswanathan, Shiva Prasad, CP21, 9:25 TueKawarabayashi, Ken-ichi, CP15, 2:00 MonKawarabayashi, Ken-ichi, CP15, 2:50 MonKerber, Michael, ALENEX, 2:00 MonKing, Valerie, CP17, 4:30 MonKingan, Sandra R., ANALCO, 5:45 SunKlonowski, Marek, ANALCO, 5:20 SunKosowski, Adrian, CP27, 5:20 TueKrenn, Daniel, ANALCO, 2:00 SunKreutzer, Stephan, CP6, 2:25 SunKrokhin, Andrei, CP19, 10:15 Tue

LLagerkvist, Victor, CP19, 9:25 TueLarsen, Kasper G., CP4, 3:40 SunLeckey, Kevin, CP13, 2:50 MonLeconte, Mathieu, CP1, 9:50 SunLee, Euiwoong, CP12, 9:50 MonLee, Troy, CP22, 2:25 TueLevin, Asaf, CP18, 6:10 MonLim, Hyeontaek, ALENEX, 6:10 MonLiu, Chih-Hung, CP24, 2:25 TueLiu, Zhenming, CP23, 2:50 Tue

MMahmoud, Hosam M., ANALCO, 3:15 SunMastrolilli, Monaldo, CP5, 3:40 SunMeng, Xiangrui, CP7, 5:45 SunMicciancio, Daniele, CP16, 5:45 MonMiracle, Sarah, CP1, 9:00 SunMitchell, Joseph, CP12, 10:40 MonMoitra, Ankur, CP21, 10:15 TueMolinaro, Marco, CP25, 6:10 TueMonemizadeh, Morteza, CP25, 5:20 TueMucha, Marcin, CP14, 3:15 Mon

62


NNanongkai, Danupon, CP23, 2:25 TueNavarro, Gonzalo, CP13, 2:25 MonNguyen, Huy, CP25, 5:45 TueNiederbrucker, Gerhard, ALENEX , 4:55 Mon

OOzeki, Kenta, CP6, 2:50 Sun

PPaes Leme, Renato, CP9, 5:45 SunPai, Mallesh M., CP9, 6:10 SunParter, Merav, CP7, 6:10 SunPeng, Richard, CP17, 5:20 MonPhillips, Jeff, CP24, 2:00 TuePilipczuk, Michal, CP6, 3:15 SunPilipczuk, Michal, CP15, 3:40 MonPost, Ian, CP18, 5:20 MonPost, Ian, CP21, 10:40 TuePraedel, Lars, CP14, 2:00 MonPrice, Eric, CP10, 9:25 MonPuglisi, Simon J., ALENEX, 3:15 Mon

RRestrepo, Ricardo, CP19, 10:40 TueRika, Inbal, CP26, 5:20 TueRoditty, Liam, CP17, 4:55 MonRoselli, Vincenzo, CP24, 3:15 TueRosenbaum, David J., CP16, 4:30 MonRucinski, Andrzej, ANALCO, 4:30 Sun

sSauerwald, Thomas, CP1, 9:25 SunScheder, Dominik A., CP19, 9:00 TueSchmidt, Melanie, CP21, 9:50 TueSchweller, Robert T., CP22, 3:15 TueSedgewick, Robert, IP1, 11:30 SunServedio, Rocco A., CP20, 9:00 TueSeshadhri, C., CP25, 4:55 TueSidiropoulos, Anastasios, CP12, 9:25 MonSinop, Ali K., CP5, 2:25 SunSlivkins, Aleksandrs, CP27, 4:55 Tue

Solomon, Shay, CP8, 4:55 SunSommer, Christian, CP8, 6:10 SunSpreer, Jonathan, ALENEX, 2:25 MonStaals, Frank, CP27, 6:10 TueStÖckel, Morten, ALENEX, 4:30 MonStorandt, Sabine, ALENEX, 10:15 Mon

tTao, Yufei, CP13, 3:15 MonThorup, Mikkel, CP4, 2:00 Sun

UUeckerdt, Torsten, CP24, 3:40 TueUno, Takeaki, ALENEX, 9:00 Mon

VVan Walderveen, Freek, CP13, 3:40 MonViola, Emanuele, CP10, 9:00 MonVishnoi, Nisheeth K., CP18, 5:45 MonVondrak, Jan, IP2, 11:30 MonVondrak, Jan, CP18, 4:55 Mon

WWang, Yusu, CP3, 10:40 SunWei, Zhewei, CP4, 3:15 SunWeinberg, Matt, CP9, 4:55 SunWieder, Udi, CP14, 2:25 MonWiese, Andreas, CP2, 10:15 SunWild, Sebastian, ALENEX, 10:40 MonWilkinson, Bryan T., CP4, 2:50 SunWulff-Nilsen, Christian, CP8, 5:45 SunWulff-Nilsen, Christian, CP26, 4:30 Tue

YYaroslavtsev, Grigory, CP20, 9:50 TueYin, Yitong, CP1, 10:15 SunYin, Yitong, CP1, 10:40 Sun

ZZadimoghaddam, Morteza, CP20, 10:15 TueZhang, Shengyu, CP22, 2:50 TueZhou, Yuan, CP23, 2:00 Tue

30 SODA, ALENEX and ANALCO 2013 Conference Program 64

Astor crowne Plaza Hotel Floor Plan

final program & abstracts · membership and customer service: (800) 447-7426 (us & canada)...

Documents