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. It 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 conferences. 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 Mathematics 3600 Market Street, 6th Floor Philadelphia, PA 19104-2688 USA Telephone: +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/da15 http://www.siam.org/meetings/alenex15 http://www.siam.org/meetings/analco15 2 SODA, ALENEX and ANALCO 2015 Conference Program

General Information

Table of Contents Anupam Gupta Thomas Sauerwald General Information...... 2 Carnegie Mellon University, USA University of Cambridge, United Kingdom Get-togethers...... 5 Sariel Har-Peled Christian Scheideler Best Paper Award...... 6 University of Illinois at Urbana- Universität Paderborn, Germany Invited Plenary Presentations...... 8 Champaign, USA Christian Sohler Schedule At A Glance...... 9 Avinatan Hassidim Technische Universität Dortmund, Program Schedule...... 11 Bar-Ilan University, Israel Germany Abstracts...... 26 Sidharth (Sid) Jaggi László A. Végh The Chinese University of Hong Kong, Speaker and Co-author Index...... 61 London School of Economics, United Hong Kong Kingdom Hotel Meeting Room Maps... Back Cover Daniel Krá Yusu Wang University of Warwick, United Kingdom The Ohio State University, USA SODA COMMITTEES Robert Krauthgamer Weizmann Institute of Science, Israel Program Committee Chair Steering Committee Ravi Kumar Mike Molloy Piotr Indyk Google, USA Massachusetts Institute of Technology, University of Toronto, Canada Katrina Ligett USA Dana Randall California Institute of Technology, USA Georgia Institute of Technology, USA Mohammad Mahdian Angelika Steger Program Committee Google Research, USA ETH Zürich, Switzerland Peyman Afshani Yury Makarychev Cliff Stein (Chair) MADALGO, Aarhus University, Toyota Technological Institute at Columbia University, USA Denmark Chicago, USA Shang-Hua Teng Pankaj K. Agarwal Jiri Matousek University of Southern California, USA Duke University, USA Charles University, Czech Republic and Alex Andoni ETH Zurich, Switzerland Microsoft Research, USA Ankur Moitra ALENEX COMMITTEES Michael Bender Massachusetts Institute of Technology, State University of New York at Stony USA Program Committee Brook, USA Shayan Oveis Gharan Co-Chairs Amit Chakrabarti University of California at Berkeley, USA Ulrik Brandes Dartmouth College, USA Debmalya Panigrahi University of Konstanz, Germany Shiri Chechik Duke University, USA David Eppstein Tel Aviv University, Israel and Microsoft Jeff M. Phillips University of California, Irvine, USA Research, USA University of Utah, USA Flavio Chierichetti Marcin Pilipczuk Program Committee Sapienza University of Rome, Italy University of Bergen, Norway and Colin Cooper University of Warsaw, Poland Umut Acar King’s College London, United Kingdom Carnegie Mellon University, USA, and Yuval Rabani INRIA, France Alina Ene Hebrew University of Jerusalem, Israel Princeton University, USA and University Lars Arge Thomas Rothvoß Aarhus University, Denmark of Warwick, United Kingdom University of Washington, USA Andrew V. Goldberg Camil Demetrescu Shubhangi Saraf Sapienza University of Rome, Italy Microsoft Research, USA Rutgers University, USA Roberto Grossi Loukas Georgiadis Università di Pisa, Italy University of Ioannina, Greece Conference Program SODA, ALENEX and ANALCO 2015 3

Andrew V. Goldberg Matthias Müller-Hannemann Conrado Martínez Microsoft Research, USA Martin-Luther-Universität, Halle- Universitat Politècnica de Catalunya, Haim Kaplan Wittenberg, Germany Spain Tel Aviv University, Israel Rajeev Raman Markus Nebel Bernard Moret University of Leicester, United Kingdom University of Kaiserslautern, Germany École Polytechnique Fédérale de Cliff Stein Daniel Panario Lausanne, Switzerland Columbia University, USA Carleton University, Canada Ian Munro Dorothea Wagner Robert Sedgewick University of Waterloo, Canada Karlsruhe Institute of Technology, Princeton University, USA Petra Mutzel Germany Wojciech Szpankowski Technische Universität Dortmund, Renato Werneck Purdue University, USA Germany Microsoft Research, USA Thomas Pajor SIAM Registration Desk Microsoft Research Silicon Valley, USA ANALCO COMMITTEES The SIAM registration desk is located in Roberto Tamassia the California Foyer. It is open during the Brown University, USA Program Committee following hours: Dorothea Wagner Co-Chairs Saturday, January 3 Karlsruhe Institute of Technology, Robert Sedgewick 5:00 PM - 8:00 PM Germany Princeton University, USA

David P. Williamson Mark Daniel Ward Sunday, January 4 Cornell University, USA Purdue University, USA 8:00 AM - 5:00 PM Program Committee Steering Committee James Fill Monday, January 5 Johns Hopkins University, USA David Applegate 8:00 AM - 5:00 PM AT&T Labs, USA Pawel Hitczenko Drexel University, USA Lars Arge Tuesday, January 6 Aarhus University, Denmark Philippe Jacquet Alcatel Lucent Bell Labs, France 8:00 AM - 5:00 PM David Bader Markus Kuba Georgia Institute of Technology, USA Technische Universität Wien, Austria Guy Blelloch Manuel Lladser Hotel Information Carnegie Mellon University, USA University of Colorado Boulder, USA Westin San Diego Gaslamp Quarter Gerth Stølting Brodal Conrado Martínez 910 Broadway Circle Aarhus University, Denmark Universitat Politècnica de Catalunya, San Diego, California 92101 USA Irene Finocchi Spain Phone Number: +1-619-239-2200 Sapienza University of Rome, Italy Ralph Neininger Toll Free Reservations (USA and Mike Goodrich (Chair) J.W. Goethe-Universität, Germany Canada): 1-888-627-8563 University of California, Irvine, USA Daniel Panario Fax: +1-619-239-0509 John Hershberger Carleton University, Canada Mentor Graphics, USA Gilles Schaeffer Hotel web address: David Johnson École Polytechnique, France http://www.starwoodhotels.com/ AT&T Labs Research, USA westin/property/overview/index. html?propertyID=1009 Bernard Moret Steering Committee École Polytechnique Fédérale de Michael Drmota Lausanne, France Technische Universität Wien, Austria Hotel Telephone Number Petra Mutzel H.K. Hwang To reach an attendee or leave a message, Technische Universität Dortmund, Institute of Statistical Science, Academia call +1-619-239-2200. If the attendee is Germany Sinica, Taiwan a hotel guest, the hotel operator can connect you with the attendee’s room. 4 SODA, ALENEX and ANALCO 2015 Conference Program

Hotel Check-in and Corporate/Institutional Members Leading the applied Check-out Times The Aerospace Corporation mathematics community … Check-in time is 3:00 PM. Air Force Office of Scientific Research Join SIAM and save! Check-out time is 12:00 PM. Aramco Services Company SIAM members save up to $130 on AT&T Laboratories - Research full registration for the ACM-SIAM Symposium on Discrete Algorithms Bechtel Marine Propulsion Laboratory (SODA15) and its associated meetings, Child Care The Boeing Company Algorithm Engineering and Experiments The concierge at The Westin San CEA/DAM (ALENEX15) and Analytic Algorithmics Diego Gaslamp Quarter recommends Department of National Defence (DND/ and Combinatorics (ANALCO15). Marion’s Childcare, a family owned CSEC) Join your peers in supporting the childcare center that the concierge premier professional society for applied has been recommending for twenty DSTO- Defence Science and Technology mathematicians and computational years. The Westin San Diego Gaslamp Organisation scientists. SIAM members receive Quarter will validate the parking for all ExxonMobil Upstream Research subscriptions to SIAM Review, SIAM babysitters from Marion’s Childcare. Hewlett-Packard News and SIAM Unwrapped, and enjoy For more information their website is: substantial discounts on SIAM books, IBM Corporation hotelchildcare.com. journal subscriptions, and conference IDA Center for Communications registrations. Research, La Jolla Corporate Members Free Student Memberships are available to IDA Center for Communications students who attend an institution that is an and Affiliates Research, Princeton Academic Member of SIAM, are members SIAM corporate members provide Institute for Computational and of Student Chapters of SIAM, or are their employees with knowledge about, Experimental Research in Mathematics nominated by a Regular Member of SIAM. access to, and contacts in the applied (ICERM) Join onsite at the registration desk, go to mathematics and computational sciences www.siam.org/joinsiam to join online or community through their membership Institute for Defense Analyses, Center download an application form, or contact benefits. Corporate membership is more for Computing Sciences SIAM Customer Service: than just a bundle of tangible products Lawrence Berkeley National Laboratory Telephone: +1-215-382-9800 (worldwide); and services; it is an expression of Lockheed Martin support for SIAM and its programs. or 800-447-7426 (U.S. and Canada only) Los Alamos National Laboratory SIAM is pleased to acknowledge its Fax: +1-215-386-7999 Mathematical Sciences Research corporate members and sponsors. In E-mail: [email protected] recognition of their support, non-member Institute Postal mail: attendees who are employed by the Max-Planck-Institute for Dynamics of Society for Industrial following organizations are entitled to the Complex Technical Systems SIAM member registration rate. and Applied Mathematics Mentor Graphics 3600 Market Street, 6th floor, Philadelphia National Institute of Standards and PA 19104-2688 USA Technology (NIST) National Security Agency (DIRNSA) Oak Ridge National Laboratory, Social Media managed by UT-Battelle for the SIAM is promoting the use of social Department of Energy media, such as Facebook and Twitter, in Sandia National Laboratories order to enhance scientific discussion at its meetings and enable attendees to connect Schlumberger-Doll Research with each other prior to, during and after Tech X Corporation conferences. If you are tweeting about U.S. Army Corps of Engineers, Engineer a conference, please use the designated Research and Development Center hashtag to enable other attendees to keep up with the Twitter conversation and to United States Department of Energy allow better archiving of our conference discussions. The hashtag for SODA is List current November 2014. #SIAMDA15. The hashtags for ALENEX and ANALCO are #ALENEX15 and #ANALCO15. Conference Program SODA, ALENEX and ANALCO 2015 5

Standard Audio/Visual Registration Fee Includes Name Badges Set-Up in Meeting Rooms • Admission to all technical sessions A space for emergency contact SIAM does not provide computers for information is provided on the back of • ANALCO/ALENEX Business any speaker. When giving an electronic your name badge. Help us help you in Meeting presentation, speakers must provide their the event of an emergency! own computers. SIAM is not responsible • Coffee breaks daily for the safety and security of speakers’ • Continental Breakfast daily Comments? computers. • Luncheon on Sunday, January 4, 2015 The Plenary Session Room will have Comments about SIAM meetings are • Proceedings (SODA USB distributed encouraged! Please send to: two (2) screens, one (1) data projector onsite; ALENEX, ANALCO and and one (1) overhead projector. Cables SODA posted online) Cynthia Phillips, SIAM Vice President or adaptors for Apple computers are not for Programs ([email protected]). supplied, as they vary for each model. • Room set-ups and audio/visual Please bring your own cable/adaptor if equipment using an Apple computer. • SODA Business Meeting Get-togethers All other concurrent/breakout rooms • Welcome Reception Welcome Reception will have one (1) screen and one (1) data Saturday, January 3 projector. Cables or adaptors for Apple computers are not supplied, as they vary Job Postings 6:00 PM - 8:00 PM for each model. Please bring your own Please check with the SIAM registration cable/adaptor if using an Apple computer. desk regarding the availability of job ALENEX/ANALCO Business Meeting Overhead projectors will be provided postings or visit http://jobs.siam.org. only if requested. Sunday, January 4 If you have questions regarding 6:45 PM – 7:45 PM availability of equipment in the meeting Conference Sponsors room of your presentation, or to request SODA Business Meeting and an overhead projector for your session, SIAM and the conference organizing Awards Presentation please see a SIAM staff member at the committee wish to extend their thanks registration desk. and appreciation to the following Monday, January 5 sponsors for their support of this 6:45 PM – 7:45 PM conference. Complimentary beer and Internet Access wine will be served. Attendees booked within the SIAM room block will have complimentary wireless Internet access in their guest Please Note rooms. All conference attendees will have SIAM is not responsible for the safety complimentary wireless Internet access in and security of attendees’ computers. the meeting space and lobby area of the Do not leave your laptop computers hotel. unattended. Please remember to turn SIAM will provide a limited number off your cell phones, pagers, etc., during of email stations for attendees during sessions. registration hours.

Recording of Presentations SIAM Books and Journals Audio and video recording of Display copies of books and presentations at SIAM meetings complimentary copies of journals are is prohibited without the written available on site. SIAM books are permission of the presenter and SIAM. 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 2:00 PM on Tuesday, January 6. 6 SODA, ALENEX and ANALCO 2015 Conference Program

Best Paper and Best Student Paper Awards

The following paper has been selected to receive both awards:

The Parameterized Complexity of k-Biclique Bingkai Lin

This paper will be presented in CP9: Session 3C, Sunday, January 4 at 4:30 PM.

See page 16 for session details. Conference Program SODA, ALENEX and ANALCO 2015 7

SIAM Activity Group on Discrete Mathematics (SIAG/DM) www.siam.org/activity/dm A GREAT WAY TO GET INVOLVED! Collaborate and interact with mathematicians and applied scientists whose work involves discrete mathematics.

ACTIVITIES INCLUDE: • Special sessions at SIAM Annual Meetings • Biennial conference on Discrete Mathematics • Co-sponsors the annual ACM-SIAM Symposium on Discrete Algorithms • Dénes König Prize • DM-Net • Website

BENEFITS OF SIAG/DM MEMBERSHIP: • Listing in the SIAG’s online membership directory • Additional $10 discount on registration at SIAM Conference on Discrete Mathematics (excludes student) • Electronic communications from your peers about recent developments in your specialty • Eligibility for candidacy for SIAG/DM office • Participation in the selection of SIAG/DM officers

ELIGIBILITY: • Be a current SIAM member.

COST: • $10 per year • Student members can join two activity groups for free!

2014-15 SIAG/DM OFFICERS • Chair: Doug West, University of Illinois • Vice-Chair: Lenore Cowen, Tufts University • Program Director: Guantao Chen, Georgia State University • Secretary: Rick Brewster, Thompson Rivers University TO JOIN: SIAG/DM: my.siam.org/forms/join_siag.htm SIAM: www.siam.org/joinsiam 8 SODA, ALENEX and ANALCO 2015 Conference Program

Invited Plenary Speakers ** All Invited Plenary Presentations will take place in California B/C--2nd Floor**

Sunday, January 4 11:30 AM - 12:30 PM IP1 Pursuit on a Graph Peter Winkler, Dartmouth College, USA

Monday, January 5 11:30 AM - 12:30 PM IP2 Optimization with Random Inputs Claire Mathieu, CNRS, Ecole Normale Supérieure, France and Brown University, USA

Tuesday, January 6 11:30 AM - 12:30 PM IP3 New Directions in Learning Theory Avrim Blum, Carnegie Mellon University, USA Conference Program SODA, ALENEX and ANALCO 2015 9

SODA15, ALENEX15 and ANALCO15 At-a-Glance

4:05 PM - 4:30 PM 2:00 PM - 4:05 PM 11:05 AM - 11:30 AM Saturday, January 3 Coffee Break Concurrent Sessions Coffee Break California A - 2nd Floor ALENEX: Session 2 California A - 2nd Floor 5:00 PM - 8:00 PM Santa Fe - 2nd Floor 11:30 AM - 12:30 PM Registration 4:30 PM - 6:35 PM CP13 Session 5A New Directions in Learning California Foyer - 2nd Floor IP3 Concurrent Sessions California B - 2nd Floor Theory ANALCO: Session 3 CP14 Session 5B Avrim Blum, Carnegie Mellon 6:00 PM - 8:00 PM Santa Fe - 2nd Floor California C - 2nd Floor University, USA Welcome Reception California B/C - 2nd Floor CP7 Session 3A CP15 Session 5C Garden Terrace - 4th Floor California B - 2nd Floor Plaza - 2nd Floor CP8 Session 3B 12:30 PM - 2:00 PM 4:05 PM - 4:30 PM California C - 2nd Floor Lunch Break Sunday, January 4 Coffee Break Attendees on their own CP9 Session 3C California A - 2nd Floor Plaza - 2nd Floor 8:00 AM - 5:00 PM 4:30 PM - 6:35 PM 2:00 PM - 4:05 PM Registration 6:35 PM - 6:45 PM Concurrent Sessions Concurrent Sessions California Foyer - 2nd Floor Intermission ALENEX: Session 3 CP22 Session 8A Santa Fe - 2nd Floor California B - 2nd Floor 8:30 AM 6:45 PM - 7:45 PM CP16 Session 6A CP23 Session 8B Continental Breakfast ALENEX and ANALCO California B - 2nd Floor California C - 2nd Floor California A - 2nd Floor Business Meeting CP17 Session 6B 9:00 AM - 11:05 AM Santa Fe - 2nd Floor CP24 Session 8C California C - 2nd Floor Plaza - 2nd Floor Concurrent Sessions CP18 Session 6C ANALCO: Session 1 4:05 PM - 4:30 PM Monday, January 5 Plaza - 2nd Floor Santa Fe - 2nd Floor Coffee Break CP1 Session 1A 6:35 PM - 6:45 PM California A - 2nd Floor California B - 2nd Floor 8:00 AM - 5:00 PM Intermission Registration 4:30 PM - 6:35 PM CP2 Session 1B California Foyer - 2nd Floor Concurrent Sessions California C - 2nd Floor 6:45 PM - 7:45 PM CP25 Session 9A 8:30 AM SODA Business Meeting California B - 2nd Floor CP3 Session 1C and Awards Presentation Plaza - 2nd Floor Continental Breakfast California A - 2nd Floor California B - 2nd Floor CP26 Session 9B California C - 2nd Floor Complimentary beer and 11:05 AM - 11:30 AM 9:00 AM - 11:05 AM wine will be served. CP27 Session 9C Coffee Break Concurrent Sessions Plaza - 2nd Floor California A - 2nd Floor ALENEX: Session 1 Santa Fe - 2nd Floor 11:30 AM - 12:30 PM Tuesday, January 6 IP1 Pursuit on a Graph CP10 Session 4A Peter Winkler, Dartmouth College, California B - 2nd Floor 8:00 AM - 5:00 PM USA CP11 Session 4B Registration Key to abbreviations California B/C - 2nd Floor California C - 2nd Floor California Foyer - 2nd Floor and symbols 12:30 PM - 2:00 PM CP12 Session 4C Luncheon Plaza - 2nd Floor = Award Presentation **Ticketed Event** 8:30 AM San Diego Ballroom - 4th Floor 11:05 AM - 11:30 AM Continental Breakfast Coffee Break California A - 2nd Floor = Business Meeting 2:00 PM - 4:05 PM California A - 2nd Floor Concurrent Sessions = Coffee Break ANALCO: Session 2 11:30 AM - 12:30 PM 9:00 AM - 11:05 AM Santa Fe - 2nd Floor IP2 Optimization with Random Concurrent Sessions Inputs CP19 Session 7A = Continental Breakfast CP4 Session 2A Claire Mathieu, CNRS, Ecole California B - 2nd Floor California B - 2nd Floor and Luncheon Normale Supérieure, France and CP20 Session 7B CP5 Session 2B Brown University, USA California C - 2nd Floor California C - 2nd Floor California B/C - 2nd Floor = Refreshments CP21 Session 7C CP6 Session 2C 12:30 PM - 2:00 PM Plaza - 2nd Floor Plaza - 2nd Floor Lunch Break Attendees on their own 10 SODA, ALENEX and ANALCO 2015 Conference Program

June 15-June 26, 2015 Delphi, Greece

RandNLA: Randomization in Numerical Linear Algebra

The sixth Gene Golub SIAM Summer School, hosted by the University of Patras, will take place at the European Cultural Centre of Delphi (ECCD), located in Delphi, Greece. The goal of RandNLA is to design novel algorithms for numerical linear algebra problems by using randomization, e.g., random sampling and random projections. It is a topic that has received a great deal of interdisciplinary interest in recent years, with contributions coming from numerical linear algebra, theoretical computer science, scientific computing, statistics, optimization, data analysis, and machine learning, as well as application areas such as genetics, physics, astronomy, and internet modeling. The summer school is designed to bring graduate students up to date on the state of the art in the theory, numerical aspects, and data analysis applications of RandNLA. Since RandNLA is quite interdisciplinary, students will be selected from a wide range of backgrounds. Thus the courses are designed to provide students with an overview of RandNLA and also an understanding of the complementary strengths and weaknesses that different traditional research areas bring to this new research area. The school will include four classes, three tutorials, and several associated project working group sessions. The topics include: • Interdisciplinary Overview of RandNLA • Numerical Analysis Perspectives on RandNLA • Theoretical Computer Science Perspectives on RandNLA • RandNLA in Applied Mathematics • Solving Symmetric Diagonally-Dominant Systems in Input Sparsity Time • Matrix Concentration Bounds and Graph Sparsification • Industrial Uses of RandNLA

The Summer school is being organized by • Petros Drineas, Rensselaer Polytechnic Institute, USA • Efstratios Gallopoulos, University of Patras, Greece • Ilse Ipsen, North Carolina State University, USA • Michael W. Mahoney, University of California at Berkeley, USA © Ben Lowery

Applicants selected to participate pay no registration. Funding for local accommodations and meal expenses will be available for all participants. Limited travel funds are also available.

Application deadline: February 1, 2015

As information is available on the courses and on how to apply, it will be posted at: www.siam.org/students/g2s3/

Sponsored by SIAM through an endowment from the estate of Gene Golub. For more information about prior summer schools and Professor Gene Golub go to www.siam.org/students/g2s3/.

Society for Industrial and Applied Mathematics 3600 Market Street, 6th Floor • Philadelphia, PA 19104-2688 USA • +1-215-382-9800 • [email protected] • www.siam.org Conference Program SODA, ALENEX and ANALCO 2015 11

SODA, ALENEX and ANALCO Program 12 SODA, ALENEX and ANALCO 2015 Conference Program

Saturday, Sunday, January 4 Sunday, January 4 January 3 ANALCO: Session 1 Registration 9:00 AM-11:05 AM 8:00 AM-5:00 PM Room:Santa Fe - 2nd Floor Registration Room:California Foyer - 2nd Floor Chair: James Fill, Johns Hopkins 5:00 PM-8:00 PM University, USA Room:California Foyer - 2nd Floor 9:00-9:20 On k-Connectivity and Minimum Vertex Degree in Random Continental Breakfast s-Intersection Graphs 8:30 AM Jun Zhao, Osman Yagan, and Virgil Welcome Reception Gligor, Carnegie Mellon University, Room:California A - 2nd Floor 6:00 PM-8:00 PM USA Room:Garden Terrace - 4th Floor 9:25-9:45 On the Algorithmic Lovász Local Lemma and Acyclic Edge Coloring Ioannis Giotis, Lefteris Kirousis, and Kostas Psaromiligkos, University of Athens, Greece; Dimitrios Thilikos, CNRS, France 9:50-10:10 A Bound for the Diameter of Random Hyperbolic Graphs Marcos Kiwi, University of Chile, Chile; Dieter Mitsche, Université de Nice, Sophia Antipolis, France 10:15-10:35 Analytic Samplers and the Combinatorial Rejection Method Jérémie Lumbroso, Princeton University, USA; Olivier Bodini and Nicolas Rolin, Université Paris XIII, France 10:40-11:00 Linear-Time Generation of Inhomogeneous Random Directed Walks Andrea Sportiello and Frédérique Bassino, Universite Paris-Nord, France Conference Program SODA, ALENEX and ANALCO 2015 13

Sunday, January 4 Sunday, January 4 Sunday, January 4 CP1 CP2 CP3 Session 1A Session 1B Session 1C 9:00 AM-11:05 AM 9:00 AM-11:05 AM 9:00 AM-11:05 AM Room:California B - 2nd Floor Room:California C - 2nd Floor Room:Plaza - 2nd Floor Chair: Debmalya Panigrahi, Duke Chair: Anupam Gupta, Carnegie Mellon Chair: Kasturi Varadarajan, University University, USA University, USA of Iowa, USA 9:00-9:20 Approximating Independent 9:00-9:20 Welfare Maximization with 9:00-9:20 Minimum Forcing Sets for Sets in Sparse Graphs Production Costs: A Primal Dual Miura Folding Patterns Nikhil Bansal, Eindhoven University of Approach Brad Ballinger, Humboldt State Technology, Netherlands Zhiyi Huang, University of Hong Kong, University, USA; Mirela Damian, Hong Kong; Anthony Kim, Stanford 9:25-9:45 Spider Covers for Prize- Villanova University, USA; David Collecting Network Activation Problem University, USA Eppstein, University of California, Takuro Fukunaga, National Institute of 9:25-9:45 Pricing Online Decisions: Irvine, USA; Robin Flatland, Siena Informatics, Japan Beyond Auctions College, USA; Jessica Ginepro, Ilan R. Cohen, Alon Eden, and Amos University of Connecticut, USA; 9:50-10:10 On Survivable Set Thomas Hull, Western New England Connectivity Fiat, Tel Aviv University, Israel; Parinya Chalermsook, Max Planck Lukasz Jez, Tel Aviv University, Israel University, USA Institute for Informatics, Germany; and University of Wroclaw, Poland 9:25-9:45 Universal Computation with Fabrizio Grandoni, University 9:50-10:10 An n-to-1 Bidder Reduction Arbitrary Polyomino Tiles in Non- of Lugano, Switzerland; Bundit for Multi-Item Auctions and Its Cooperative Self-Assembly Laekhanukit, McGill University, Applications Sándor P. Fekete, Technische Canada Andrew C. Yao, Tsinghua University, P. Universität Braunschweig, Germany; R. China Jacob Hendricks, Matthew Patitz, and 10:15-10:35 A Note on the Ring Trent Rogers, University of Arkansas, Loading Problem 10:15-10:35 On the Complexity USA; Robert T. Schweller, University Martin Skutella, Technical University of of Computing An Equilibrium in of Texas - Pan American, USA Berlin, Germany Combinatorial Auctions Shahar Dobzinski, Weizmann Institute 9:50-10:10 Efficient and Robust 10:40-11:00 New Approximation of Science, Israel; Hu Fu, Microsoft Persistent Homology for Measures Schemes for Unsplittable Flow on a Research, USA; Robert Kleinberg, Mickaël Buchet, Frédéric Chazal, and Path Cornell University, USA Jatin Batra, Naveen Garg, and Amit Steve Oudot, INRIA Saclay Ile-de- Kumar, IIT Delhi, India; Tobias 10:40-11:00 Combinatorial Auctions France, France; Donald Sheehy, Mömke, Saarland University, Germany; Via Posted Prices University of Connecticut, USA Andreas Wiese, MPII Saarbrücken, Michal Feldman, Tel Aviv University, 10:15-10:35 Zigzag Persistence Via Germany Israel; Nick Gravin and Brendan Reflections and Transpositions Lucier, Microsoft Research, USA Clément Maria, INRIA Sophia Antipolis, France; Steve Oudot, INRIA Saclay Ile-de-France, France 10:40-11:00 Improved Bounds for Orthogonal Point Enclosure Query and Point Location in Orthogonal Subdivisions in R3 Saladi Rahul, University of Minnesota, Twin Cities, USA

Coffee Break 11:05 AM-11:30 AM Room:California A - 2nd Floor 14 SODA, ALENEX and ANALCO 2015 Conference Program

Sunday, January 4 Sunday, January 4 Sunday, January 4 IP1 ANALCO: Session 2 CP4 Pursuit on a Graph 2:00 PM-4:05 PM Session 2A 11:30 AM-12:30 PM Room:Santa Fe - 2nd Floor 2:00 PM-4:05 PM Room:California B/C - 2nd Floor Chair: Mark Daniel Ward, Purdue Room:California B - 2nd Floor Chair: Piotr Indyk, Massachusetts University, USA Chair: Yusu Wang, Ohio State Institute of Technology, USA 2:00-2:20 Cuts in Increasing Trees University, USA Pursuit games---motivated historically Antoine Genitrini, UPMC, France; 2:00-2:20 Speeding Up the Four by military tactics---are a natural for Olivier Bodini, Université Paris XIII, Russians Algorithm by About One graphical settings, and take many forms. France More Logarithmic Factor We will present some recent results 2:25-2:45 Repeated Fringe Subtrees in Timothy M. Chan, University of involving (among other things) drunks, Random Rooted Trees Waterloo, Canada Kakeya sets and a 'ketchup graph.’ Dimbinaina Ralaivaosaona and Stephan 2:25-2:45 More Applications of the Lastly, we discuss what we think is Wagner, Stellenbosch University, Polynomial Method to Algorithm the most important open problem in the South Africa Design field. 2:50-3:10 Bootstrap Percolation on Amir Abboud, Ryan Williams, and Huacheng Yu, Stanford University, Peter Winkler Periodic Trees USA Dartmouth College, USA Milan Bradonjic and Iraj Saniee, Bell Labs, Alcatel-Lucent, USA 2:50-3:10 Beating Exhaustive Search 3:15-3:35 Variance of Size in Regular for Quantified Boolean Formulas and Graph Tries Connections to Circuit Complexity Luncheon Philippe Jacquet, Alcatel-Lucent Bell Rahul Santhanam, University of **Ticketed Event** Laboratories, France; Abram Magner, Edinburgh, United Kingdom; Ryan 12:30 PM-2:00 PM Purdue University, USA Williams, Stanford University, USA 3:15-3:35 Degree-3 Treewidth Room:San Diego Ballroom - 4th Floor 3:40-4:00 Lattice Paths of Slope 2/5 Michael Wallner, Technische Sparsifiers Please visit the SIAM Registration Desk Universitaet Wien, Austria; Cyril Chandra Chekuri, University of Illinois if you require a ticket. Banderier, Universite Paris-Nord, at Urbana-Champaign, USA; Julia France Chuzhoy, Toyota Technological Institute at Chicago, USA 3:40-4:00 Improved Bounds for the Flat Wall Theorem Julia Chuzhoy, Toyota Technological Institute at Chicago, USA Conference Program SODA, ALENEX and ANALCO 2015 15

Sunday, January 4 Sunday, January 4 Sunday, January 4 CP5 CP6 ANALCO: Session 3 Session 2B Session 2C 4:30 PM-5:20 PM 2:00 PM-4:05 PM 2:00 PM-4:05 PM Room:Santa Fe - 2nd Floor Room:California C - 2nd Floor Room:Plaza - 2nd Floor Chair: Daniel Panario, Carleton Chair: Kasturi Varadarajan, University Chair: Sungjin Im, University of University, Canada of Iowa, USA California, Merced, USA 4:30-4:50 Analysis of Branch Misses in 2:00-2:20 Fast Lattice Point 2:00-2:20 (2 ∆ –1)-Edge-Coloring Is Quicksort Enumeration with Minimal Overhead Much Easier Than Maximal Matching Conrado Martínez, Universidad Daniele Micciancio and Michael Walter, in the Distributed Setting Politecnica de Catalunya, Spain; University of California, San Diego, Michael Elkin, Ben Gurion University, Markus Nebel and Sebastian USA Israel; Seth Pettie, University of Wild, University of Kaiserslautern, Germany 2:25-2:45 Short Paths on the Voronoi Michigan, Ann Arbor, USA; Hsin- Graph and Closest Vector Problem Hao Su, University of Michigan, USA 4:55-5:15 On Distributed Cardinality with Preprocessing 2:25-2:45 Plurality Consensus in the Estimation: Random Arcs Recycled Nicolas Bonifas, École Polytechnique, Gossip Model Marcin Kardas, Miroslaw Kutylowski, France and IBM, USA; Daniel Dadush, Luca Becchetti, Universita di Roma, and Jakub Lemiesz, Wroclaw New York University, USA La Sapienza, Italy; Andrea Clementi, University of Technology, Poland 2:50-3:10 On Largest Volume Simplices Universita’ di Roma Tor Vergata, and Sub-Determinants Italy; Emanuele Natale, Francesco Marco Di Summa, Università degli Studi Pasquale, and Riccardo Silvestri, di Padova, Italy; Friedrich Eisenbrand, Universita di Roma, La Sapienza, Italy Yuri Faenza, and Carsten Moldenhauer, 2:50-3:10 Distributed Computation of École Polytechnique Fédérale de Large-Scale Graph Problems Lausanne, Switzerland Hartmut Klauck and Danupon 3:15-3:35 Approximating Hereditary Nanongkai, Nanyang Technological Discrepancy Via Small Width Ellipsoids University, Singapore; Gopal Aleksandar Nikolov, Rutgers University, Pandurangan, Nanyang Technological USA; Kunal Talwar, Unaffiliated University, Singapore; Peter Robinson, National University of Singapore, 3:40-4:00 Approximate Nearest Line Search in High Dimensions Singapore Sepideh Mahabadi, Massachusetts 3:15-3:35 Gossip vs. Markov Chains, Institute of Technology, USA and Randomness-Efficient Rumor Spreading Zeyu Guo, California Institute of Technology, USA; He Sun, Max Planck Institute for Informatics, Germany 3:40-4:00 Sequential Random Permutation, List Contraction and Tree Contraction Are Highly Parallel Guy Blelloch, Carnegie Mellon University, USA; Jeremy Fineman, Georgetown University, USA; Phillip Gibbons, Intel Labs and Carnegie Mellon University, USA; Yan Gu and Julian Shun, Carnegie Mellon University, USA

Coffee Break 4:05 PM-4:30 PM Room:California A - 2nd Floor 16 SODA, ALENEX and ANALCO 2015 Conference Program

Sunday, January 4 Sunday, January 4 Sunday, January 4 CP7 CP8 CP9 Session 3A Session 3B Session 3C 4:30 PM-6:35 PM 4:30 PM-6:35 PM 4:30 PM-6:35 PM Room:California B - 2nd Floor Room:California C - 2nd Floor Room:Plaza - 2nd Floor Chair: Alexandr Andoni, Unaffiliated Chair: Yusu Wang, Ohio State Chair: Julia Chuzhoy, Toyota 4:30-4:50 Robust Probabilistic University, USA Technological Institute at Chicago, USA Inference 4:30-4:50 Internal Pattern Matching 4:30-4:50 The Parameterized Yishay Mansour, Tel Aviv University Queries in a Text and Applications Complexity of K and Microsoft Research, Israel; Aviad Tomasz Kociumaka and Jakub Bingkai Lin, University of Tokyo, Japan Rubinstein, University of California, Radoszewski, University of Warsaw, 4:55-5:15 Characterizing the Easy-to- Berkeley, USA; Moshe Tennenholtz, Poland; Wojciech Rytter, University Find Subgraphs from the Viewpoint of Microsoft Research and Technion - of Warsaw and Nicolaus Copernicus Polynomial-Time Algorithms, Kernels, Israel Institute of Technology, Israel University in Torun, Poland; Tomasz and Turing Kernels 4:55-5:15 Learning Privately with Walén, University of Warsaw, Poland Bart Jansen, Eindhoven University of Labeled and Unlabeled Examples Technology, Netherlands; Daniel Marx, 4:55-5:15 Cell-Probe Bounds for Hungarian Academy of Sciences, Amos Beimel, Ben Gurion University, Online Edit Distance and Other Israel; Kobbi Nissim and Uri Stemmer, Pattern Matching Problems Hungary Ben-Gurion University, Israel Raphael Clifford, Markus Jalsenius, 5:20-5:40 Solving D-Sat Via Backdoors 5:20-5:40 Learning from Satisfying and Benjamin G. Sach, University of to Small Treewidth Assignments Bristol, United Kingdom Fedor Fomin and Daniel Lokshtanov, University of Bergen, Norway; Ilias Diakonikolas, University of 5:20-5:40 A New Characterization of Edinburgh, United Kingdom; Anindya Maximal Repetitions by Lyndon Trees Neeldhara Misra, Indian Institute of De, Institute for Advanced Study, Hideo Bannai, Kyushu University, Science, Bangalore, India; Ramanujan USA; Rocco A. Servedio, Columbia Japan; Tomohiro I, Technische M. Sridharan, University of Bergen, University, USA Universität Dortmund, Germany; Norway; Saket Saurabh, Institute of Mathematical Sciences, India 5:45-6:05 Approximate Resilience, Shunsuke Inenaga, Yuto Nakashima, Monotonicity, and the Complexity of Masayuki Takeda, and Kazuya 5:45-6:05 An Exact Characterization Agnostic Learning Tsuruta, Kyushu University, Japan of Tractable Demand Patterns for Dana Dachman-Soled, University of Maximum Disjoint Path Problems 5:45-6:05 Wavelet Trees Meet Suffix Dániel Marx, Hungarian Academy of Maryland, USA; Vitaly Feldman, Trees Sciences, Hungary; Paul Wollan, IBM Research, USA; Li-Yang Tan, Maxim Babenko, National Research University of Rome La Sapienza, Italy Columbia University, USA; Andrew University Higher School of Wan, University of California, Economics, Russia; Paweł 6:10-6:30 Tight Lower Bound for the Berkeley, USA; Karl Wimmer, Gawrychowski, Max Planck Institute Channel Assignment Problem Duquesne University, USA for Informatics, Germany; Tomasz Arkadiusz Socala, University of Warsaw, 6:10-6:30 Linear-Programming Based Kociumaka, University of Warsaw, Poland Approximation Algorithms for Multi- Poland; Tatiana Starikovskaya, Vehicle Minimum Latency Problems National Research University Higher Intermission Ian Post and Chaitanya Swamy, School of Economics, Russia University of Waterloo, Canada 6:10-6:30 Sharp Bounds on Formation- 6:35 PM-6:45 PM Free Sequences Seth Pettie, The University of Michigan, Ann Arbor, USA ALENEX and ANALCO Business Meeting 6:45 PM-7:45 PM Room:Santa Fe - 2nd Floor Conference Program SODA, ALENEX and ANALCO 2015 17

Monday, January 5 Monday, January 5 Monday, January 5 ALENEX: Session 1 CP10 9:00 AM-11:05 AM Session 4A Registration Room:Santa Fe - 2nd Floor 9:00 AM-11:05 AM 8:00 AM-5:00 PM Chair: Lars Arge, Aarhus University, Room:California B - 2nd Floor Room:California Foyer - 2nd Floor Denmark Chair: Robert Krauthgamer, Weizmann 9:00-9:20 Cache Replacement with Institute of Science, Israel Memory Allocation 9:00-9:20 Improved Region-Growing Continental Breakfast Shahram Ghandeharizadeh, University of Southern California, USA; Sandy and Combinatorial Algorithms for 8:30 AM Irani and Jenny Lam, University of k-Route Cut Problems California, Irvine, USA Guru Guruganesh, Carnegie Mellon Room:California A - 2nd Floor University, USA; Laura Sanita and 9:25-9:45 A Data-Aware Fm-Index Chaitanya Swamy, University of Hongwei Huo, Longgang Chen, and Waterloo, Canada Heng Zhao, Xidian University, China; Jeffrey Scott Vitter, University 9:25-9:45 On Uniform Capacitated K-Median Beyond the Natural Lp of Kansas, USA; Yakov Nekrich, Relaxation University of Waterloo, Canada; Qiang Shi Li, Toyota Technological Institute at Yu, Xidian University, China Chicago, USA 9:50-10:10 Improved Single-Term Top-K 9:50-10:10 Dynamic Facility Location Document Retrieval via Exponential Clocks Simon Gog, Karlsruhe Institute of Hyung-Chan An, Ashkan Norouzi- Technology, Germany; Gonzalo Fard, and Ola Svensson, École Navarro, University of Chile, Chile Polytechnique Fédérale de Lausanne, 10:15-10:35 (Semi-)External Algorithms Switzerland For Graph Partitioning and Clustering 10:15-10:35 Bi-Factor Approximation Yaroslav Akhremtsev, Peter Sanders, and Algorithms for Hard Capacitated Christian Schulz, Karlsruhe Institute of K-Median Problems Technology, Germany Jaroslaw Byrka, University of Wroclaw, 10:40-11:00 The 1-2-3-Toolkit for Poland; Krzysztof Fleszar, University Building Your Own Balls-into-Bins of Würzburg, Germany; Bartosz Algorithm Rybicki, University of Wroclaw, Pierre Bertrand, Ecole Normale Poland; Joachim Spoerhase, Superieure de Cachan, France; University of Würzburg, Germany Christoph Lenzen, Max Planck Institute 10:40-11:00 An Improved for Informatics, Germany Approximation for k-Median, and Positive Correlation in Budgeted Optimization Jaroslaw Byrka, University of Wroclaw, Poland; Thomas Pensyl, University of Maryland, College Park, USA; Bartosz Rybicki, University of Wroclaw, Poland; Aravind Srinivasan, University of Maryland, USA; Khoa Trinh, University of Maryland, College Park, USA 18 SODA, ALENEX and ANALCO 2015 Conference Program

Monday, January 5 Monday, January 5 Monday, January 5 CP11 CP12 IP2 Session 4B Session 4C Optimization with Random 9:00 AM-11:05 AM 9:00 AM-11:05 AM Inputs Room:California C - 2nd Floor Room:Plaza - 2nd Floor 11:30 AM-12:30 PM Chair: Michael Kapralov, IBM T.J. Chair: Julia Chuzhoy, Toyota Room:California B/C - 2nd Floor Watson Research Center, USA Technological Institute at Chicago, USA Chair: Piotr Indyk, Massachusetts 9:00-9:20 The Amortized Cost of 9:00-9:20 Strong Inapproximability Institute of Technology, USA Finding the Minimum Results on Balanced Rainbow- Haim Kaplan, Or Zamir, and Uri Zwick, Colorable Hypergraphs The traditional approach to NP-hard Tel Aviv University, Israel Venkatesan Guruswami and Euiwoong problems in combinatorial optimization Lee, Carnegie Mellon University, USA aims to design an algorithm with the 9:25-9:45 Approximate Range best possible approximation factor. Emptiness in Constant Time and 9:25-9:45 The Matching Polytope Many problems, however, are resistant Optimal Space Does Not Admit Fully-Polynomial Size to approximation, even though simple Mayank Goswami, Max Planck Institute Relaxation Schemes heuristics are said to be unreasonably for Informatics, Germany Gábor Braun and Sebastian Pokutta, effective in practice. To circumvent Georgia Institute of Technology, USA 9:50-10:10 Set Membership with a Few worst-case lower bounds, it is natural Bit Probes 9:50-10:10 Towards a Characterization to focus on random inputs, but how Mohit Garg and Jaikumar Radhakrishnan, of Constant-Factor Approximable Min random should they be? Classical Csps Tata Institute of Fundamental Research, average-case analysis goes all out Victor Dalmau, Universitat Pompeu- India and assumes fully random, artificially Fabra, Barcelona, Spain; Andrei 10:15-10:35 Deterministic Fully generated inputs, and then the results Krokhin, Durham University, United Dynamic Data Structures for Vertex tend to be applicable within a narrow Kingdom; Rajsekar Manokaran, KTH Cover and Matching scope only. Instead, a growing body Stockholm, Sweden Sayan Bhattacharya and Monika of work studies inputs with only a Henzinger, University of Vienna, 10:15-10:35 The Simplex Algorithm Is little bit of randomness, either on Austria; Giuseppe Italiano, University NP-Mighty only one aspect of the input or with of Rome II, Tor Vergata, Italy Yann Disser, Technische Universitaet only a little bit of random noise. I will 10:40-11:00 A Linear-Size Logarithmic Berlin, Germany; Martin Skutella, give examples from combinatorial Stretch Path-Reporting Distance Technische Universität Berlin, optimization, online and streaming Oracle for General Graphs Germany models, and discuss ways to define Michael Elkin, Ben Gurion University, 10:40-11:00 Sperner’s Colorings, “minimally random” input models. Israel; Seth Pettie, University of Hypergraph Labeling Problems and Claire Mathieu Michigan, Ann Arbor, USA Fair Division CNRS, Ecole Normale Supérieure, Maryam Mirzakhani, Stanford France and Brown University, USA University, USA; Jan Vondrak, IBM Almaden Research Center, USA

Coffee Break Lunch Break 11:05 AM-11:30 AM 12:30 PM-2:00 PM Room:California A - 2nd Floor Attendees on their own Conference Program SODA, ALENEX and ANALCO 2015 19

Monday, January 5 Monday, January 5 Monday, January 5 ALENEX: Session 2 CP13 CP14 2:00 PM-4:05 PM Session 5A Session 5B Room:Santa Fe - 2nd Floor 2:00 PM-4:05 PM 2:00 PM-4:05 PM Chair: David Eppstein, University of Room:California B - 2nd Floor Room:California C - 2nd Floor California, Irvine, USA Chair: Robert Krauthgamer, Weizmann Chair: Anupam Gupta, Carnegie 2:00-2:20 Computing MaxMin Edge Mellon University, USA Length Triangulations Institute of Science, Israel Sándor P. Fekete, Winfried Hellmann, 2:00-2:20 Online Principal Component 2:00-2:20 Approximating the Best (log n) Michael Hemmer, Arne Schmidt, and Analysis Nash Equilibrium in no -Time Julian Troegel, Technische Universität Christos Boutsidis, Rensselaer Breaks the Exponential Time Hypothesis Braunschweig, Germany Polytechnic Institute, USA; Dan Garber, Technion - Israel Institute of Mark Braverman, Young Kun Ko, and 2:25-2:45 Branch-and-Reduce Technology, Israel; Zohar Karnin and Omri Weinstein, Princeton University, Exponential/FPT Algorithms in USA Practice: A Case Study of Vertex Edo Liberty, Yahoo! Labs, USA Cover 2:25-2:45 Tighter Low-Rank 2:25-2:45 Perfect Bayesian Equilibria Takuya Akiba and Yoichi Iwata, Approximation Via Sampling the in Repeated Sales University of Tokyo, Japan Leveraged Element Nikhil R. Devanur, Yuval Peres, and Srinadh Bhojanapalli, University of Balasubramanian Sivan, Microsoft 2:50-3:10 Faster Linear-Space Research, USA Orthogonal Range Searching in Texas at Austin, USA; Prateek Jain, Arbitrary Dimensions Microsoft Research, USA; Sujay 2:50-3:10 A Stable Marriage Requires Yuzuru Okajima and Kouichi Maruyama, Sanghavi, University of Texas at Communication NEC Solution Innovators, Ltd., Japan Austin, USA Yannai Gonczarowski and Noam Nisan, Hebrew University of Jerusalem, 3:15-3:35 Region-Based 2:50-3:10 Sketching for M-Estimators: Israel; Rafail Ostrovsky and Will Approximation Algorithms for Visibility A Unified Approach to Robust Between Imprecise Locations Regression Rosenbaum, University of California, Maarten Löffler, Utrecht University, Kenneth Clarkson, IBM Corporation, Los Angeles, USA The Netherlands; Irina Kostitsyna, USA; David Woodruff, IBM Almaden 3:15-3:35 The Value 1 Problem Technische Universiteit Eindhoven, Research Center, USA Under Finite-Memory Strategies for The Netherlands; Rodrigo Silveira, 3:15-3:35 The Polyhedron-Hitting Concurrent Mean-Payoff Games Universitat Politecnica de Catalunya, Problem Krishnendu Chatterjee and Rasmus Spain; Kevin Buchin, Technische Ventsislav Chonev, Joel Ouaknine, and Ibsen-Jensen, IST, Austria Universiteit Eindhoven, The James Worrell, University of Oxford, 3:40-4:00 Robust Price of Anarchy Netherlands United Kingdom Bounds Via Lp and Fenchel Duality 3:40-4:00 Engineering Motif Search for 3:40-4:00 On Termination of Integer Janardhan Kulkarni, Duke University, Large Graphs Linear Loops USA; Vahab Mirrokni, Google Andreas Björklund, Lund University, Joel Ouaknine, Joao Sousa Pinto, and Research, USA Sweden; Petteri Kaski, Aalto James Worrell, University of Oxford, University, Finland; Lukasz Kowalik, United Kingdom University of Warsaw, Poland; Juho Lauri, Aalto University, Finland 20 SODA, ALENEX and ANALCO 2015 Conference Program

Monday, January 5 Monday, January 5 Monday, January 5 CP15 ALENEX: Session 3 CP16 Session 5C 4:30 PM-6:35 PM Session 6A 2:00 PM-4:05 PM Room:Santa Fe - 2nd Floor 4:30 PM-6:35 PM Room:Plaza - 2nd Floor Chair: Michael Goodrich, University of Room:California B - 2nd Floor California, Irvine, USA Chair: Debmalya Panigrahi, Duke Chair: Anupam Gupta, Carnegie Mellon University, USA 4:30-4:50 Experimental Evaluation of University, USA Multi-Round Matrix Multiplication on 2:00-2:20 New Approximations for MapReduce 4:30-4:50 Optimal Approximation Broadcast Scheduling Via Variants of Francesco Silvestri and Matteo for Submodular and Supermodular α-Point Rounding Ceccarello, University of Padova, Optimization with Bounded Curvature Sungjin Im, University of California, Italy Maxim Sviridenko, Yahoo! Labs, USA; Merced, USA; Maxim Sviridenko, Jan Vondrák, IBM Almaden Research Yahoo! Labs, USA 4:55-5:15 Approximating Betweenness Center, USA; Justin Ward, University Centrality in Large Evolving Networks of Warwick, United Kingdom 2:25-2:45 A Dynamic Programming Elisabetta Bergamini, Henning Framework for Non-Preemptive Meyerhenke, and Christian Staudt, 4:55-5:15 Comparing Apples Scheduling Problems on Multiple Karlsruhe Institute of Technology, and Oranges: Query Tradeoff in Machines Germany Submodular Maximization Sungjin Im, University of California, Niv Buchbinder, Tel Aviv University, Merced, USA; Shi Li, Toyota 5:20-5:40 Analysis and Experimental Israel; Moran Feldman, École Technological Institute at Chicago, Evaluation of Time-Dependent Polytechnique Fédérale de Lausanne, Distance Oracles USA; Benjamin Moseley, Washington Switzerland; Roy Schwartz, Princeton Spyros Kontogiannis and Georgia University, St. Louis, USA; Eric Torng, University, USA Michigan State University, USA Papastavrou, University of Ioannina, Greece; George Michalopoulos and 5:20-5:40 Revealing Optimal Thresholds 2:50-3:10 On (1,ε)-Restricted Andreas Paraskevopoulos, University for Generalized Secretary Problem Assignment Makespan Minimization of Patras, Greece; Dorothea Wagner, Via Continuous Lp: Impacts on Deeparnab Chakrabarty, Microsoft Online K-Item Auction and Bipartite Karlsruhe Institute of Technology, Research, India; Sanjeev Khanna, K-Matching with Random Arrival Order Germany; Christos Zaroliagis, University of Pennsylvania, USA; Shi T-H. Hubert Chan, Fei Chen, and University of Patras, Greece Li, Toyota Technological Institute at Shaofeng H.-C. Jiang, University of Chicago, USA 5:45-6:05 An I/O-Efficient Distance Hong Kong, Hong Kong Oracle for Evolving Real-World 5:45-6:05 A Simple O(log log(rank))- 3:15-3:35 A Fully Polynomial-Time Graphs Approximation Scheme for Speed Competitive Algorithm for the Matroid David Veith, Goethe University Secretary Problem Scaling with Sleep State Frankfurt, Germany; Deepak Ajwani, Antonios Antoniadis, Max Planck Moran Feldman and Ola Svensson, École Bell Labs, Ireland; Ulrich Meyer, Institute for Informatics, Germany; Polytechnique Fédérale de Lausanne, Goethe University Frankfurt, Germany Chien-Chung Huang, Chalmers Switzerland; Rico Zenklusen, ETH University of Technology, Sweden; 6:10-6:30 2-Connectivity in Directed Zürich, Switzerland Graphs: An Experimental Study Sebastian Ott, Max Planck Institute for 6:10-6:30 Online Submodular William Di Luigi, University of Informatics, Germany Maximization with Preemption Bologna, Italy; Loukas Georgiadis, 3:40-4:00 Rejecting Jobs to Minimize Niv Buchbinder, Tel Aviv University, University of Ioannina, Greece; Israel; Moran Feldman, École Load and Maximum Flow-Time Giuseppe Italiano, Universita’ di Anamitra Roychoudhury, IBM Research, Polytechnique Fédérale de Lausanne, Roma Tor Vergata, Italy; Luigi Laura, India; Syamantak Das, Naveen Garg, Switzerland; Roy Schwartz, Princeton Universita di Roma, La Sapienza, and Amit Kumar, IIT Delhi, India University, USA Italy; Nikos Parotsidis, University of Ioannina, Greece

Coffee Break 4:05 PM-4:30 PM Room:California A - 2nd Floor Conference Program SODA, ALENEX and ANALCO 2015 21

Monday, January 5 Monday, January 5 Tuesday, CP17 CP18 January 6 Session 6B Session 6C 4:30 PM-6:35 PM 4:30 PM-6:35 PM Room:California C - 2nd Floor Room:Plaza - 2nd Floor Registration Chair: David Woodruff, IBM Almaden Chair: Piotr Indyk, Massachusetts 8:00 AM-5:00 PM Research Center, USA Institute of Technology, USA Room:California Foyer - 2nd Floor 4:30-4:50 Streaming Algorithms for 4:30-4:50 Capacity of Interactive Estimating the Matching Size in Planar Communication over Erasure Channels Graphs and Beyond and Channels with Feedback Hossein Esfandiari, MohammadTaghi Ran Gelles, Princeton University, USA; Continental Breakfast Hajiaghayi, and Vahid Liaghat, Bernhard Haeupler, Carnegie Mellon 8:30 AM University of Maryland, College University, USA Room:California A - 2nd Floor Park, USA; Morteza Monemizadeh, 4:55-5:15 Limitations on Testable Affine- University of Frankfurt, Germany; Invariant Codes in the High-Rate Krzysztof Onak, IBM T.J. Watson Regime Research Center, USA Venkatesan Guruswami, Carnegie Mellon 4:55-5:15 Parameterized Streaming: University, USA; Madhu Sudan, Maximal Matching and Vertex Cover Microsoft Research New England, Rajesh Chitnis, University of Maryland, USA; Ameya Velingker, and Carol College Park, USA; Graham Wang, Carnegie Mellon University, Cormode, University of Warwick, USA United Kingdom; MohammadTaghi 5:20-5:40 Lp/sdp Hierarchy Lower Hajiaghayi, University of Maryland, Bounds for Decoding Random Ldpc College Park, USA; Morteza Codes Monemizadeh, Goethe University Badih Ghazi, Massachusetts Institute Frankfurt am Main, Germany of Technology, USA; Euiwoong Lee, 5:20-5:40 A Polylogarithmic Space Carnegie Mellon University, USA Deterministic Streaming Algorithm 5:45-6:05 On the Quickest Flow for Approximating Distance to Problem in Dynamic Networks -- A Monotonicity Parametric Min-Cost Flow Approach Timothy R. Naumovitz, Rutgers Patrick Jaillet and Maokai Lin, University, USA; Michael Saks, Massachusetts Institute of Technology, Rutgers University, USA USA 5:45-6:05 Streaming Lower Bounds for 6:10-6:30 Combinatorial Algorithm for Approximating Max-Cut Restricted Max-Min Fair Allocation Michael Kapralov, IBM T.J. Watson Chidambaram Annamalai, Christos Research Center, USA; Sanjeev Kalaitzis, and Ola Svensson, École Khanna, University of Pennsylvania, Polytechnique Fédérale de Lausanne, USA; Madhu Sudan, Microsoft Switzerland Research New England, USA 6:10-6:30 Decomposing a Graph Into Expanding Subgraphs Intermission Guy Moshkovitz and Asaf Shapira, Tel 6:35 PM-6:45 PM Aviv University, Israel SODA Business Meeting and Awards Presentation 6:45 PM-7:45 PM Room:California B - 2nd Floor Complimentary wine and beer will be served. See page 6 for award information. 22 SODA, ALENEX and ANALCO 2015 Conference Program

Tuesday, January 6 Tuesday, January 6 Tuesday, January 6 CP19 CP20 CP21 Session 7A Session 7B Session 7C 9:00 AM-11:05 AM 9:00 AM-11:05 AM 9:00 AM-11:05 AM Room:California B - 2nd Floor Room:California C - 2nd Floor Room:Plaza - 2nd Floor Chair: Debmalya Panigrahi, Duke Chair: Alexandr Andoni, Unaffiliated Chair: Piotr Indyk, Massachusetts University, USA 9:00-9:20 Approximation Schemes for Institute of Technology, USA 9:00-9:20 Online Network Design Partitioning: Convex Decomposition 9:00-9:20 FPTAS for Counting Algorithms Via Hierarchical and Surface Approximation Monotone CNF Decompositions Sayan Bandyapadhyay, Santanu Jingcheng Liu, University of California, Seeun W. Umboh, University of Bhowmick, and Kasturi Varadarajan, Berkeley, USA; Pinyan Lu, Microsoft Wisconsin, Madison, USA University of Iowa, USA Research Asia 9:25-9:45 Online Stochastic Matching 9:25-9:45 A Unified Framework for 9:25-9:45 Spatial Mixing and the with Unequal Probabilities Clustering Constrained Data Without Connective Constant: Optimal Aranyak Mehta, Google, Inc., USA; Bo Locality Property Bounds Waggoner, Harvard University, USA; Hu Ding and Jinhui Xu, State University Alistair Sinclair, University of Morteza Zadimoghaddam, Google, of New York at Buffalo, USA California, Berkeley, USA; Piyush Inc., USA 9:50-10:10 A Quasi-Ptas for the Two- Srivastava, California Institute of Technology, USA; Daniel Štefankovic, 9:50-10:10 Fast Algorithms for Online Dimensional Geometric Knapsack Stochastic Convex Programming Problem University of Rochester, USA; Yitong Shipra Agrawal and Nikhil R. Devanur, Anna Adamaszek and Andreas Wiese, Yin, Nanjing University, China Microsoft Research, USA MPII Saarbrücken, Germany 9:50-10:10 The Switch Markov Chain for Sampling Irregular Graphs 10:15-10:35 The Optimal Absolute 10:15-10:35 On the Richter- Ratio for Online Bin Packing Thomassen Conjecture About Catherine Greenhill, University of New Janos Balogh and Jozsef Bekesi, Pairwise Intersecting Closed Curves South Wales, Australia University of Szeged, Hungary; János Pach, École Polytechnique 10:15-10:35 Phase Transitions Gyorgy Dosa, University of Pannonia, Fédérale de Lausanne, Switzerland in Random Dyadic Tilings and Hungary; Jiri Sgall, Charles University, and Alfréd Rényi Institute of Rectangular Dissections Prague, Czech Republic; Rob van Stee, Mathematics, Hungary; Natan Rubin, Sarah Cannon, Sarah Miracle, and Max Planck Institute for Informatics, Ben-Gurion University, Israel; Gábor Dana Randall, Georgia Institute of Germany Tardos, Alfréd Rényi Institute of Technology, USA Mathematics, Budapest 10:40-11:00 Using Optimization to 10:40-11:00 The Speed of Evolution Break the Epsilon Barrier: A Faster and 10:40-11:00 Density and Regularity Nisheeth K. Vishnoi, École Simpler Width-Independent Algorithm Theorems for Semi-Algebraic Polytechnique Fédérale de Lausanne, for Solving Positive Linear Programs in Hypergraphs Switzerland Parallel Jacob Fox, Massachusetts Institute Zeyuan Allen-Zhu and Lorenzo Orecchia, of Technology, USA; Janos Pach, Massachusetts Institute of Technology, Courant Institute of Mathematical USA Sciences, New York University, USA; Coffee Break Andrew Suk, University of Illinois, Chicago, USA 11:05 AM-11:30 AM Room:California A - 2nd Floor Conference Program SODA, ALENEX and ANALCO 2015 23

Tuesday, January 6 Tuesday, January 6 Tuesday, January 6 IP3 CP22 CP23 New Directions in Learning Session 8A Session 8B Theory 2:00 PM-4:05 PM 2:00 PM-4:05 PM 11:30 AM-12:30 PM Room:California B - 2nd Floor Room:California C - 2nd Floor Room:California B/C - 2nd Floor Chair: Piotr Indyk, Massachusetts Chair: Alexandr Andoni, Unaffiliated Chair: Alexandr Andoni, Unaffiliated Institute of Technology, USA 2:00-2:20 Finding Four-Node The classic PAC-learning model focuses 2:00-2:20 Compatible Connectivity- Subgraphs in Triangle Time Virginia Vassilevska Williams, Joshua on the problem of learning a single Augmentation of Planar Disconnected R. Wang, Ryan Williams, and unknown function, over data drawn from Graphs Greg Aloupis, Université Libre de Huacheng Yu, Stanford University, a distribution that is unrelated to what Bruxelles, Belgium; Luis Barba, USA is being learned, and that resides on a Carleton University, Canada; Paz single machine. A number of current 2:25-2:45 Subcubic Equivalences Carmi, Ben-Gurion University of directions in machine learning differ Between Graph Centrality Problems, the Negev, Israel; Vida Dujmovic, from this setting in crucial ways. These Apsp and Diameter University of Ottawa, Canada; Fabrizio include distributed learning, multi-task Amir Abboud, Stanford University, Frati, University of Sydney, Australia; learning, semi-supervised learning, and USA; Fabrizio Grandoni, University Pat Morin, Carleton University, Canada deep/representation learning. This talk of Lugano, Switzerland; Virginia will survey some theoretical results 2:25-2:45 Geometric k Shortest Paths Vassilevska Williams, Stanford in these directions and a number of Sylvester Eriksson-Bique, Courant University, USA open algorithmic problems that these Institute, New York University, 2:50-3:10 Minors and Dimension directions bring. USA; John Hershberger, Mentor Bartosz Walczak, Georgia Insitute of Graphics Corporation, USA; Valentin Avrim Blum Technology, USA and Jagiellonian Polishchuk, Linköping University, Carnegie Mellon University, USA University, Poland Sweden; Bettina Speckmann, 3:15-3:35 Forbidden Structure Technische Universiteit Eindhoven, Characterization of Circular-Arc The Netherlands; Subhash Suri, Graphs and a Certifying Recognition University of California, Santa Algorithm Barbara, USA; Topi L. Talvitie, Mathew Francis, Indian Statistical Lunch Break University of Helsinki, Finland; Kevin Institute, India; Pavol Hell, Simon 12:30 PM-2:00 PM Verbeek and Hakan Yildiz, University Fraser University, Canada; Juraj of California, Santa Barbara, USA Stacho, Columbia University, USA Attendees on their own 2:50-3:10 Triangulation Refinement 3:40-4:00 Four Terminal Planar Delta- and Approximate Shortest Paths in Wye Reducibility Via Rooted K2,4 Weighted Regions Minors Siu-Wing Cheng, Hong Kong University Lino Demasi and Bojan Mohar, Simon of Science and Technology, Hong Fraser University, Canada Kong; Jiongxin Jin, Google, Inc., USA; Antoine Vigneron, King Abdullah University of Science & Technology (KAUST), Saudi Arabia 3:15-3:35 Optimal Detection of Intersections Between Convex Polyhedra Luis Barba, Carleton University, Canada; Stefan Langerman, Université Libre de Bruxelles, Belgium 3:40-4:00 Detecting Weakly Simple Polygons Hsien-Chih Chang, University of Illinois at Urbana-Champaign, USA; Jeff Erickson, University of Illinois, USA; Chao Xu, University of Illinois, Urbana-Champaign, USA 24 SODA, ALENEX and ANALCO 2015 Conference Program

Tuesday, January 6 Tuesday, January 6 Tuesday, January 6 CP24 CP25 CP26 Session 8C Session 9A Session 9B 2:00 PM-4:05 PM 4:30 PM-6:35 PM 4:30 PM-6:35 PM Room:Plaza - 2nd Floor Room:California B - 2nd Floor Room:California C - 2nd Floor Chair: Michael Kapralov, IBM T.J. Chair: Alexandr Andoni, Unaffiliated Chair: Piotr Indyk, Massachusetts Institute of Technology, USA Watson Research Center, USA 4:30-4:50 Property Testing on Product 2:00-2:20 An Algorithmic Framework Distributions: Optimal Testers for 4:30-4:50 Approximately Stable, for Obtaining Lower Bounds for Bounded Derivative Properties School Optimal, and Student- Random Ramsey Problems Deeparnab Chakrabarty, Microsoft Truthful Many-to-One Matchings (via Rajko Nenadov, Nemanja Skoric, Research, India; Kashyap Dixit, Differential Privacy) Sampath Kannan, University of and Angelika Steger, ETH Zürich, Pennsylvania State University, USA; Pennsylvania, USA; Jamie Switzerland Madhav Jha and C. Seshadhri, Sandia National Laboratories, USA Morgenstern, Carnegie Mellon 2:25-2:45 Robust Hamiltonicity of University, USA; Aaron Roth and Random Directed Graphs 4:55-5:15 Testing Poisson Binomial Zhiwei Steven Wu, University of Distributions Asaf Ferber, Rajko Nenadov, Andreas Pennsylvania, USA Noever, Ueli Peter, and Nemanja Jayadev Acharya and Constantinos Skoric, ETH Zürich, Switzerland Daskalakis, Massachusetts Institute of 4:55-5:15 Robust Randomized Technology, USA Matchings 2:50-3:10 Surprise Probabilities in Jannik Matuschke and Martin Skutella, Markov Chains 5:20-5:40 Testing Identity of Structured Technische Universität, Berlin, Distributions James Norris, University of Cambridge, Germany; José Soto, Universidad de Ilias Diakonikolas, University of United Kingdom; Yuval Peres, Chile, Chile Microsoft Research, USA; Alex Zhai, Edinburgh, United Kingdom; Daniel Stanford University, USA Kane, University of California, San 5:20-5:40 The Structure of the Core in Diego, USA; Vladimir Nikishkin, Assignment Markets 3:15-3:35 Characterization of Cutoff for Yash Kanoria, Columbia Business Reversible Markov Chains University of Edinburgh, United Kingdom School, USA; Daniela Saban and Jay Riddhipratim Basu and Jonathan Sethuraman, Columbia University, Hermon, University of California, 5:45-6:05 The Complexity of USA Berkeley, USA; Yuval Peres, Microsoft Estimating Rényi Entropy Research, USA Jayadev Acharya, Massachusetts 5:45-6:05 A Dynamic Model of Barter Institute of Technology, USA; Alon Exchange 3:40-4:00 Lopsidependency in the Orlitsky, Ananda Theertha Suresh, Ross Anderson, Itai Ashlagi, and David Moser-Tardos Framework: Beyond the Gamarnik, Massachusetts Institute Lopsided Lovász Local Lemma and Himanshu Tyagi, University of of Technology, USA; Yash Kanoria, David Harris, University of Maryland, California, San Diego, USA Columbia Business School, USA USA 6:10-6:30 Algorithmic Regularity for Polynomials and Applications 6:10-6:30 Bayesian Truthful Arnab Bhattacharyya, Indian Institute Mechanisms for Job Scheduling from of Science, Bangalore, India; Pooya Bi-Criterion Approximation Algorithms Constantinos Daskalakis and Matt Coffee Break Hatami, University of Chicago, Weinberg, Massachusetts Institute of 4:05 PM-4:30 PM USA; Madhur Tulsiani, Toyota Technology, USA Room:California A - 2nd Floor Technological Institute, USA Conference Program SODA, ALENEX and ANALCO 2015 25

Tuesday, January 6 CP27 Session 9C 4:30 PM-6:35 PM Room:Plaza - 2nd Floor Chair: Robert Krauthgamer, Weizmann Institute of Science, Israel 4:30-4:50 Contagious Sets in Expanders Amin Coja-Oghlan, Goethe University, Germany; Uriel Feige, Weizmann Institute of Science, Israel; Michael Krivelevich, Tel Aviv University, Israel; Daniel Reichman, Cornell University, USA 4:55-5:15 2-Edge Connectivity in Directed Graphs Loukas Georgiadis, University of Ioannina, Greece; Giuseppe Italiano, Universita’ di Roma Tor Vergata, Italy; Luigi Laura, Universita di Roma, La Sapienza, Italy; Nikos Parotsidis, University of Ioannina, Greece 5:20-5:40 Tight Bounds on Vertex Connectivity Under Vertex Sampling Keren Censor-Hillel, Technion - Israel Institute of Technology, Israel; Mohsen Ghaffari, Massachusetts Institute of Technology, USA; George Giakkoupis, INRIA, France; Bernhard Haeupler, Carnegie Mellon University, USA; Fabian Kuhn, University of Freiburg, Germany 5:45-6:05 Fast Generation of Random Spanning Trees and the Effective Resistance Metric Aleksander Madry, École Polytechnique Fédérale de Lausanne, Switzerland; Damian Straszak and Jakub Tarnawski, University of Wroclaw, Poland 6:10-6:30 Connectivity in Random Forests and Credit Networks Ashish Goel, Stanford University, USA; Sanjeev Khanna, University of Pennsylvania, USA; Sharath Raghvendra, Virginia Tech, USA; Hongyang Zhang, Stanford University, USA 26 SODA, ALENEX and ANALCO 2015 Conference Program

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. DA15 Abstracts 27

IP1 graph partitioning algorithm is scalable and can compute Pursuit on a Graph high quality partitions in time that is comparable to the running time of an efficient internal memory implementa- Pursuit games—motivated historically by military tion. tactics—are a natural for graphical settings, and take many forms. We will present some recent results involving Yaroslav Akhremtsev (among other things) drunks, Kakeya sets and a “ketchup Karlsruhe Institute of Technology graph.’ Lastly, we discuss what we think is the most [email protected] important open problem in the field. Peter Sanders Peter Winkler Karlsruhe Institue of Technology Dartmouth College Faculty of Informatics Depatment of Mathematics [email protected] [email protected] Christian Schulz IP2 Karlsruhe Institut of Technology Faculty of Informatics Optimization with Random Inputs [email protected] The traditional approach to NP-hard problems in combi- natorial optimization aims to design an algorithm with the CP0 best possible approximation factor. Many problems, how- ever, are resistant to approximation, even though simple Approximating Betweenness Centrality in Large heuristics are said to be unreasonably effective in practice. Evolving Networks To circumvent worst-case lower bounds, it is natural to fo- Betweenness centrality ranks the importance of nodes by cus on random inputs, but how random should they be? their participation in all shortest paths of the network. Classical average-case analysis goes all out and assumes Therefore computing exact betweenness values is imprac- fully random, artificially generated inputs, and then the tical in large networks. For static networks, approximation results tend to be applicable within a narrow scope only. based on randomly sampled paths has been shown to be Instead, a growing body of work studies inputs with only significantly faster in practice. However, for dynamic net- a little bit of randomness, either on only one aspect of works, no approximation algorithm for betweenness cen- the input or with only a little bit of random noise. I will trality is known that improves on static recomputation. We give examples from combinatorial optimization, online and address this deficit by proposing two incremental approxi- streaming models, and discuss ways to define ”minimally mation algorithms (for weighted and unweighted connected random” input models. graphs) which provide a provable guarantee on the absolute Claire Mathieu approximation error. Our experimental study shows that CNRS, Ecole Normale Sup´erieure, France & our algorithms are the first to make in-memory compu- Brown University tation of a betweenness ranking practical for million-edge [email protected] semi-dynamic networks. Moreover, our results show that the accuracy is even better than the theoretical guarantees in terms of absolute errors and the rank of nodes is well IP3 preserved. New Directions in Learning Theory Elisabetta Bergamini, Henning Meyerhenke The classic PAC-learning model focuses on the problem Institute of Theoretical Informatics of learning a single unknown function, over data drawn Karlsruhe Institute of Technology (KIT) from a distribution that is unrelated to what is being [email protected], learned, and that resides on a single machine. A number [email protected] of current directions in machine learning differ from this setting in crucial ways. These include distributed learn- Christian Staudt ing, multi-task learning, semi-supervised learning, and Karlsruhe Institute of Technology deep/representation learning. This talk will survey some [email protected] theoretical results in these directions and a number of open algorithmic problems that these directions bring. CP0 Avrim Blum The 1-2-3-Toolkit for Building Your Own Balls- Carnegie Mellon University into-Bins Algorithm [email protected] We examine a generic class of simple distributed balls-into- bins algorithms and compute accurate estimates of the re- CP0 maining balls and the load distribution after each round. (Semi-)External Algorithms For Graph Partition- Each algorithm is classified by (i) the load that bins ac- ing and Clustering cept in a given round and (ii) the number of messages each ball sends in a given round. Our algorithms employ In this paper, we develop semi-external and external mem- a novel ranking mechanism resulting in notable improve- ory algorithms for graph partitioning and clustering prob- ments. Simulations independently verify our results and lems. Graph partitioning and clustering are key tools for their high accuracy. processing and analyzing large complex networks. Our algorithm is able to partition and cluster huge complex Pierre Bertrand networks with billions of edges on cheap commodity ma- Ecole Normale Sup´erieure Cachan chines. Experiments demonstrate that the semi-external D´epartement de Maths 28 DA15 Abstracts

[email protected] sparse cases which allows to tradeoff round number with the amount of data shuffled in each round and the amount Christoph Lenzen of memory required by reduce functions. Then, we present MPI for Informatics an extensive study of this library on an in-house cluster [email protected] and on Amazon Web Services aiming at showing its per- formance and at comparing monolithic and multi-round approaches. The experiments show that, even without a CP0 low level optimization, it is possible to design multi-round Cuts in Increasing Trees algorithms with a small overhead.

Increasing trees have been extensively studied, since it is Francesco Silvestri, Matteo Ceccarello a simple model for many natural phenomena. Our paper University of Padova focuses on sub-families of increasing trees. We measure [email protected], [email protected] the number of connected components obtained after having re- moved the nodes whose labels are smaller than a given value. This measure of cut-length allows, for example, to CP0 analyse in average an algorithm for tree-labelling. It is no- ticeable that we give exact formulae for the distribution of Computing MaxMin Edge Length Triangulations trees according to their size and cut-lengths. Our approach is based on a construction using grafting processes. We consider the problem of computing a triangulation for Antoine Genitrini a given planar set of points with shortest edge as long as UPMC-LIP6 possible. We resolve an open problem dating back to 1991 [email protected] by showing this to be NP-hard. Making use of integer pro- gramming formulations, we compute optimal solutions for benchmark instances; applying a number of refinements, Olivier Bodini we are able to do this for instances with up to 200 points. LIPN, Universit´e Paris 13 [email protected] S´andor P. Fekete TU Braunschweig CP0 Department of Computer Science Bootstrap Percolation on Periodic Trees [email protected]

We study bootstrap percolation with the threshold param- Winfried Hellmann, Michael Hemmer, Arne Schmidt, eter θ ≥ 2 and the initial probability p on infinite periodic Julian Troegel trees that are defined as follows. Each node of a tree has TU Braunschweig degree selected from a finite predefined set of non-negative [email protected], [email protected], integers and starting from any node, all nodes at the same [email protected], [email protected] graph distance from it have the same degree. We show the existence of the critical threshold pf (θ) ∈ (0, 1) such that p>p θ with high probability, (i) if f ( ) then the periodic tree CP0 becomes fully active, while (ii) if p

[email protected] ory with numerous practical applications. In this paper we consider various notions of 2-edge and 2-vertex connectiv- ity in directed graphs (digraphs). In particular, we consider CP0 the computation of the 2-vertex- and 2-edge-connected Improved Single-Term Top-K Document Retrieval components and blocks in practice. These notions have a much richer and more complicated structure than their On natural language text collections, finding the k doc- counterparts in undirected graphs, which has only been re- uments most relevant to a query is generally solved with cently investigated. We present efficient implementations inverted indexes. On general string collections, however, of both previously proposed and new algorithms for com- more sophisticated data structures are necessary. Navarro puting the 2-vertex- and 2-edge-connected components and and Nekrich [SODA 2012] showed that a linear-space in- blocks of a digraph, and evaluate their performance experi- dex can solve such top-k queries in optimal time O(m + k), mentally on large graphs taken from a variety of application where m is the query length. Konow and Navarro [DCC areas. We conduct an extensive experimental study that 2013] implemented the scheme, managing to solve top-k sheds light on the relative difficulty of computing these no- queries within microseconds with an index using 3.34.0 tions of 2-connectivity in directed graphs in practice. bytes per character. In this paper we introduce a new implementation using significantly less space and retaining William Di Luigi similar query times. For short queries, which are the most University of Bologna, Italy difficult, our new index actually outperforms the previous [email protected] one, as well as all the other solutions in the literature. We also show that our index can be built on very large text Loukas Georgiadis collections, and that it can handle phrase queries efficiently University of Ioannina, Greece on natural language text collections. [email protected] Simon Gog Karsruhe Insitute of Technology Giuseppe Italiano [email protected] Universita di Roma “Tor Vergata’ [email protected] Gonzalo Navarro University of Chile Luigi Laura [email protected] “Sapienza’ Universita di Roma [email protected]

CP0 Nikos Parotsidis A Data-Aware Fm-Index University of Ioannina, Greece [email protected] In this paper we present some practical modifications of the higher-order entropy-compressed text indexing method of Foschini et al. [Foschini et al. 2006] based upon CP0 the Burrows-Wheeler transform and the FM-index. Our Branch-and-Reduce Exponential/FPT Algorithms method, called FM-Adaptive, retains the theoretical per- in Practice: A Case Study of Vertex Cover formance of previous work and introduces some improve- ments in practice. At the same time, broad experiments in- We investigate the gap between theory and practice for ex- dicate that our index achieves superior performance, espe- act branching algorithms. In theory, branch-and-reduce cially in terms of compression, in comparison to the state- algorithms currently have the best time complexity for of-the-art indexing techniques. The source code is available numerous important problems. On the other hand, in online. practice, state-of-the-art methods are based on different approaches, and the empirical efficiency of such theoreti- Hongwei Huo, Longgang Chen, Heng Zhao cal algorithms have seldom been investigated probably be- Xidian University cause they are seemingly inefficient because of the plethora [email protected], [email protected], of complex reduction rules. In this paper, we design a [email protected] branch-and-reduce algorithm for the vertex cover problem using the techniques developed for theoretical algorithms Jeffrey Scott Vitter and compare its practical performance with other state- The University of Kansas of-the-art empirical methods. The results indicate that [email protected] branch-and-reduce algorithms are actually quite practical and competitive with other state-of-the-art approaches for Yakov Nekrich several kinds of instances, thus showing the practical im- University of Waterloo pact of theoretical research on branching algorithms. [email protected] Takuya Akiba, Yoichi Iwata Qiang Yu The University of Tokyo Xidian University [email protected], [email protected] [email protected] CP0 CP0 Engineering Motif Search for Large Graphs 2-Connectivity in Directed Graphs: An Experi- mental Study In the graph motif problem, we are given as input a vertex- colored graph H and a multiset of colors M.Ourtaskis Graph connectivity is a fundamental concept in graph the- to decide whether H has a connected set of vertices whose 30 DA15 Abstracts

multiset of colors agrees with M. The graph motif problem query algorithms. The common preprocessing phase is is NP-complete but admits parameterized algorithms that based on a novel, quite efficient one-to-all approximation run in linear time in the size of H. We demonstrate that method for creating approximate distance summaries. (ii) algorithms based on constrained multilinear sieving are vi- We conduct an extensive experimental study with three able in practice, scaling to graphs with hundreds of millions query algorithms and six landmark sets. Our results are of edges as long as M remains small. quite encouraging, achieving remarkable speedups (by two orders of magnitude) and quite small approximation guar- Andreas Bj¨orklund antees (even PTAS), over the time-dependent variant of Lund University Dijkstra’s algorithm. [email protected] Spyros Kontogiannis Petteri Kaski University of Ioannina Aalto University [email protected] Department of Information and Computer Science & HIIT Georgia Papastavrou petteri.kaski@aalto.fi University of Ioannina, Greece [email protected] Lukasz Kowalik University of Warsaw George Michalopoulos, Andreas Paraskevopoulos Institute of Informatics University of Patras, Greece [email protected] [email protected], [email protected]

Juho Lauri Dorothea Wagner Aalto University Karlsruhe Institute of Technology Department of Information and Computer Science & Institute of Theoretical Informatics HIIT [email protected] juho.lauri@aalto.fi Christos Zaroliagis University of Patras, Greece CP0 [email protected] A Bound for the Diameter of Random Hyperbolic Graphs CP0 Random hyperbolic graphs were recently introduced as a Cache Replacement with Memory Allocation model for large networks. Their rigorous study was ini- α> / C> tiated using the following model: for 1 2, 0, We consider a variant of the generalized caching problem in set R =2lnn + C and define the graph G =(V,E)on n v ∈ V which the cache management policy must not only specify nodes as follows: For each , generate i.i.d. polar which items to evict to make room for an incoming item, coordinates (rv,θv) using the joint density function f(r, θ), θ , π r but must also specify a location in memory where each with v chosen uniformly from [0 2 )and v with density object can be placed contiguously. The problem is moti- f r α sinh(αr) ≤ r

ering the circle with random arcs. We prove the correctness Θ(nρ ), with ρ depending on G minimized when G is a of the algorithm and by the methods of complex analysis we tree. We also give an expression for ρ for grids. examine the properties of the applied estimator. We also show that the construction of the proposed algorithm is a Philippe Jacquet backbone for a simple distributed summation algorithm. Alcatel Lucent [email protected] Marcin Kardas, Miroslaw Kutylowski Wroclaw University of Technology Abram Magner Department of Computer Science Department of Computer Science [email protected], Purdue University [email protected] [email protected]

Jakub Lemiesz Wroclaw University of Technology CP0 [email protected] Faster Linear-Space Orthogonal Range Searching in Arbitrary Dimensions

CP0 We consider the problem of orthogonal range searching in Analytic Samplers and the Combinatorial Rejec- linear space for any d dimensions. The kd-tree achieves tion Method O(n(d−1)/d) query time for range counting, which has been considered to be the best complexity bound in practice, Abstract not available at time of publication. while the non-overlapping k-range achieves O(n)time J´er´emie Lumbroso in theory. In this paper, we propose a new succinct O n(d−2)/d n Universit´e Pierre et Marie Curie data structure with ( log ) query time for range [email protected] counting. In experiments, it significantly outperformed the kd-tree for high selectivity queries.

CP0 Yuzuru Okajima Region-Based Approximation Algorithms for Visi- VALWAY Technology Center, NEC Solution Innovators, bility Between Imprecise Locations Ltd. [email protected] In this paper we present new geometric algorithms for ap- proximating the visibility between two imprecise locations Kouichi Maruyama amidst a set of obstacles, where the imprecise locations are NEC Solution Innovators, Ltd. modeled by continuous probability distributions. Our tech- [email protected] niques are based on approximating distributions by a set of regions rather than on approximating by a discrete point sample. In this way we obtain guaranteed error bounds, CP0 and the results are more robust than similar results based Linear-Time Generation of Inhomogeneous Ran- on discrete point sets. We implemented our techniques and dom Directed Walks present an experimental evaluation. The experiments show that the actual error of our region-based approximation Directed two-dimensional random walks modelise diffusion, scheme converges quickly when increasing the complexity and are related to Galton–Watson trees. These random of the regions. objects can be generated in linear time, through an algo- rithm of Devroye, based on the fact that the steps form an Maarten L¨offler exchangeable sequence of random variables. We consider Utrecht University here the random generation of a more general family of m.loffl[email protected] structures, in which the transition rates, instead of being fixed, evolve in time. The steps are not exchangeable any- Irina Kostitsyna more. This describes time-dependent diffusion, and, for Technical University Eindhoven trees arising from exploration processes on finite random [email protected] graphs, accounts for effects of excluded volume. Appli- cations include rooted accessible K-maps, which provide Rodrigo Silveira the uniform generation of random minimal automata. We Universitat Polit`ecnica de Catalunya present an algorithm for this wider family of problems, for [email protected] which the achieved (bit-)complexity remains linear. Andrea Sportiello Kevin Buchin LIPN, Universit´eParisNord TU Eindhoven Villetaneuse, FRANCE [email protected] [email protected]

CP0 Fr´ed´erique Bassino Variance of Size in Regular Graph Tries LIPN, Univ. Paris Nord Villetaneuse, FRANCE Graph tries are tries built from label functions on a graph [email protected] G. Here we determine asymptotics for the variance of the size of a G-trie on a memoryless source with a uniform alphabet distribution, where G comes from a large class of CP0 infinite, M-regular DAGs. We show that the variance is An I/O-Efficient Distance Oracle for Evolving 32 DA15 Abstracts

Real-World Graphs mentally that on certain hardware the classically optimal choice of the pivot as median of a sample gets harmful be- We present an I/O-efficient distance oracle that is able cause the rollback costs of mispredicted branches become to answer online queries with a constant number of I/O. dominating. In this paper, we give the first precise an- Furthermore, we developed batched queries that have an alytical investigation of branch mispredictions in (classic) O 1 amortized I/O-complexity of ( B ) I/Os per query. On- Quicksort and Yaroslavskiy’s dual-pivot Quicksort variant line queries can be processed in milliseconds on SSDs and as implemented in Oracle’s Java 7 library. It is still not batched queries within microseconds even on HDDs. All fully understood why the latter is 10 % faster in experi- resultshavebeenachievedonrealworldgraphs.Weex- ments. For different branch prediction strategies, we give plain the experimental results and discuss improvements the expected number of branch misses caused by the afore- for the future. mentioned Quicksort variants when pivots are chosen from a sample. We conclude that the difference in branch misses David Veith is too small to explain the superiority of the dual-pivot al- Goethe University, Frankfurt gorithm. [email protected] Conrado Mart´ınez Deepak Ajwani Department of Computer Science Bell Labs Ireland Univ. Polit`ecnica de Catalunya [email protected] [email protected]

Ulrich Meyer Markus Nebel, Sebastian Wild Goethe University, Frankfurt Computer Science Department [email protected] University of Kaiserslautern [email protected], s [email protected]

CP0 Repeated Fringe Subtrees in Random Rooted Trees CP0 On k-Connectivity and Minimum Vertex Degree in A fringe subtree of a rooted tree consists of a node and Random s-Intersection Graphs all its descendants. We show that the average number of fringe subtrees that occur at least r times is of asymp- Random s-intersection graphs have recently received much totic order n/(log n)3/2 for every r ≥ 2ifann-nodetreeis interest in a wide range of application areas. Broadly taken uniformly at random from a simply generated family. speaking, a random s-intersection graph is constructed by Moreover, we also prove a strong concentration result for first assigning each vertex a set of items in some random the size of the smallest tree that does not occur as a fringe manner, and then putting an undirected edge between all subtree. pairs of vertices that share at least s items (the graph is called a random intersection graph when s =1).A Dimbinaina Ralaivaosaona, Stephan Wagner special case of particular interest is a uniform random s- Stellenbosch University, South Africa intersection graph, where each vertex independently selects [email protected], [email protected] the same number of items uniformly at random from a com- mon item pool. Another important case is a binomial ran- dom s-intersectiongraph,whereeachitemfromapoolis CP0 independently assigned to each vertex with the same proba- Lattice Paths of Slope 2/5 bility. Both models have found numerous applications thus far including cryptanalysis, and modeling recommender We analyze some enumerative and asymptotics properties systems, secure sensor networks, online social networks, of Dyck paths under a line of slope 2/5. This answers trust networks and small-world networks (uniform random to Knuth’s problem #4 from his “Flajolet lecture’ during s-intersection graphs), as well as clustering analysis, clas- the conference “Analysis of Algorithms’ (AofA’2014). Our sification, and the design of integrated circuits (binomial approach relies on the work of Banderier and Flajolet for random s-intersection graphs). In this paper, for bino- asymptotics and enumeration of directed lattice paths. A mial/uniform random s-intersection graphs, we present re- key ingredient is the “kernel method’, an old trick of Knuth sults related to k-connectivity and minimum vertex degree. himself. We also show how to obtain similar results for Specifically, we derive the asymptotically exact probabili- other slopes. ties and zero-one laws for the following three properties: (i) k-vertex-connectivity, (ii) k-edge-connectivity and (iii) Michael Wallner the property of minimum vertex degree being at least k. TU Wien [email protected] Jun Zhao, Osman Yagan, Virgil Gligor Carnegie Mellon University Cyril Banderier [email protected], [email protected], [email protected] Universit´e Paris-Nord [email protected] CP1 Approximating Independent Sets in Sparse Graphs CP0 Analysis of Branch Misses in Quicksort We consider the independent set problem on graphs with maximum degree d. Currently, an O(d log log d/ log d)ap- 2 Analyzing classic elementary operations is not sufficient proximation and a (d/ log d) UGC hardness is known. We when features of modern processors like pipelined execu- show the following: (i) The natural LP strengthened by tion and memory hierarchies influence running time. For O(log4(d)) levels of the mixed-hierarchy has integrality gap Quicksort, for example, it has been demonstrated experi- O˜(d/ log2 d). However, our proof is non-constructive and DA15 Abstracts 33

does not give a good algorithm. (ii) An O(d/ log d)ap- [email protected] proximation using polylog(d)-levels of the mixed hierarchy that runs in time nO(1) exp(logO(1) d), improving upon the previous approximation by log log d. CP1 On Survivable Set Connectivity Nikhil Bansal Eindhoven University of Technology In the Set Connectivity problem, we are given an n-node [email protected] edge-weighted undirected graph and a collection of h set pairs (Si,Ti), where Si and Ti are subsets of nodes. The goal is to compute a min-cost subgraph H so that, for each CP1 (Si,Ti), H has a path between Si and Ti. New Approximation Schemes for Unsplittable Flow We initiate the study of the Survivable Set Connectivity on a Path problem (SSC) – the generalization of Set Connectivity where we are additionally given an integer requirement We study the unsplittable flow on a path problem. Given ki >=1foreach(Si,Ti), and we want to find a min-cost is a path with capacities on its edges and a set of tasks. subgraph H that has ki edge-disjoint paths between Si The goal is to find a subset of the tasks of maximum to- and Ti. We prove the following results: tal profit such that all task demands can be routed si- 1− multaneously without violating the capacity constraints. • An approximation hardness of 2log n for SSC, for When the task densities—defined as the ratio of a task’s any >0 assuming NP does not has a quasi- profit and demand—lie in a constant range, we obtain a polynomial time algorithm. PTAS. We improve the QPTAS of Bansal et al. by re- • A bicriteria approximation algorithm for SSC that moving the assumption that the demands need to lie in a computes a solution H such that cost(H) <=opt∗ quasi-polynomial range. Our third result is a PTAS for polylog(nh) and connectivity >= ki/ log n for all the case where we are allowed to shorten the paths of the S ,T  ( i i). So, by relaxing connectivity, we obtain a tasks by at most an -fraction. Each of these results uses cost-guarantee beyond the lower bound. a sparsification lemma which could be of independent in- terest. The lemma shows that in any (optimal) solution Parinya Chalermsook there exists an O()-fraction (measured by weight) of its Max-Planck-Institut f¨ur Informatik taskswhoseremovalcreates,oneachedge,aslackwhichis [email protected] at least as large as the (1/)th largest demand using that edge. Fabrizio Grandoni Jatin Batra, Naveen Garg, Amit Kumar IDSIA, University of Lugano IIT Delhi [email protected] [email protected], [email protected], [email protected] Bundit Laekhanukit Simons Institute & I-CORE Tobias M¨omke [email protected] Saarland University [email protected] CP1 Andreas Wiese A Note on the Ring Loading Problem MPII Saarbr¨ucken The Ring Loading Problem is an optimal routing problem [email protected] arising in the planning of optical communication networks which use bidirectional SONET rings. In mathematical terms, it is an unsplittable multicommodity flow problem CP1 on undirected ring networks. We prove that any split rout- Spider Covers for Prize-Collecting Network Acti- ing solution can be turned into an unsplittable solution vation Problem while increasing the load on any edge of the ring by no 7 D D more than + 5 ,where is the maximum demand value. In the network activation problem, each edge in a graph is This improves upon a classical result of Schrijver, Seymour, associated with an activation function that decides whether and Winkler (1998) who obtained a slightly larger bound the edge is activated from weights assigned to its end nodes. 3 D 11 D of + 2 . We also present an improved lower bound of 10 The feasible solutions of the problem are node weights 101 D (previously 100 ) on the best possible bound and disprove such that the activated edges form graphs of required con- a famous and long-standing conjecture of Schrijver et al. nectivity, and the objective is to find a feasible solution in this context. minimizing its total weight. In this paper, we consider a prize-collecting version of the network activation prob- Martin Skutella lem and present the first nontrivial approximation algo- Institut f¨ur Mathematik, Sekr. MA 5, Tech Uni Berlin, rithms. Our algorithms are based on a new linear pro- Strae des 17. Juni 136, 10623 Berlin, Germany gramming relaxation of the problem. They round optimal [email protected] solutions for the relaxation by repeatedly computing node weights activating subgraphs, called spiders. For the prob- lem with element- and node- connectivity requirements, we CP2 also present a new potential function on uncrossable biset Pricing Online Decisions: Beyond Auctions families and use it to analyze our algorithms. We consider dynamic pricing schemes in online settings Takuro Fukunaga where selfish agents generate online events. Previous work National Institute of Informatics on online mechanisms has dealt almost entirely with the 34 DA15 Abstracts

goal of maximizing social welfare or revenue in an auction For valuations with complements our approximations de- settings. This paper deals with quite general settings and grade linearly with the level of complementarity. minimizing social costs. We show that appropriately com- puted posted prices allow one to achieve essentially the Michal Feldman same performance as the best online algorithm. Unlike on- Tel-Aviv University line algorithms that learn about the event, and then make [email protected] enforceable decisions, prices are posted without knowing the future events or even the current event, and are thus Nick Gravin, Brendan Lucier inherently dominant strategy incentive compatible. In par- Microsoft Research ticular we show that one can give efficient posted price [email protected], [email protected] mechanisms for metrical task systems, some instances of the k-server problem, and metrical matching problems. CP2 Ilan R. Cohen Welfare Maximization with Production Costs: A Tel Aviv University Primal Dual Approach Tel Aviv University [email protected] We study online combinatorial auctions with production costs using the online primal dual framework. For arbi- Alon Eden,AmosFiat trary (strictly convex and differentiable) production cost Tel Aviv University functions, we characterize the optimal competitive ratio [email protected], fi[email protected] achievable by online mechanisms/algorithms. We show that online posted pricing mechanisms, which are incen- Lukasz Jez tive compatible, can achieve competitive ratios arbitrarily Tel Aviv University close to the optimal, and construct lower bound instances University of Wroclaw, Institute of Computer Science on which no online algorithms, not necessarily incentive [email protected] compatible, can do better. Zhiyi Huang CP2 The University of Hong Kong On the Complexity of Computing An Equilibrium [email protected] in Combinatorial Auctions Anthony Kim We study the computation of pure Nash equilbria with Stanford University good social welfare in the game of combinatorial auctions [email protected] where each item is sold separately but simultaneously via a second-price auction. We show that when bidders’ valu- ations are submodular, in many settings computing equi- CP2 libria with good welfare is as easy as computing good al- An n-to-1 Bidder Reduction for Multi-Item Auc- locations. For subadditive valuations, computing an equi- tions and Its Applications librium requires exponential communication, and for XOS valuations, this requires techniques different from the ones We introduce a novel approach for reducing the maximal currently known. revenue k-item n-bidder auction with additive valuation to k-item 1-bidder auctions. This approach leads to a simple Shahar Dobzinski deterministic mechanism that achieves a constant fraction The Weizmann Institute of Science, Israel of the optimal revenue when items are independent. In [email protected] addition, it leads to several structural results, including a constant-ratio relationship between the optimal revenue Hu Fu, Robert Kleinberg obtainable by dominant strategy mechanisms and that ob- Cornell University tainable by Bayesian mechanisms. [email protected], [email protected] Andrew C. Yao Tsinghua University CP2 [email protected] Combinatorial Auctions Via Posted Prices

We study anonymous posted price mechanisms for Com- CP3 binatorial Auctions(CA) in a Bayesian framework. In Efficient and Robust Persistent Homology for Mea- such a mechanism item prices are posted, then the con- sures sumers approach the seller sequentially in an arbitrary or- der, each taking her favorite bundle from among the un- A new paradigm for point cloud data analysis has emerged sold items. These mechanisms are simple, transparent and recently, where point clouds are treated as empirical mea- Dominant Strategy Incentive Compatible(DSIC). For frac- sures instead of compact sets. A notion of distance to a tionally subadditive(includes submodular) agents’ prefer- measure has been defined and shown to be stable with re- ences, we find prices that obtain at least half of the op- spect to perturbations. It can easily be computed point- timal welfare in expectation. Our result is constructive: wise in the case of a point cloud, but its sublevel-sets, which given black-box access to a CA algorithm A, sample access carry the geometric information, remain hard to compute to the prior, appropriate query access to the sampled val- or approximate. This makes it challenging to adapt many uations, we compute prices that generate at least half of powerful techniques based on the Euclidean distance to a the expected welfare of A. As a corollary, we get the first point cloud to the larger setting of the distance to a mea- polytime (in n and m) constant-factor DSIC mechanism for sure on a metric space. We propose an efficient and re- Bayesian submodular CA given access to demand queries. liable scheme to approximate the topological structure of DA15 Abstracts 35

the family of sublevel-sets of the distance to a measure. containing only shape-P tiles. We obtain an algorithm for approximating the persistent homology of the distance to an empirical measure working S´andor P. Fekete in arbitrary metric spaces. Precise quality and complexity TU Braunschweig guarantees are given with a discussion on the behavior of Department of Computer Science our approach in practice. [email protected]

Micka¨el Buchet,Fr´ed´eric Chazal, Steve Oudot Jacob Hendricks INRIA Saclay Ile-de-France University of Arkansas [email protected], [email protected], University of Arkansas [email protected] [email protected]

Donald Sheehy Matthew Patitz, Trent Rogers University of Connecticut University of Arkansas [email protected] [email protected], [email protected]

Robert T. Schweller CP3 University of Texas - Pan American Minimum Forcing Sets for Miura Folding Patterns [email protected]

We introduce the study of forcing sets in mathematical CP3 origami. The origami material folds flat along mountain or valley creases. A subset F ofcreasesisforcingifthe Zigzag Persistence Via Reflections and Transposi- mountain/valley assignment can be deduced from F .We tions develop efficient algorithms for constructing a minimum forcing set of a Miura-ori map, and for deciding whether a We introduce a simple algorithm for computing zigzag per- given set of creases is forcing or not. We also provide tight sistence, designed in the same spirit as the standard per- bounds on the size of a forcing set. sistence algorithm. Our algorithm reduces a single matrix, maintains an explicit set of chains encoding the persistent homology of the current zigzag, and updates it under sim- Brad Ballinger plex insertions and removals. The total worst-case running Humboldt State University time matches the usual cubic bound. A noticeable differ- [email protected] ence with the standard persistence algorithm is that we do not insert or remove new simplices ”at the end” of the Mirela Damian zigzag, but rather ”in the middle”. To do so, we use arrow Villanova University reflections and transpositions, in the same spirit as reflec- [email protected] tion functors in quiver theory. Our analysis introduces two new kinds of reflection called the ”injective and surjective David Eppstein diamonds” and the ”transposition diamond”, for which we University of California, Irvine are able to predict the changes in the interval decomposi- Department of Computer Science tion and associated compatible bases. [email protected] Cl´ement Maria Robin Flatland INRIA Sophia Antipolis-M´editerran´ee Siena College [email protected] fl[email protected] Steve Oudot Jessica Ginepro INRIA Saclay Ile-de-France University of Connecticut [email protected] [email protected] CP3 Thomas Hull Western New England University Improved Bounds for Orthogonal Point Enclosure [email protected] Query and Point Location in Orthogonal Subdivi- sions in R3

CP3 In this paper, new results for two fundamental data struc- ture problems in the field of computational geometry are Universal Computation with Arbitrary Polyomino presented: orthogonal point enclosure query (OPEQ)in3d Tiles in Non-Cooperative Self-Assembly and point location in orthogonal subdivisions in 3d.Allthe results are in the pointer machine model of computation. We explore the power of geometry to overcome limitations In OPEQ the objective is to report those (prepreocessed) of non-cooperative self-assembly. We define a generaliza- rectangles which are stabbed by a query point. OPEQ has tion of the abstract Tile Assembly Model with tile systems been optimally solved (in terms of space and query time) consisting of collections of polyomino tiles, the Polyomino in 1d and 2d many decades back. For OPEQ in 3d,op- Tile Assembly Model, and investigate the computational timal solution is known only when rectangles are orthant. powers of polyTAM systems at temperature-1,whereat- The key result of this work is an almost optimal solution tachment among tiles occurs without glue cooperation. We for more general rectangles in 3d. For point location in prove that for any polyomino P of size ≥ 3thereexistsa orthogonal subdivisions (a special case of OPEQ where computationally universal temperature-1 polyTAM system rectangles are disjoint), a data structure is presented which 36 DA15 Abstracts

occupies O(n) space and answers the query in O(log3/2 n) Toyota Technological Institute at Chicago time, improving√ the previously best known query time by [email protected] roughly log n factor.

Saladi Rahul CP4 University of Minnesota, Twin Cities Beating Exhaustive Search for Quantified Boolean [email protected] Formulas and Connections to Circuit Complexity

We present new algorithms for the satisfiability problem for CP4 quantified Boolean formulas (QBFs), and consequences of Speeding Up the Four Russians Algorithm by faster QBF algorithms for circuit complexity. For example, About One More Logarithmic Factor we show that when the number of quantifier blocks in the QBF is either ω(log n)oro(log n/ log log n)(wheren is the We present a new combinatorial algorithm for Boolean ma- number of variables), the 2n running time of exhaustive trix multiplication that runs in O(n3(log log n)3/ log3 n) search can be beaten in a significant way. On the other time. This improves the previous combinatorial algo- hand, we prove that beating exhaustive search for QBF rithm by Bansal and Williams [FOCS’09] that runs in when the number of quantifier blocks is O(log n)would O(n3(log log n)2/ log9/4 n) time. Whereas Bansal and imply major circuit complexity lower bounds. Williams’ algorithm uses regularity lemmas for graphs, the new algorithm is simple and uses entirely elementary tech- Rahul Santhanam niques: table lookup, word operations, plus a deceptively University of Edinburgh straightforward divide-and-conquer. [email protected]

Timothy M. Chan Ryan Williams University of Waterloo Stanford University [email protected] [email protected]

CP4 CP4 Degree-3 Treewidth Sparsifiers More Applications of the Polynomial Method to Algorithm Design We study treewidth sparsifiers. Informally, given a graph G of treewidth k, a treewidth sparsifier H is a minor of In low-depth circuit complexity, the polynomial method is G, whose treewidth is close to k, |V (H)| is small, and the a way to prove lower bounds by translating weak circuits maximum vertex degree in H is bounded. In this paper we into low-degree polynomials, then analyzing their proper- describe an algorithm that, given a graph G of treewidth ties. Recently, this method found an application to algo- k H G , computes a topological minor of such that (i) the rithm design: Williams (STOC 2014)√ used it to compute H k/ k |V H | O k4 n treewidth of is Ω( polylog( )); (ii) ( ) = ( ); all-pairs shortest paths in n3/2Ω( log ) time on dense n- H and (iii) the maximum vertex degree in is 3. The running node graphs. In this paper, we extend this methodology to |V G | k time of the algorithm is polynomial in ( ) and . solve a number of problems faster than previously known methods. We first give an algorithm for Boolean Or- Chandra Chekuri thogonal Detection, which is to detect among two sets University of Illinois at Urbana-Champaign d A, B ⊆{0, 1} of size n if there is an x ∈ A and y ∈ B such [email protected] that x, y = 0. For vectors of dimension d = c(n)logn,we solve the problem in n2−1/O(log c(n)) time. We apply this as Julia Chuzhoy a subroutine in several other new algorithms: Batch Par- Toyota Technological Institute at Chicago tial Match, DNF evaluation, LCS with don’t cares, Longest [email protected] Substrings with small edit distance, Symmetric Boolean CSPs.

CP4 Amir Abboud, Ryan Williams, Huacheng Yu Improved Bounds for the Flat Wall Theorem Stanford University [email protected], [email protected], The Flat Wall Theorem of Robertson and Seymour states [email protected] that there is some function f, such that for all inte- gers w, t > 1, every graph G containing a wall of size f(w, t), contains either (i) a Kt-minor; or (ii) a small set CP5 A ⊂ V G w G \ A ( )ofvertices,andaflatwallofsize in . Short Paths on the Voronoi Graph and Closest Vec- Kawarabayashi, Thomas and Wollan recently showed a tor Problem with Preprocessing proof of this theorem with the following two sets of pa- f w, t t24 t2 w |A| O t24 n rameters: (1) ( )=Θ( ( + )) with = ( ), We give a O(2 ) expected time and space algorithm for Θ(t24 ) and (2) f(w, t)=w2 with |A|≤t − 5. In this pa- CVPP (the preprocessing version of the Closest Vector per we improve their bounds to f(w, t)=Θ(t(t + w)) with Problem). This improves on the O(4n) deterministic run- |A|≤t − 5. For the special case where the maximum time of the Micciancio Voulgaris algorithm for CVPP,based vertex degree in G is bounded by D, we show that, if G on Voronoi cell techniques, at the cost of a polynomial contains a wall of size Ω(Dt(t+w)), then either G contains amount of randomness (which only affects runtime, not a Kt-minor, or there is a flat wall of size w in G.Wecom- correctness). As a byproduct of our work, we also give an plement our result for the low-degree case by proving an optimal relationship between geometric and path distance almost matching lower bound. on the Voronoi graph.

Julia Chuzhoy Nicolas Bonifas DA15 Abstracts 37

Ecole Polytechnique & IBM itary Discrepancy of a hypergraph, defined as the maxi- [email protected] mum discrepancy of a restriction of the hypergraph to a subset of its vertices, is a measure of its complexity. Re- Daniel Dadush cent work by Nikolov, Talwar and Zhang (2013) showed New York University a polynomial time O˜(log3 n)-approximation to hereditary [email protected] discrepancy, as a by-product of their work in differential privacy. In this paper, we give a direct simple O(log3/2 n)- approximation algorithm for this problem. We show that CP5 up to this approximation factor, the hereditary discrepancy On Largest Volume Simplices and Sub- is characterized by the optimal value of a geometric con- Determinants vex program that seeks to minimize the largest ∞ norm of any point in an ellipsoid containing the columns of the We show that the problem of finding the simplex of largest incidence matrix of the hypergraph. This characterization d volume in the convex hull of n points in Q can be approxi- promises to be a useful tool in discrepancy theory. matedwithafactorofO(log d)d/2 in polynomial time. This improves upon the previously best known approximation Aleksandar Nikolov guarantee of d(d−1)/2 by Khachiyan. On the other hand, Rutgers University we show that there exists a constant c>1 such that this [email protected] problem cannot be approximated with a factor of cd, un- less P=NP. Our hardness result holds even if n = O(d), in Kunal Talwar which case there exists ac ¯d-approximation algorithm that No affiliation relies on recent sampling techniques, wherec ¯ is again a con- [email protected] stant. We show that similar results hold for the problem of finding the largest absolute value of a subdeterminant of a d × n matrix. CP5

Marco Di Summa Fast Lattice Point Enumeration with Minimal Universita degli Studi di Padova Overhead [email protected] Enumeration algorithms are arguably the best currently known methods to solve lattice problems. However, there is Friedrich Eisenbrand, Yuri Faenza, Carsten Moldenhauer an uncomfortable gap between our theoretical understand- Ecole Polytechnique Federale de Lausanne ing and practical performance of enumeration algorithms. friedrich.eisenbrand@epfl.ch, yuri.faenza@epfl.ch, We introduce a new class of lattice enumeration algorithms carsten.moldenhauer@epfl.ch that simultaneously achieve asymptotic efficiency and prac- ticality, matching or surpassing the performance of practi- CP5 cal algorithms already in moderately low dimension. Approximate Nearest Line Search in High Dimen- Daniele Micciancio sions University of California - San Diego [email protected] We consider the Approximate Nearest Line Search (NLS) problem. Given a set L of N lines in high dimensional Euclidean space Rd, the goal is to build a data structure Michael Walter that, given a query point q ∈ Rd, reports a line  ∈ L s.t. University of California, San Diego its distance to q is within (1 + ) factor of the distance of [email protected] the closest line to q. This is a natural generalization of the Approximate Nearest Neighbor problem for point sets (ANN), and is a natural first step to build nearest-neighbor CP6 data structures for objects other than points. We present Gossip vs. Markov Chains, and Randomness- a data structure that, for any fixed >0, reports the Efficient Rumor Spreading approximate nearest line in time (d+logN +1/)O(1) using 2 O(N + d)O(1/ ) space. This is the first high-dimensional We study gossip algorithms for the rumor spreading prob- data structure for this problem with poly-log query time lem which asks one node to deliver a rumor to all nodes and polynomial space. The best previous, due to Magen, in an unknown network, and every node is only allowed required quasi-polynomial space. The bounds we achieve to call one neighbor in each round. In this work we in- match the performance of the best algorithm for the ANN troduce two fundamentally new techniques in studying the problem. rumor spreading problem: (1) We establish a new connec- tion between the rumor spreading process in an arbitrary Sepideh Mahabadi graph and certain Markov chains, and show that the mixing MIT time of a certain Markov chain suffices to bound the ru- [email protected] mor spreading time in an arbitrary graph. (2) We construct a reduction from rumor spreading processes to branching programs. Based on this reduction, we show that, for any CP5 n-vertex expander graph, there is a protocol which informs Approximating Hereditary Discrepancy Via Small every node in O(log n) rounds with high probability, and Width Ellipsoids uses O(log n · log log n) random bits in total.

The Discrepancy of a hypergraph is the minimum attain- Zeyu Guo able value, over two-colorings of its vertices, of the max- Caltech imum absolute imbalance of any hyperedge. The Hered- [email protected] 38 DA15 Abstracts

He Sun [email protected] Max Planck Institute for Informatics [email protected] Peter Robinson Department of Computer Science National University of Singapore CP6 [email protected] Plurality Consensus in the Gossip Model

We study Plurality Consensus in the Gossip Model over a CP6 network of n anonymous agents. Each agent supports an Sequential Random Permutation, List Contraction initial opinion or color. The goal is to provide a proto- and Tree Contraction Are Highly Parallel col that brings the system into the configuration in which all agents support the (initial) plurality. We consider the We show that simple sequential randomized iterative al- Undecided-State Dynamics, a well-known protocol which gorithms for random permutation, list contraction, and uses just one more state than those necessary to store col- tree contraction are highly parallel. In particular, if itera- ors. We show that the speed of convergence of this proto- tions of the algorithms are run as soon as all of their de- col depends on the initial color configuration as a whole. pendencies have been resolved, the resulting computations We capture this dependence by introducing the notion of have logarithmic depth (parallel time) with high probabil- monochromaticdistancemd(¯c) which measures the dis- ity. Our proofs make an interesting connection between tance of the initial color configurationc ¯ from the closest the dependence structure of two of the problems and ran- monochromatic one. In the complete graph, we prove that dom binary trees. Building upon this analysis, we describe this dynamics converges within O(md(¯c)logn) rounds. We linear-work, polylogarithmic-depth algorithms for the three prove that this upper bound is almost tight and we adapt problems. Although asymptotically no better than the the Undecided-State Dynamics to obtain a fast, random many prior parallel algorithms for the given problems, their walk-based protocol for plurality consensus on regular ex- advantages include very simple and fast implementations, panders. and returning the same result as the sequential algorithm. Experiments on a 40-core machine show reasonably good Luca Becchetti performance relative to the sequential algorithms. Sapienza University of Rome [email protected] Guy Blelloch Computer Science Department Andrea Clementi Carnegie Mellon University University of Rome Tor Vergata [email protected] [email protected] Jeremy Fineman Emanuele Natale, Francesco Pasquale, Riccardo Silvestri Georgetown University Sapienza University of Rome jfi[email protected] [email protected], [email protected], sil- [email protected] Phillip Gibbons Intel Labs and Carnegie Mellon University [email protected] CP6 Distributed Computation of Large-Scale Graph Yan Gu, Julian Shun Problems Carnegie Mellon University [email protected], [email protected] Motivated by the increasing need for fast distributed pro- cessing of large-scale graphs such as the Web graph and various social networks, we study a number of fundamen- CP6 tal graph problems in the message-passing model, where (2Δ − 1)-Edge-Coloring Is Much Easier Than Max- we have k machines that jointly perform computation on imal Matching in the Distributed Setting an arbitrary n-node (typically, n k) input graph. The graph is assumed to be randomly partitioned among the Graph coloring is a central problem in distributed com- k ≥ 2 machines (a common implementation in many real puting. Both vertex- and edge-coloring problems have world systems). The communication is point-to-point, and been extensively studied in this context. In this paper the goal is to minimize the time complexity, i.e., the num- we show that a (2Δ − 1)-edge-coloring can be computed  n > ber of communication rounds, of solving various fundamen- in time√ smaller than log for any 0, specifically, tal graph problems. in eO( log log n) rounds. This establishes a separation be- tween the (2Δ − 1)-edge-coloring and Maximal√ Match- Hartmut Klauck ing problems, as the latter is known to require Ω( log n) Division of Mathematical Sciences time. We devise a (1 + )Δ-edge-coloring algorithm for Nanyang Technological University any constant >0. The running time of this algorithm [email protected] is O(log∗ Δ+ log n ). For Δ = (log n)1+Ω(1) the run- Δ1−o(1) ning time of our algorithm is only O(log∗ n). This con- Danupon Nanongkai stitutes a drastic improvement of the previous logarithmic Nanyang Technological University, Singapore bound. Our result for (2Δ − 1)-edge-coloring also follows [email protected] from our more general results concerning (1 − )-locally sparse graphs. Gopal Pandurangan Division of Mathematical Sciences Michael Elkin Nanyang Technological University Ben-Gurion University DA15 Abstracts 39

[email protected] Andrew Wan Simons Institute, UC Berkeley Seth Pettie [email protected] University of Michigan, Ann Arbor [email protected] Karl Wimmer Duquesne University Hsin-Hao Su [email protected] University of Michigan EECS Department [email protected] CP7 Robust Probabilistic Inference

CP7 Robust probabilistic inference is an extension of probabilis- Learning from Satisfying Assignments tic inference, where some of the observations are adversar- ially corrupted. We model it as a zero-sum game between This paper studies the problem of learning “low- the adversary, who can select a modification rule, and the complexity” probability distributions over the Boolean hy- predictor, who wants to accurately predict the state of na- percube {−1, 1}n. In our model the learning algorithm ture. Given a black-box access to a Bayesian inference in is given uniform random satisfying assignments of an un- the classic (adversary-free) setting, our near optimal policy known f ∈Cand its goal is to output a high-accuracy ap- runs in polynomial time in the number of observations and proximation of the uniform distribution over f −1(1). We the number of possible modification rules. show that the two most widely studied classes of Boolean functions in learning theory — LTFs and DNFs — have Yishay Mansour efficient distribution learning algorithms in our model. Tel Aviv University Microsoft Research, Hertzelia Ilias Diakonikolas [email protected] University of Edinburgh [email protected] Aviad Rubinstein UC Berkeley Anindya De [email protected] DIMACS/IAS [email protected] Moshe Tennenholtz Microsoft Research Rocco A. Servedio Technion - Israel Institute of Technology Columbia University [email protected] [email protected] CP7 CP7 Learning Privately with Labeled and Unlabeled Approximate Resilience, Monotonicity, and the Examples Complexity of Agnostic Learning Inspired by the (non-private) models of semi-supervised A function f is d-resilient if all its Fourier coefficients of and active learning, where the focus is on the sample degree at most d are zero, i.e. f is uncorrelated with all complexity of labeled examples, we consider private semi- low-degree parities. We study the notion of approximate supervised learners that operate on a random sample where resilience of Boolean functions, where we say that f is α- only a (hopefully small) portion of this sample is labeled. approximately d-resilient if f is α-closetoa[−1, 1]-valued The learners have no control over which of the sample el- ements are labeled. Our main result is that the labeled d-resilient function in 1 distance. We show that approxi- mate resilience essentially characterizes the complexity of sample complexity of private learners is characterized by agnostic learning of a concept class C over the uniform the VC dimension. distribution. Focusing on monotone Boolean functions, α Amos Beimel, Kobbi Nissim, Uri Stemmer we exhibit√ the existence of near-optimal -approximately  α n α> Ben Gurion University Ω( )-resilient monotone functions for all 0. Our [email protected], [email protected], stem- constructions are based on general resilience analysis and [email protected] amplification techniques we introduce. These structural results, together with the characterization, imply nearly optimal lower bounds for agnostic learning of monotone CP7 juntas. Linear-Programming Based Approximation Algo- Dana Dachman-Soled rithms for Multi-Vehicle Minimum Latency Prob- University of Maryland lems [email protected] an We consider various multi-vehicle versions of the minimum latency problem. There is a fleet of k vehicles located at one Vitaly Feldman or more depot nodes, and we seek a collection of routes for IBM Research - Almaden these vehicles that visit all nodes so as to minimize the to- [email protected] tal latency incurred, which is the sum of the client waiting times. We obtain an 8.497-approximation for the version Li-Yang Tan where vehicles may be located at multiple depots and a Columbia University 7.183-approximation for the version where all vehicles are [email protected] located at the same depot, both of which are the first im- 40 DA15 Abstracts

provements on this problem in a decade. Perhaps more University of Warsaw significantly, our algorithms exploit various LP-relaxations International Institute of Molecular and Cell Biology for minimum-latency problems. We show how to effectively [email protected] leverage two classes of LPs—configuration LPs and bidi- rected LP-relaxations—thatareoftenbelievedtobequite powerful but have only sporadically been effectively lever- CP8 aged for network-design and vehicle-routing problems. Sharp Bounds on Formation-Free Sequences

Ian Post, Chaitanya Swamy An (r, s)-formation is the concatenation of s permutations University of Waterloo over an r-letter alphabet. Formation-free sequences are [email protected], [email protected] a generalization of standard Davenport-Schinzel sequences (where r = 2) and can be used to obtain good bounds on the extremal function of any forbidden subsequence. CP8 More recently, formation-free sequences have been applied A New Characterization of Maximal Repetitions to bounding the size of sets of permutations with fixed VC- by Lyndon Trees dimension. Improving on earlier work by Klazar, Nivasch, and Cibulka and Kynˇcl, we prove sharp bounds on the We give a new characterization of maximal repetitions (or length of (r, s)-formation-free sequences, for every r and s, runs) in strings, using a tree defined on recursive standard and on sequences avoiding “doubled’ (r, s)-formations. factorizations of Lyndon words, called the Lyndon tree. This leads to a very simple proof of ρ(n) < 1.5n,where Seth Pettie ρ(n) is the maximum number of runs in a string of length University of Michigan n. We also present a new, conceptually simple linear-time [email protected] algorithm for computing all the runs in a string, not relying on the Lempel-Ziv factorization. CP8 Hideo Bannai Cell-Probe Bounds for Online Edit Distance and Department of Informatics Other Pattern Matching Problems Kyushu University [email protected] We give new cell-probe lower bounds for Hamming dis- tance and convolution in a stream. In contrast to previ- Tomohiro I ous results, these bounds are non-trivial even for single bit Department of Computer Science cells. Using these new ideas, we then show the first known TU Dortmund time lower bounds for both online edit distance and LCS. [email protected] Finally, for the online edit distance problem we give a cell- probe algorithm which is exponentially faster than the best Shunsuke Inenaga, Yuto Nakashima, Masayuki Takeda, known RAM algorithm. Kazuya Tsuruta Department of Informatics Raphael Clifford, Markus Jalsenius, Benjamin G. Sach Kyushu University University of Bristol, UK [email protected], cliff[email protected], [email protected], [email protected], [email protected] [email protected], CP8 CP8 Wavelet Trees Meet Suffix Trees Internal Pattern Matching Queries in a Text and Applications We start by presenting a faster construction algorithm for the famous wavelet trees — a data structure with ”myr- We develop a data structure for internal pattern match- iad virtues”. We then combine wavelet and suffix trees in ing in a text, which finds in constant time the occur- a clever way to develop a new data structure, a wavelet rences of one subword x in another subword y provided suffix tree, which operates on substrings of a string rather that |y| = O(|x|). The data structure has linear size and than single characters. Sample applications of wavelet suf- admits a linear-time construction algorithm. We then use fix trees are: substring suffix rank, substring suffix selec- it to answer queries about primitivity and periods of sub- tion, and computation of a run-length-encoded Burrows- words, general substring compression, and cyclic equiva- Wheeler transform of a substring. lence of subwords. Efficient construction allows to improve the algorithm finding δ-subrepetitions (a generalized ver- Maxim Babenko sion of runs). Recently, our data structure has been applied National Research University Higher School of Economics for further queries about subwords of a given text. [email protected]

Tomasz Kociumaka, Jakub Radoszewski Pawel Gawrychowski University of Warsaw Max-Planck-Institut f [email protected], [email protected] [email protected]

Wojciech Rytter Tomasz Kociumaka University of Warsaw University of Warsaw Nicolaus Copernicus University in Toru´n [email protected] [email protected] Tatiana Starikovskaya Tomasz Wale´n National Research University Higher School of Economics DA15 Abstracts 41

[email protected] [email protected]

Daniel Lokshtanov CP9 University of Bergen Characterizing the Easy-to-Find Subgraphs from [email protected] the Viewpoint of Polynomial-Time Algorithms, Kernels, and Turing Kernels Neeldhara Misra Indian Institute of Science We study two fundamental problems related to finding sub- [email protected] graphs: given graphs G and H, test whether H is isomor- phic to a subgraph of G, or determine the maximum num- ber of vertex-disjoint H-subgraphs that can be packed in Ramanujan M. Sridharan G. We investigate these problems when the graph H be- University of Bergen longs to a fixed hereditary family F. Our goal is to study [email protected] which classes F make the two problems tractable in one of the following senses: (a) (randomized) polynomial-time Saket Saurabh solvable, (b) admits a polynomial (many-one) kernel, or (c) IMS admits a polynomial Turing kernel. [email protected]

Bart Jansen Eindhoven University of Technology CP9 [email protected] An Exact Characterization of Tractable Demand Daniel Marx Patterns for Maximum Disjoint Path Problems Computer and Automation Research Institute, Hungarian Academy of Sciences (MTA SZTAKI) We study the following general disjoint paths problem: [email protected] given a supply graph G,asetT ⊆ V (G)ofterminals, ademandgraphH on the vertices T , and an integer k, the task is to find a set of k pairwise vertex-disjoint valid CP9 paths, where we say that a path of the supply graph G is The Parameterized Complexity of K-Biclique valid if its endpoints are in T and adjacent in the demand graph H.ForaclassH of graphs, we denote by Max- Given a graph G and a parameter k,thek-Biclique prob- imum Disjoint H-Paths G the restriction of this problem lem asks whether contains a complete bipartite subgraph when the demand graph H is assumed to be a member of Kk,k.Weprovethatk-Biclique is W [1]-hard by giving H k k . We study the fixed-parameter tractability of this fam- an fpt-reduction from -Clique to -Biclique, thus solving ily of problems, parameterized by k. Our main result is a this longstanding open problem. We also provide a prob- Maximum k complete characterization of the FPT cases of abilistic√ reduction, which indicates that -Biclique has no Disjoint H-Paths H o k for every hereditary class of graphs: f(k) ·n ( )-time algorithm for any computable function f it turns out that complexity depends on the existence of unless a randomized version of Exponential Time Hypoth- large induced matchings and large induced skew bicliques esis fails. Our result implies the dichotomy classification of in the demand graph H. the parameterized complexity of cardinality constrain sat- isfaction problem and the inapproximability of maximum D´aniel Marx k-intersection problem. Hungarian Academy of Sciences (MTA SZTAKI), Bingkai Lin Budapest,Hungary University of Tokyo [email protected] [email protected] Paul Wollan University of Rome ”La Sapienza” CP9 [email protected] Solving D-Sat Via Backdoors to Small Treewidth φ t For a CNF formula and integer ,astrong backdoor set CP9 to treewidth t is a set of variables such that each possi- ble partial assignment τ to this set reduces φ toaformula Tight Lower Bound for the Channel Assignment whose incidence graph is of treewidth at most t.Aweak Problem backdoor set to treewidth t is a set of variables such that φ there is a partial assignment to this set that reduces to a We study the complexity of the Channel Assignment prob- satisfiable formula of treewidth at most t. Our main con- d φ lem. An open problem asks whether Channel Assignment tribution is an algorithm that, given a -CNF formula admits an O(cn)-time algorithm, for a constant c indepen- k O(k)|φ| and an integer ,intime2 , either finds a satisfying dent of the weights on the edges. We answer this question φ φ assignment of , or reports correctly that is not satisfi- in the negative i.e. we show that there is no 2o(n log n)-time able, or concludes correctly that φ has no weak or strong t k algorithm solving Channel Assignment unless the Expo- backdoor set to treewidth of size at most . Prior to our nential Time Hypothesis fails. Note that the currently best work, such results were know only for the very special case ∗ O n n known algorithm works in time O (n!) = 2 ( log ) so our of t = 1 (Gaspers and Szeider, ICALP 2012). lower bound is tight. Fedor Fomin Dep. of Informatics Arkadiusz Socala University of Bergen University of Warsaw 42 DA15 Abstracts

[email protected] [email protected]

CP10 CP10 On Uniform Capacitated K-Median Beyond the Bi-Factor Approximation Algorithms for Hard Ca- Natural Lp Relaxation pacitated K-Median Problems k In this paper, we study the uniform capacitated k-median In the classical -median problem the goal is to select a k problem. Our result is an exp(O(1/2))-approximation al- subset of at most facilities in order to minimize the total gorithm for the problem that violates the cardinality con- cost of opened facilities and established connections be- straint by a factor of 1 + , by considering a novel LP tween clients and opened facilities. We consider the capac- relaxation for the problem. itated version of the problem, where a single facility may only serve a limited number of clients. It is known that vi- Shi Li olating capacities by a factor of 2 + ε is sufficient to obtain Toyota Technological Institute at Chicago constant factor approximation of the connection cost. We [email protected] substantially extend this result in the following two direc- tions. First, we extend the 2 + ε capacity violating algo- rithm to the more general k-facility location problem with CP10 uniform capacities, where opening facilities incurs a loca- tion specific opening cost. Second, we obtain a constant Dynamic Facility Location via Exponential Clocks factor approximation algorithm for the k-median problem also in the case of non-uniform capacities with a slightly We present a new LP-rounding algorithm for facility lo- bigger violation factor of 3 + ε. cation problems, which yields the first constant approxi- mation algorithm for the dynamic facility location prob- Jaroslaw Byrka lem. This problem is a generalization of the classic fa- University of Wroclaw cility location problem, proposed by Eisenstat, Math- [email protected] ieu, and Schabanel to model the dynamics of evolving social/infrastructure networks. Our algorithm exhibits Krzysztof Fleszar several properties distinguishing it from previous LP- University of W¨urzburg roundings for facility location problems, which are crucial krzysztof.fl[email protected] for its application to the dynamic problem.

Hyung-Chan An, Ashkan Norouzi-Fard, Ola Svensson Bartosz Rybicki EPFL University of Wroclaw hyung-chan.an@epfl.ch, ashkan.norouzifard@epfl.ch, [email protected] ola.svensson@epfl.ch Joachim Spoerhase University of W¨urzburg CP10 [email protected] An Improved Approximation for k-Median, and Positive Correlation in Budgeted Optimization CP10 k-median is a classic problem in combinatorial optimiza- Improved Region-Growing and Combinatorial Al- k tion. We improve upon the best known approximation ra- gorithms for -Route Cut Problems .  .  tio of Li and Svensson from 2 732 + to 2 611 + by im- k proving two separate stages of their algorithm. We do this We study the -route generalizations of various cut prob- lems, the most general of which is k-route multicut, by exploiting the structure of their worst case instance. We r also show a general dependent rounding technique which wherein we have source-sink pairs and the goal is to delete a minimum-cost set of edges to reduce the edge- bounds positive as well as negative correlation, and allows k us to decrease the run-time dependence on . connectivity of every pair to below . We present vari- ous approximation and hardness results that improve the Jaroslaw Byrka state-of-the-art for these problems in several cases. Our University of Wroclaw algorithms are based on combinatorial techniques and on [email protected] a new powerful region-growing approach. Guru Guruganesh Thomas Pensyl Carnegie Mellon University University of Maryland, College Park [email protected] [email protected] Laura Sanita, Chaitanya Swamy Bartosz Rybicki University of Waterloo University of Wroclaw [email protected], [email protected] [email protected]

Aravind Srinivasan CP11 University of Maryland A Linear-Size Logarithmic Stretch Path-Reporting [email protected] Distance Oracle for General Graphs

Khoa Trinh For any n-vertex undirected graph G =(V,E)and University of Maryland, College Park k =1, 2,..., Thorup-Zwick’s distance oracle has size DA15 Abstracts 43

O(kn1+1/k ). It produces paths with stretch at most 2k −1, per update. and its query time is, as was recently shown by Chechik, O(1). A major drawback of the Thorup-Zwick oracle is Sayan Bhattacharya that its size is Ω(n · log n). Mendel and Naor devised an University of Vienna oracle with space O(n1+1/k)andstretchO(k), but their or- Vienna, Austria acle is not path-reporting. In this paper we devise a path- [email protected] reporting distance oracle with size O(n1+1/k), stretch O(k) and query time O(n), for any constant >0. Monika Henzinger University of Vienna Michael Elkin [email protected] Ben-Gurion University [email protected] Giuseppe Italiano University of Rome, Tor Vergata Seth Pettie Rome, Italy University of Michigan, Ann Arbor [email protected] [email protected] CP11 CP11 The Amortized Cost of Finding the Minimum Set Membership with a Few Bit Probes We obtain an essentially optimal tradeoff between the amortized cost of the three basic priority queue operations We consider the bit-probe complexity of the set member- insert, delete and find-min in the comparison model. More ship problem, where a set S of size at most n from a uni- specifically, we show that verse of size m is to be represented as a short bit vector in   order to answer membership queries of the form “Is x in n A(find-min)=Ω , S?” by adaptively probing the bit vector at t places. Let (2 + )A(insert)+A(delete) m n t s( , , ) be the minimum number of bits of storage needed  n  for such a scheme. Improving the results of Alon and Feige A find min O n , ( - )= A insert A delete +log (2009), we obtain better bounds for s(m,n,2) and s(m,n,3). 2 ( )+ ( ) for any fixed >0, where n is the number of items in the priority queue and A(insert), A(delete)andA(find-min) Mohit Garg, Jaikumar Radhakrishnan are the amortized costs of the insert, delete and find-min Tata Institute of Fundamental Research operations, respectively. Our lower bound holds even if [email protected], [email protected] randomization is allowed. Surprisingly, such fundamen- tal bounds on the amortized cost of the operations were not known before. Brodal, Chaudhuri and Radhakrishnan, CP11 obtained similar bounds for the worst-case complexity of find-min Approximate Range Emptiness in Constant Time and Optimal Space Haim Kaplan, Or Zamir,UriZwick Tel Aviv University We study a generalization of Bloom Filters. The ARE [email protected], [email protected], [email protected] problem asks to represent elements from a universe in or- der to answer emptiness queries on a range, with a prob- ability of false positives allowed. We provide space-error CP12 tradeoffs, and then show how one can achieve constant The Simplex Algorithm Is NP-Mighty query time while matching the space lower bound up to a lower order additive term. This result is achieved through We propose to classify the power of algorithms by the com- a novel succinct data structure for (exact 1d) range empti- plexity of the problems that they can be used to solve. ness/reporting. Instead of restricting to the problem an algorithm was de- signed to solve, we include problems that are implicitly Mayank Goswami during the algorithm’s execution. We show that the (Net- Max-Planck Institute for Informati work) Simplex Method and the Successive Shortest Path Algorithms and Complexity. Algorithm can be used to solve any problem in NP. As a [email protected] consequence, we obtain several novel hardness results. Yann Disser, Martin Skutella CP11 TU Berlin [email protected], [email protected] Deterministic Fully Dynamic Data Structures for berlin.de Vertex Cover and Matching

We present the first deterministic data structures for main- CP12 taining approximate minimum vertex cover and maximum √ Towards a Characterization of Constant-Factor Ap- matching in a fully dynamic graph in o( m) time per up- proximable Min Csps date. In particular, for minimum vertex cover we provide  deterministic data structures for maintaining a (2 + )ap- We study the approximability of Minimum Constraint Sat- O n/2 proximation in (log ) amortized time per update. For isfaction Problems, where the goal is to minimize the num-  maximum matching, we show how to maintain a (3 + ) ber of unsatisfied constraints in a given CSP instance. The 1/3 2 approximation in O(m / ) amortized time per update, basic LP relaxation is known to be optimal for constant- and a (4+) approximation in O(m1/3/2) worst-case time factor approximating Min CSP unless the Unique Games 44 DA15 Abstracts

Conjecture fails. Using the algebraic approach to the CSP, lems and Fair Division we show that a new natural algebraic condition is neces- sary and sufficient for the finiteness of the integrality gap We prove three results about colorings of the simplex rem- of this LP. iniscent of Sperner’s Lemma, with applications in hard- ness of approximation and fair division. First, we prove Victor Dalmau a coloring lemma which implies an optimal Unique-Games University Pompeu Fabra, Barcelona hardness of (k − 1 − )-approximation for the Hypergraph [email protected] Labeling problem with label set [k]. We also show that a (k − 1)-approximation can be achieved. Second, we show Andrei Krokhin a Sperner-admissible labeling of the simplex with an ap- Durham University plication in the context of fair division. Third, we prove [email protected] that there are subdivisions of the simplex with a fractional labeling such that every hyperedge in the subdivision uses only labelings with 1 or 2 colors. This means that a natural Rajsekar Manokaran LP cannot distinguish instances of Hypergraph Labeling KTH Stockholm that can be labeled so that every hyperedge uses at most 2 [email protected] colors, and instances that must have a rainbow hyperedge. We prove that this problem is indeed NP-hard for k =3.

CP12 Maryam Mirzakhani Strong Inapproximability Results on Balanced Stanford University Rainbow-Colorable Hypergraphs [email protected]

Consider a K-uniform hypergraph H =(V,E). A coloring Jan Vondrak c : V →{1, 2,...,k} with k colors is rainbow if every hyper- IBM Almaden Research Center edge e contains at least one vertex from each color, and is [email protected] called perfectly balanced when each color appears the same number of times. A simple polynomial-time algorithm finds a 2-coloring if H admits a perfectly balanced rainbow k- CP13 coloring. For a hypergraph that admits an almost balanced Tighter Low-Rank Approximation Via Sampling rainbow coloring, we prove that it is NP-hard to find an the Leveraged Element independent set of size , for any >0. Consequently, we cannot weakly color (avoiding monochromatic hyperedges) We propose a new randomized algorithm for computing a it with O(1) colors. With k = 2, it implies strong hard- low-rank approximation to a given matrix. Taking an ap- ness for discrepancy minimization of systems of bounded proach different from existing literature, our method first set-size. We give a recipe for converting a promising test involves a specific biased sampling, with an element being distribution and a suitable choice of a outer PCP to a hard- chosen based on the leverage scores of its row and column, ness result, and use this recipe to prove additional results and then involves weighted alternating minimization over almost in a black-box manner. the factored form, to minimize error only on these sam- ples. Our method can leverage input sparsity, yet produce Venkatesan Guruswami, Euiwoong Lee approximations in spectral norm. Carnegie Mellon University [email protected], [email protected] Srinadh Bhojanapalli UT Austin [email protected] CP12 The Matching Polytope Does Not Admit Fully- Prateek Jain Polynomial Size Relaxation Schemes Microsoft Research [email protected] Rothvo established that every linear program expressing the matching polytope has exponential many inequalities. Sujay Sanghavi We generalize this to sharp bounds on the polyhedral inap- Electrical and Computer Enigneering proximability of the matching polytope: for 0 ¡ e ¡ 1, every University of Texas at Austin polyhedral (1 + e/n)-approximation requires exponential [email protected] many inequalities, where n is the number of vertices. Thus matching is the first problem in P, whose natural linear encoding does not admit a fully polynomial-size relaxation CP13 scheme (the polyhedral version of FPTAS). Online Principal Component Analysis

G´abor Braun We consider the online version of the Principal Component x ,...,x ∈ Rd×n ISYE Georgia Intitute of Technology Analysis problem: given a set of vectors 1 n and a target dimension k

Christos Boutsidis CP12 Yahoo! Labs Sperner’s Colorings, Hypergraph Labeling Prob- [email protected] DA15 Abstracts 45

Dan Garber James Worrell Technion Department of Computer Science [email protected] Oxford University [email protected] Zohar Karnin Yahoo Labs [email protected] CP13 karnin Sketching for M-Estimators: A Unified Approach to Robust Regression Edo Liberty We give algorithms for M-estimators minx Ax − bG, Yahoo Labs n×d n n [email protected] where A ∈ R , b ∈ R ,andyG for y ∈ R is speci- fied by a non-negative function G with yG ≡ i G(yi). We give a (1 + )-approximation to the Huber measure CP13 in O(nnz(A)logn+poly(d(log n)/)) time, where nnz(A)is The Polyhedron-Hitting Problem the number of non-zero entries of A.WethensolvetheM- estimation problem in nnz(A)+poly(d) time for any convex We consider polyhedral versions of Kannan and Lipton’s G with at most quadratic growth, finding a constant factor Orbit Problem (STOC ’80 and JACM ’86)—determining approximation with high probability. whether a target polyhedron V may be reached from a starting point x under repeated applications of a linear Kenneth Clarkson transformation A in an ambient vector space Qm.Inthe IBM context of program verification, very similar reachability [email protected] questions were also considered and left open by Lee and Yannakakis in (STOC ’92). We present what amounts to a David Woodruff complete characterisation of the decidability landscape for IBM Almaden Research Center, USA the Polyhedron-Hitting Problem, expressed as a function [email protected] of the dimension m of the ambient space, together with the dimension of the polyhedral target V : more precisely, for each pair of dimensions, we either establish decidability, CP14 or show hardness for longstanding number-theoretic open A Stable Marriage Requires Communication problems. The Gale-Shapley algorithm for the Stable Marriage Prob- Ventsislav Chonev lem is known to take Θ(n2) steps to find a stable marriage Department of Computer Science in the worst case (with n women and n men). Using a re- University of Oxford duction to the communication complexity of the disjoint- [email protected] ness problem, we show that Ω(n2) Boolean queries of any type are indeed required to find a stable marriage. This Joel Ouaknine, James Worrell lower bound also generalizes to randomized algorithms, Department of Computer Science finding approximately stable marriages, and several related Oxford University problems. [email protected], [email protected] Yannai Gonczarowski Departments of Mathematics and Computer Science CP13 The Hebrew University of Jerusalem On Termination of Integer Linear Loops [email protected]

A fundamental problem in program verification concerns Noam Nisan the termination of simple linear loops of the form: Hebrew University of Jerusalem [email protected] x ← u; while Bx ≥ c do x ← Ax + a where x is a vector of variables, u, a,andc are integer Rafail Ostrovsky vectors, and A and B are integer matrices. Assuming the UCLA matrix A is diagonalisable, we give a decision procedure [email protected] for the problem of whether, for all initial integer vectors u, such a loop terminates. The correctness of our algorithm Will Rosenbaum relies on sophisticated tools from algebraic and analytic Department of Mathematics number theory, Diophantine geometry, and real algebraic University of California, Los Angeles geometry. To the best of our knowledge, this is the first [email protected] substantial advance on a 10-year-old open problem of Ti- wari (2004) and Braverman (2006). CP14 Joel Ouaknine The Value 1 Problem Under Finite-Memory Department of Computer Science Strategies for Concurrent Mean-Payoff Games Oxford University [email protected] The talk will be about two-player, zero-sum concurrent mean-payoff games with payoffs in [0,1]. We present an Joao Sousa Pinto algorithm for finding the set of states, such that for play University of Oxford starting from any state in that set, for each >0, player 1 [email protected] has a finite memory strategy ensuring mean-payoff 1 − , 46 DA15 Abstracts

against all strategies for player 2. We also argue that in and remains the same throughout. One key aspect of this such states player 1 has a strategy that does not use mem- game is revelation of the buyer’s type through his actions: ory. while the seller might try to learn this value to extract more revenue, the buyer is motivated to hide it to induce lower Krishnendu Chatterjee, Rasmus Ibsen-Jensen prices. If the seller is able to commit to future prices, it IST Austria is known that the best he can do is extract the Myerson [email protected], rasmusibsen- optimal revenue each day. In a more realistic scenario, the [email protected] seller is unable to commit and must play a perfect Bayesian equilibrium. We explore this setting without commitment and find several surprises. CP14 Approximating the Best Nash Equilibrium in Nikhil R. Devanur, Yuval Peres no(log N)-Time Breaks the Exponential Time Hy- Microsoft Research, Redmond pothesis [email protected], [email protected]  We study the computational complexity of finding an - Balasubramanian Sivan approximate Nash equilibrium with good social welfare. University of Wisconsin-Madison Hazan and Krauthgamer and subsequent improvements [email protected] showed that finding an -approximate Nash equilibrium with good social welfare in a two player game and many variants of this problem is at least as hard as finding CP15 a planted clique of size O(log n) in the random graph A Dynamic Programming Framework for Non- G(n, 1/2). We show that any polynomial time algorithm Preemptive Scheduling Problems on Multiple Ma- that finds an -approximate Nash equilibrium with good chines social welfare refutes (the worst-case) Exponential Time Hypothesis by Impagliazzo and Paturi, confirming the re- In this paper, we consider a variety of scheduling prob- cent conjecture by Aaronson, Impagliazzo and Moshkovitz. lems where n jobs with release times are to be scheduled O˜ n1/2 Specifically, it would imply a 2 ( ) algorithm for SAT. non-preemptively on a set of m identical machines. The Our lower bound matches the quasi-polynomial time al- problems considered are machine minimization, (weighted) gorithm by Lipton, Markakis and Mehta for solving the throughput maximization and min-sum objectives such as problem. (weighted) flow time and (weighted) tardiness. We de- velop a novel quasi-polynomial time dynamic programming Mark Braverman, Young Kun Ko,OmriWeinstein framework that gives O(1)-speed O(1)-approximation algo- Princeton University rithms for these problems. [email protected], [email protected], owe- [email protected] Sungjin Im University of California, Merced [email protected] CP14 Robust Price of Anarchy Bounds Via Lp and Shi Li Fenchel Duality Toyota Technological Institute at Chicago [email protected] In this paper, we develop a new framework based on LP and Fenchel duality for bounding the robust price of anarchy for a large class of games. We use our framework to give to Benjamin Moseley the first PoA bounds for temporal routing games on graphs Washington University in St. Louis and energy minimization games in machine scheduling. We [email protected] demonstrate the wide applicability of our framework by giving new proofs of the PoA bounds for several classical Eric Torng games, such as congestion games. Michigan State University [email protected] Janardhan Kulkarni Deparment of Computer Science, Duke University Durham, NC, 27708. CP15 [email protected] New Approximations for Broadcast Scheduling Via Variants of α-Point Rounding Vahab Mirrokni Google Research In the pull-based broadcast scheduling, there are n unit- [email protected] sized pages of information available at the server. When the server transmits a page, all outstanding requests for the page are simultaneously satisfied. In this paper, we CP14 give new approximation results for two popular objectives, Perfect Bayesian Equilibria in Repeated Sales minimizing the maximum flow time and maximizing the throughput. In particular, we obtain the first PTAS for A special case of Myerson’s classic result describes the the former objective. Our key techniques for these im- revenue-optimal equilibrium when a seller offers a single provements are novel variants of α-point rounding that can item to a buyer. We study a natural repeated sales exten- effectively reduce congestion in schedule. sion of this model: a seller offers to sell a single fresh copy of an item to the same buyer every day via a posted price. Sungjin Im The buyer’s value for the item is unknown to the seller University of California, Merced but is drawn initially from a publicly known distribution F [email protected] DA15 Abstracts 47

Maxim Sviridenko In this paper we propose a “rejection model” which requires Yahoo! Labs no resource augmentation but which permits the online al- [email protected] gorithm to not serve an epsilon-fraction of the requests. We consider the problems of minimizing the maximum load on any machine and minimizing (weighted) flow time in the CP15 restricted assignment setting. On (1,)-Restricted Assignment Makespan Mini- mization Anamitra Roychoudhury IBM Research-India Makespan minimization on unrelated machines is a clas- [email protected] sic problem in approximation algorithms. In this paper we consider the (1,)-restricted assignment problem where Syamantak Das each job is either heavy (pj = 1) or light (pj = ), for some Indian Institute of Technology Delhi parameter >0. Each job can be assigned to a specified [email protected] set of machines. Our main result is a (2 − δ)-approximate polynomial time algorithm for the (1,)-restricted assign- δ> Naveen Garg, Amit Kumar ment problem for a fixed constant 0. IIT Delhi Deeparnab Chakrabarty [email protected], [email protected] Microsoft Research India [email protected] CP16 Revealing Optimal Thresholds for Generalized Sec- Sanjeev Khanna retary Problem Via Continuous Lp: Impacts on University of Pennsylvania Online K-Item Auction and Bipartite K-Matching [email protected] with Random Arrival Order

Shi Li We consider the general (J, K)-secretary problem. An al- Toyota Technological Institute at Chicago gorithm observes the relative merits of arriving items and [email protected] is allowed to make J selections. The objective is to maxi- mize the expected number of items selected among the K best items. We give a construction procedure for an op- CP15 timal algorithm involving JK thresholds via a continuous A Fully Polynomial-Time Approximation Scheme linear programming method. We further provide new re- for Speed Scaling with Sleep State sults on online K-item auction and bipartite K-matching with random arrival order. We study deadline-based preemptive scheduling of jobs in a computing environment equipped with both dynamic speed T-H. Hubert Chan, Fei Chen, Shaofeng H.-C. Jiang scaling and sleep state capabilities, in order to minimize the The University of Hong Kong total energy consumption. We present a fully polynomial- [email protected], [email protected], [email protected] time approximation scheme for this problem. Previously, only a 4/3-approximation was known (Albers and Anto- niadis, SODA 2012). CP16 Comparing Apples and Oranges: Query Tradeoff Antonios Antoniadis in Submodular Maximization Max-Planck-Institut f¨ur Informatik [email protected] Fast algorithms for submodular maximization problems have a vast potential use. Our main result is a new al- Chien-Chung Huang gorithm for the general problem of maximizing a submod- Chalmers University ular function subject to a matroid independence constraint. [email protected] Our algorithm is faster than the previous algorithm of Badanidiyuru and Vondr´ak (2014) when the rank is sub- Sebastian Ott linear in the size of the ground set. Furthermore, our algo- Max-Planck-Institut f¨ur Informatik rithm establishes a surprising tradeoff between the number [email protected] of value and independence oracle queries. Niv Buchbinder CP15 Statistics and Operations Research Dept. Tel Aviv University, Israel Rejecting Jobs to Minimize Load and Maximum [email protected] Flow-Time

Online algorithms are usually analyzed using the notion Moran Feldman of competitive ratio which compares the algorithm solu- EPFL tion to that obtained by an online adversary for the worst moran.feldman@epfl.ch possible input sequence. Often this measure turns out to be too pessimistic, and one popular approach especially Roy Schwartz for scheduling problems has been that of “resource aug- Princeton University mentation” which was first proposed by Kalyanasundaram [email protected] and Pruhs. Although resource augmentation has been very successful in dealing with a variety of objective functions, there are problems for which even a (arbitrary) constant CP16 speedup cannot lead to a constant competitive algorithm. Online Submodular Maximization with Preemp- 48 DA15 Abstracts

tion and the column-subset selection problem.

We study a natural online variant of submodular function Maxim Sviridenko maximization in which the algorithm has to maintain an Yahoo! Labs online solution obeying certain constraints. Upon arrival of [email protected] an element, the algorithm has to decide whether to accept the element and may preempt previously chosen elements. Jan Vondr´ak The goal is to maximize a submodular function over the IBM Almaden Research Center maintained solution. We derive upper and lower bounds [email protected] on the competitive ratio for the unconstrained case and for cardinality constraints. Justin Ward University of Warwick Niv Buchbinder [email protected] Statistics and Operations Research Dept. Tel Aviv University, Israel [email protected] CP17 Streaming Algorithms for Estimating the Matching Moran Feldman Size in Planar Graphs and Beyond EPFL moran.feldman@epfl.ch We consider the problem of estimating the size of a maxi- mum matching when the edges are revealed in a streaming Roy Schwartz fashion. When the input graph is planar, we present a sim- Princeton University ple and elegant streaming algorithm that with high proba- [email protected] bility estimates the size of a maximum matching within a constant factor using O(n2/3) space, where n is the num- ber of vertices. The approach generalizes to the family of CP16 graphs that have bounded arboricity, which include graphs ASimpleO(log log(rank))-Competitive Algorithm with an excluded constant-size minor. We furthermore de- for the Matroid Secretary Problem sign a reduction from the Boolean Hidden Matching Prob- lem to show that there is no randomized streaming algo- Only recently progress has been made in obtaining rithm that estimates the size of the maximum matching to / o(log(rank))-competitive algorithms for the matroid secre- within a factor better than 3/2 and uses only o(n1 2)bits tary problem. Chakraborty and Lachish (2012) presented a of space. O( log(rank))-competitive procedure, and Lachish (2014) Hossein Esfandiari recently presented a O(log log(rank))-competitive algo- University of Maryland at College Park rithm. Both algorithms are involved with complex anal- [email protected] yses. Using different tools, we present a considerably simpler O(log log(rank))-competitive algorithm. Our al- gorithm can be interpreted as a distribution over a simple MohammadTaghi Hajiaghayi type of matroid secretary algorithms which are easy to an- University of Maryland, College Park alyze. [email protected]

Moran Feldman, Ola Svensson Vahid Liaghat EPFL University of Maryland at College Park moran.feldman@epfl.ch, ola.svensson@epfl.ch [email protected]

Rico Zenklusen Morteza Monemizadeh ETH Zurich University of Frankfurt [email protected] [email protected]

Krzysztof Onak CP16 IBM TJ Watson Research Center Optimal Approximation for Submodular and Su- [email protected] permodular Optimization with Bounded Curvature

We design new approximation algorithms for the prob- CP17 lems of optimizing submodular and supermodular func- Streaming Lower Bounds for Approximating Max- tions with bounded curvature subject to a single matroid Cut constraint. Our approach is based on modifications of the continuous greedy algorithm and non-oblivious local We consider the problem of estimating the value of max cut search, and allows us to approximately maximize the sum in a graph in the streaming model of computation. At one of a nonnegative, nondecreasing submodular function and extreme, there is a trivial 2-approximation for this problem a (possibly negative) linear function. We show how to that uses only O(log n) space, namely, count the number of reduce both submodular maximization and supermodular edges and output half of this value as the estimate for max minimization to this general problem when the objective cut value. On the other extreme, a near-optimal solution function has bounded total curvature. We prove that the to the max cut value can be obtained in O˜(n)spaceby approximation results we obtain are the best possible in storing a cut sparsifier. An intriguing question is if poly- the value oracle model, even in the case of a cardinality logarithmic space suffices to obtain a non-trivial approx- constraint. Finally, we give two concrete applications of imation to the max-cut value (that is, beating the factor our results in the settings of maximum entropy sampling, 2). Our main result is that any streaming algorithm that DA15 Abstracts 49

√ breaks the 2-approximation barrier requires Ω(˜ n)space [email protected], asafi[email protected] even if the edges of the input graph are presented in ran- dom order. We also show that for any >0, any streaming algorithm that obtains a (1+)-approximation when edges CP17 arrive in adversarial order requires n1−O() space. A Polylogarithmic Space Deterministic Streaming Algorithm for Approximating Distance to Mono- Michael Kapralov tonicity MIT [email protected] We give the first deterministic streaming algorithm that approximates the distance to monotonicity within a 1 + ε Sanjeev Khanna factor for any fixed ε>0 and runs in space polylogarithmic University of Pennsylvania in the length of the sequence and the range of the numbers. [email protected] The best previous deterministic algorithm achieving√ the same approximation factor required space Ω( n). Previ- Madhu Sudan ous polylogarithmic space algorithms were either random- Microsoft Research New England ized, or had approximation factor no better than 2. [email protected] Timothy R. Naumovitz, Michael Saks Rutgers University CP17 [email protected], [email protected] Parameterized Streaming: Maximal Matching and Vertex Cover CP18 In this paper, we introduce a new approach to handling Combinatorial Algorithm for Restricted Max-Min graph streams, by instead seeking solutions for the param- Fair Allocation eterized versions of these problems. Here, we are given a k parameter and the objective is to decide whether there is We study the basic allocation problem of assigning re- k a solution bounded by . By combining kernelization tech- sources to players so as to maximize fairness. This is one niques with randomized sketch structures, we obtain the of the few natural problems that enjoys the intriguing sta- first streaming algorithms for the parameterized versions tus of having a better estimation algorithm than approxi- of Maximal Matching and Vertex Cover. mation algorithm. Indeed, a certain configuration-LP can be used to estimate the value of the optimal allocation to Rajesh Chitnis within a factor of 4 + . In contrast, however, the best Department of Computer Science known approximation algorithm for the problem has an University of Maryland at College Park, USA unspecified large constant guarantee. In this paper we sig- [email protected] nificantly narrow this gap by giving a 13-approximation algorithm for the problem. Graham Cormode University of Warwick Chidambaram Annamalai, Christos Kalaitzis, Ola [email protected] Svensson EPFL MohammadTaghi Hajiaghayi chidambaram.annamalai@epfl.ch, chris- University of Maryland, College Park tos.kalaitzis@epfl.ch, ola.svensson@epfl.ch [email protected]

Morteza Monemizadeh CP18 Institute for Computer Science Capacity of Interactive Communication over Era- Goethe-University Frankfurt am Main sure Channels and Channels with Feedback [email protected] We consider interactive communication performed over two simple types of noisy channels: binary error channels with CP17 noiseless feedback and binary erasure channels. In both Decomposing a Graph Into Expanding Subgraphs cases, the noise model is adversarial. Assuming at most ε-fraction of the bits can be corrupted, we show coding A paradigm that was successfully applied in the study of schemes that simulate any alternating interactive protocol both pure and algorithmic problems in graph theory can with rate 1−Θ(H(ε)). All our simulations are simple, ran- be colloquially summarized as stating that any graph is domized, and computationally efficient. The rates of our close to being the disjoint union of expanders. Our goal in coding schemes stand in contrast to the interactive com- this paper is to show that in several of the instantiations munication rates supported by random or adversarial error of the above approach, the quantitative bounds that were channels without feedback. Such a gap has no equivalent obtained are essentially best possible. These instantiations in the standard one-way communication setting. include the Unique Games algorithm of Trevisan [FOCS ’05], the eigenspace enumeration method of Arora, Barak Ran Gelles and Steurer [FOCS ’10], and a classical result of Lipton, Princeton University RoseandTarjanongraphseparators. [email protected]

Guy Moshkovitz, Asaf Shapira Bernhard Haeupler Tel Aviv University Carnegie Mellon U. 50 DA15 Abstracts

[email protected] and absolute distance d is testable with O(N/d)queries, and has a dimension of ≈ N − (log N)log d. On the other hand, the smallest possible co-dimension for a distance d CP18 ≈ d N code (without any testability requirement) is 2 log , Lp/sdp Hierarchy Lower Bounds for Decoding achieved by BCH codes. This raises the natural question Random Ldpc Codes of understanding the optimal co-dimension of a distance d LTC. One promising approach for constructing LTCs is Random (dv,dc)-regular LDPC codes are well-known to to focus on affine-invariant codes. In this work, we show achieve the Shannon capacity of the binary symmetric that a recent family of affine-invariant constructions is es- channel (for sufficiently large dv and dc) under exponen- sentially optimal among linear affine-invariant LTCs that tial time decoding. However, polynomial time algorithms contain the Reed-Muller code of the appropriate degree. are only known to correct a much smaller fraction of errors. One of the most powerful polynomial-time algorithms with Venkatesan Guruswami a formal analysis is the LP decoding algorithm of Feldman Carnegie Mellon University et al. which is known to correct an Ω(1/dc) fraction of er- [email protected] rors. In this work, we show that fairly powerful extensions of LP decoding, based on the Sherali-Adams and Lasserre Madhu Sudan hierarchies, fail to correct much more errors than the ba- Microsoft Research New England sic LP-decoder. In particular, we show that for any (resp. [email protected] infinitely many) values of dv and dc, a linear number of rounds of the Sherali-Adams LP (resp. Lasserre SDP) hi- Ameya Velingker erarchy cannot correct more than an O(1/dc)fractionof Carnegie Mellon University errors on a random (dv,dc)-regular LDPC code. Computer Science Department Badih Ghazi [email protected] MIT [email protected] Carol Wang Carnegie Mellon University Euiwoong Lee [email protected] Carnegie Mellon University [email protected] CP19 Fast Algorithms for Online Stochastic Convex Pro- CP18 gramming On the Quickest Flow Problem in Dynamic Net- works – A Parametric Min-Cost Flow Approach We introduce the online stochastic Convex Program- ming(CP) problem, a very general version of stochastic We consider the Quickest Flow (QF) problem in dynamic online problems which allows arbitrary concave objectives networks with a single source s and a single sink t:givenan and convex feasibility constraints. Online stochastic pack- amount of flow F , find the minimum time needed to send it ing and covering form special cases of online stochastic from s to t, and the corresponding optimal flow over time. CP. We present fast algorithms for these problems, which We introduce new mathematical formulations and derive achieve near-optimal regret guarantees for both the i.i.d. optimality conditions for QF. Based on the optimality con- and the random permutation models of stochastic inputs. ditions, we develop a cost-scaling algorithm that leverages Our techniques make explicit the connection of primal-dual the parametric nature of the problem. The algorithm solves paradigm and online learning to online stochastic CP QF with integer arc costs in O(nm log(n2/m)log(nC)) time, where n, m,andC are the number of nodes, arcs, Shipra Agrawal and the maximum arc cost, respectively. Our algorithm Microsoft runs in the same time bound as the cost-scaling algorithm [email protected] by Goldberg and Tarjan for solving the min-cost flow prob- lem. This result shows for the first time that QF can be Nikhil R. Devanur solved within the same time bound as one of the fastest Microsoft Research algorithms for the min-cost flow problem. [email protected]

Patrick Jaillet MIT CP19 [email protected] Using Optimization to Break the Epsilon Barrier: A Faster and Simpler Width-Independent Algo- Maokai Lin rithm for Solving Positive Linear Programs in Par- Operations Reseach Center, MIT allel [email protected] We study the design of nearly-linear-time algorithms for approximately solving positive linear programs. Both the CP18 parallel and the sequential deterministic versions of these −4 −1/4 Limitations on Testable Affine-Invariant Codes in algorithms require O˜(ε ) iterations (i.e., have a O(T ) the High-Rate Regime convergence rate), a dependence that has not been im- proved since the introduction of these methods in 1993 by Locally testable codes (LTCs) of constant minimum (abso- Luby and Nisan. Moreover, previous algorithms and their lute) distance that allow the tester to make a nearly linear analyses rely on update steps and convergence arguments number of queries have become the focus of attention re- that are combinatorial in nature, but do not seem to arise cently. The binary Reed-Muller code of block length N naturally from an optimization viewpoint. In this paper, DA15 Abstracts 51

we leverage insights from optimization theory to construct University of Szeged a novel algorithm that breaks the O˜(ε−4) barrier. Our Hungary algorithm has a simple analysis and a clear motivation. [email protected], [email protected] Our work introduces a number of techniques, such as the combined application of gradient and mirror descent, and Gyorgy Dosa a truncated version of the multiplicative weight update, University of Pannonia which may be of independent interest. Hungary [email protected] Zeyuan Allen-Zhu MIT CSAIL Jiri Sgall [email protected] Computer Science Institute Charles University, Czech Republic Lorenzo Orecchia [email protected]ff.cuni.cz MIT [email protected] Rob van Stee Max Planck Institute for Informatics CP19 [email protected] Online Network Design Algorithms Via Hierarchi- cal Decompositions CP20 We develop a new approach for online network design and A Unified Framework for Clustering Constrained obtain improved competitive ratios for several problems. Data Without Locality Property At the heart of our work is a novel application of embed- dings into hierarchically well-separated trees (HSTs) to the We consider a class of constrained clustering problems d analysis of online network design algorithms. Using this in R ,whered could be rather high. A common fea- approach, we obtain simple greedy-style online algorithms ture of these problems is that their optimal clusterings no and analyses. In this work, we apply it to Steiner network longer have the locality property (due to the additional (with edge-duplication), rent-or-buy, connected facility lo- constraints), which is a key property required by many ex- cation and prize-collecting Steiner forest. isting clustering algorithms. To overcome this difficulty, we present a unified framework, and improve considerably Seeun W. Umboh the best known results. University of Wisconsin-Madison [email protected] Hu Ding,JinhuiXu Department of Computer Science and Engineering State University of New York at Buffalo CP19 huding@buffalo.edu, jinhui@buffalo.edu Online Stochastic Matching with Unequal Proba- bilities CP20 We study a variant of online bipartite matching in which On the Richter-Thomassen Conjecture About Pair- edges are labeled with the probability with which a match wise Intersecting Closed Curves along this edge “succeeds’. Prior work gave a competitive ratio of 0.567 for vanishing probabilities when all probabil- A long standing conjecture of Richter and Thomassen ities in the graph are equal. We consider general, unequal states that the total number of intersection points between probabilities, and show a ratio of 0.534 for the vanishing any n pairwise intersecting simple closed Jordan curves in case with a different algorithmic approach: maximizing the plane is at least (1 − o(1))n2. We confirm the above marginal contributions while carefully controlling the level conjecture in several important cases, including the case of adaptiveness. (1) when all curves are convex, and (2) when the family of Aranyak Mehta curves can be partitioned into two equal classes such that Google each curve from the first class is touching every curve from [email protected] the second class. An important ingredient of our proofs is the following statement: Let S be a family of the graphs of n continuous real functions defined on R,nothreeof Bo Waggoner which pass through the same point. If there are nt pairs of Harvard touching curves in S, then the number of crossing points [email protected]  is Ω(nt log t/ log log t).

Morteza Zadimoghaddam J´anos Pach Google EPFL, Lausanne [email protected] R´enyi Institute, Budapest [email protected] CP19 Natan Rubin The Optimal Absolute Ratio for Online Bin Pack- Department of Computer Science ing Ben Gurion University We present an online bin packing algorithm with absolute [email protected] competitive ratio 5/3, which is optimal. G´abor Tardos Janos Balogh, Jozsef Bekesi R´enyi Institute, Budapest 52 DA15 Abstracts

[email protected] Knapsack Problem

We consider the two-dimensional geometric knapsack prob- CP20 lem. Given a collection of rectangular axis-parallel items Density and Regularity Theorems for Semi- with weights, we want to find a maximum weight subset of Algebraic Hypergraphs the items that can be packed into a rectangular knapsack. The goal is to compute the optimal collection of items to- A k-uniform semi-algebraic hypergraph H is a pair (P, E), gether with a feasible packing. For the general case the best d known result is a (2 + )-approximation algorithm, while where P is a subset of R and E is a collection of k-tuples the only hardness result is NP-hardness. Our main result {p ,...,pk}⊂P such that (p ,...,pk) ∈ E if and only 1 1 is a (1 + )-approximation algorithm that runs in quasi- if the kd coordinates of the pi-s satisfy a boolean combi- polynomial time, provided that the input data consists of nation of a finite number of polynomial inequalities. Im- (quasi-)polynomially bounded integers. proving a theorem of Fox, Gromov, Lafforgue, Naor, and Pach, we establish a polynomial regularity lemma: For any Anna Adamaszek, Andreas Wiese >0, the vertex set of every k-uniform semi-algebraic hy- c MPII Saarbr¨ucken pergraph can be partitioned into at most (1/) parts such [email protected], [email protected] that all but an -fraction of the k-tuples of parts are com- plete or empty. We also establish an improved lower bound on the best constant δ>0 such that the vertex classes of CP21 every dense k-partite k-uniform semi-algebraic hypergraph H =(P1 ∪ ... ∪ Pk,E)haveδ|Pi|-element subsets that The Switch Markov Chain for Sampling Irregular induces a complete k-partite hypergraph. We give 3 geo- Graphs metric applications of our results. The problem of efficiently sampling from a set of (undi- Jacob Fox rected) graphs with a given degree sequence has many ap- MIT plications. One approach to this problem uses a simple [email protected] Markov chain, which we call the switch chain, to perform the sampling. The switch chain is known to be rapidly mix- Janos Pach ing for regular degree sequences. We prove that the switch Courant Institute chain is rapidly mixing for any degree sequence with min- d [email protected] imum degree at least 1 and√ with maximum degree max ≤ d ≤ 1 M M which satisfies 3 max 4 ,where is the sum of Andrew Suk the degrees. The mixing time bound obtained is only an order of n larger than that established in the regular case, University of Illinois at Chicago n [email protected] where is the number of vertices. Catherine Greenhill CP20 University of New South Wales School of Mathematics and Statistics Approximation Schemes for Partitioning: Convex [email protected] Decomposition and Surface Approximation

Building on the discovery of new types of separators by CP21 Adamaszek and Wiese, there has recently been significant progress on approximability of planar packing problems as FPTAS for Counting Monotone CNF well as certain covering problems. Following these devel- A monotone CNF formula is a Boolean formula in conjunc- opments, we revisit two NP-hard geometric partitioning tive normal form where each variable appears positively. problems - convex decomposition and surface approxima- We design a deterministic fully polynomial-time approxi- tion. These problems bring up additional issues that are mation scheme (FPTAS) for counting the number of satis- worth examining in the context of the wider applicability fying assignments for a given monotone CNF formula when of the separator methodology; we show how these issues each variable appears in at most 5 clauses. Equivalently, can be addressed. this is also an FPTAS for counting set covers where each Sayan Bandyapadhyay set contains at most 5 elements. If we allow variables to Department of Computer Science appear in a maximum of 6 clauses (or sets to contain 6 ele- The University of Iowa ments), it is NP-hard to approximate it. Thus, this gives a [email protected] complete understanding of the approximability of counting for monotone CNF formulas. It is also an important step towards a complete characterization of the approximability Santanu Bhowmick for all bounded degree Boolean #CSP problems. In ad- University of Iowa dition, we study the hypergraph matching problem, which [email protected] arises naturally towards a complete classification, and show an FPTAS for counting 3D matchings of hypergraphs with Kasturi Varadarajan maximum degree 4. Department of Computer Science The University of Iowa Jingcheng Liu [email protected] UC Berkeley [email protected]

CP20 Pinyan Lu A Quasi-Ptas for the Two-Dimensional Geometric Microsoft Research Asia DA15 Abstracts 53

[email protected] [email protected]

CP21 CP21 The Speed of Evolution Phase Transitions in Random Dyadic Tilings and In this paper we study the mixing times of a Wright-Fisher Rectangular Dissections model for an asexual population. Despite their importance, and having been widely studied, there has been a lack of We study rectangular dissections of an n × n lattice region rigorous results establishing fast mixing times for many into rectangles of area n,wheren =2k foraneveninteger relevant ranges of model parameters even for the case of k. We show there is a natural edge-flipping Markov chain two genotypes (single locus). The main result of this paper that connects the state space. A similar chain is known to is a tight analytical bound on the mixing time of a Wright- connect the state space when restricted to dyadic tilings, Fisher model for two genotypes when there is no restriction where each rectangle has the form R =[s2u, (s +1)2u] × on the mutation rate or the fitness. [t2v , (t+1)2v], where s, t, u and v are nonnegative integers. We consider a weighted version where, given a parameter Nisheeth K. Vishnoi λ>0, we generate each rectangular dissection (or dyadic EPFL tiling) σ with probability proportional to λ|σ|, where |σ| is EPFL the total edge length. We show there is a phase transition nisheeth.vishnoi@epfl.ch in the dyadic setting: when λ<1, the edge-flipping chain mixes in time O(n2 log n), and when λ>1, the mixing time CP22 is exp(Ω(n2)). In the general setting, we show the chain requires exponential time when λ>1andwhenλ<1. Optimal Detection of Intersections Between Con- vex Polyhedra

Sarah Cannon, Sarah Miracle, Dana Randall For a polyhedron P in Rd,denoteby|P | its combina- Georgia Institute of Technology torial complexity, i.e., the number of faces of all dimen- [email protected], [email protected], ran- sions of the polyhedra. In this paper, we revisit the classic [email protected] problem of preprocessing polyhedra independently so that given two preprocessed polyhedra P and Q in Rd,each translated and rotated, their intersection can be tested rapidly. For d = 3 we show how to perform such a test in CP21 O(log |P |+log |Q|) time after linear preprocessing time and space. This running time is the best possible and improves Spatial Mixing and the Connective Constant: Op- O |P | |Q| timal Bounds upon the last best known query time of (log log ) by Dobkin and Kirkpatrick (1990). We then generalize our method to any constant dimension d,achievingthesame We study the problem of deterministic approximate count- optimal O(log |P | +log|Q|) query time using a representa- ing of matchings and independent sets in graphs of tion of size O(|P |d/2+ε) for any ε>0 arbitrarily small. bounded connective constant. More generally, we consider This answers an even older question posed by Dobkin and the problem of evaluating the partition functions of the Kirkpatrick 30 years ago. monomer-dimer model and the hard core model on such graphs. The connective constant is a natural notion of av- Luis Barba erage degree of a graph which has been studied extensively Carleton University in combinatorics and mathematical physics, and can be [email protected] bounded by a constant even for certain unbounded degree graphs such as those sampled from the sparse Erd˝os–R´enyi Stefan Langerman model G(n, d/n). For both models, we prove the best pos- Universit´e Libre de Bruxelles sible rates of decay of correlations in graphs with a given [email protected] bound on the connective constant. We then use these opti- mal decay of correlations results to obtain FPTASs for the two problems on graphs of bounded connective constant. CP22 Our results improve upon previous results which were often Compatible Connectivity-Augmentation of Planar applicable only to bounded degree graphs. Disconnected Graphs

Alistair Sinclair Motivated by applications to graph morphing, we consider University of California, Berkeley the following compatible connectivity-augmentation prob- n G [email protected] lem: We are given a labelled -vertex planar graph, , that has r ≥ 2 connected components, and k ≥ 2iso- morphic planar straight-line drawings, G1,...,Gk,ofG. Piyush Srivastava We wish to augment G by adding vertices and edges to California Institute of Technology make it connected in such a way that these vertices and [email protected] edges can be added to G1,...,Gk as points and straight- line segments, respectively, to obtain k planar straight-line Daniel tefankovic drawings isomorphic to the augmentation of G.Weshow University of Rochester that adding Θ(nr1−1/k)edgesandverticestoG is always [email protected] sufficient and sometimes necessary to achieve this goal.

Yitong Yin Greg Aloupis Nanjing University Universite Libre de Bruxelles 54 DA15 Abstracts

[email protected] weighted planar subdivision, were the cost of a path within a face is the product of its length and the face weight. We Luis Barba present a fully polynomial-time approximation scheme for Carleton University this problem. Our algorithm relies on a new triangulation [email protected] refinement method that produces a triangulation such that no triangle has two small angles. Paz Carmi Ben-Gurion University of the Negev Siu-Wing Cheng [email protected] The Hong Kong University of Science and Technology [email protected] Vida Dujmovic University of Ottawa Jiongxin Jin [email protected] Google Inc. [email protected] Fabrizio Frati The University of Sydney Antoine Vigneron [email protected] King Abdullah University of Science and Technology (KAUST) [email protected] Pat Morin Carleton University [email protected] CP22 Detecting Weakly Simple Polygons CP22 Geometric k Shortest Paths A closed curve in the plane is weakly simple if it is the limit (in the Frchet metric) of a sequence of simple closed We consider computing k homotopically distinct shortest curves. We describe an algorithm to determine whether a paths in a polygonal domain. We give algorithms for find- closed walk of length n in a simple plane graph is weakly ing the paths, and present the kth shortest path map – a simple in O(n log n) time, improving an earlier O(n3)-time structure allowing one to report efficiently the length of the algorithm of Cortese et al. [Discrete Math. 2009]. As an im- k shortest paths to a query point. mediate corollary, we obtain the first efficient algorithm to determine whether an arbitrary n-vertex polygon is weakly Sylvester Eriksson-Bique simple; our algorithm runs in O(n2 log n)time.Wealsode- Courant Institute, NYU scribe algorithms that detect weak simplicity in O(n log n) [email protected] time for two interesting classes of polygons. Finally, we discuss subtle errors in several previously published defini- John Hershberger tions of weak simplicity. Mentor Graphics Corporation john [email protected] Hsien-Chih Chang, Jeff Erickson, Chao Xu University of Illinois, Urbana-Champaign Valentin Polishchuk [email protected], jeff[email protected], Communications and Transport Systems [email protected] ITN, Link¨oping University [email protected] CP23 Bettina Speckmann Subcubic Equivalences Between Graph Centrality Dept. of Mathematics and Computer Science Problems, Apsp and Diameter TU Eindhoven [email protected] In the analysis of social networks there are many important centrality parameters such as radius, diameter, between- Subhash Suri ness centrality, and reach centrality. The known algorithms University of California, Santa Barbara for computing these measures run in time that is roughly [email protected] cubic in the number of nodes. We present a set of reduc- tions showing that these algorithms cannot be improved Topi L. Talvitie without improving the best known algorithms for the clas- University of Helsinki sical problems All Pairs Shortest Paths (APSP) and graph topi.talvitie@helsinki.fi diameter. Moreover, we prove that most graph centrality measures are equivalent to either APSP or diameter. Kevin Verbeek, Hakan Yildiz Amir Abboud Computer Science Stanford University University of California Santa Barbara [email protected] [email protected], [email protected] Fabrizio Grandoni CP22 IDSIA, University of Lugano Triangulation Refinement and Approximate Short- [email protected] est Paths in Weighted Regions Virginia Vassilevska Williams We consider the problem of computing a shortest path in a Stanford University DA15 Abstracts 55

[email protected] generalizes both the aforementioned results and verifies a conjecture of Joret et al. The proof relies on the Robertson- Seymour and Grohe-Marx structural decomposition theo- CP23 rems. Forbidden Structure Characterization of Circular- Arc Graphs and a Certifying Recognition Algo- Bartosz Walczak rithm Jagiellonian University [email protected] A circular-arc graph is the intersection graph of arcs of a circle. It is a well-studied graph model with numerous natural applications. A certifying algorithm is an algo- CP23 rithm that outputs a certificate, along with its answer (be Finding Four-Node Subgraphs in Triangle Time it positive or negative), where the certificate can be used to easily justify the given answer. We present the first We present new algorithms for finding induced four-node forbidden structure characterization of circular-arc graphs subgraphs in a given graph, which run in time roughly that and the first polynomial-time certifying algorithm for the of detecting a clique on three nodes. recognition of circular-arc graphs. • The best known algorithms for triangle finding in an ω Mathew Francis n-node graph take O(n ) time, where ω<2.373 is the Indian Statistical Institute, Chennai Centre matrix multiplication exponent. We give a general Computer Science Unit randomized technique for finding any induced four- [email protected] node subgraph, except for the clique or independent set on 4 nodes, in O˜(nω) time with high probability. Pavol Hell The algorithm can be derandomized in some cases. School of Computer Science • For sparse graphs with m edges, the best known Simon Fraser University triangle finding algorithm runs in O(m2ω/(ω+1)) ≤ [email protected] . ω/ ω O(m1 41) time. We give a randomized O˜(m2 ( +1)) time algorithm for finding any induced four-node sub- Juraj Stacho graph other than C4, K4 and their complements. Columbia University in the City of New York Some cases can be derandomized. For C4 or its com- Department of Industrial Engineering & Operations plement, we get slightly slower algorithms. Research [email protected] Virginia Vassilevska Williams, Joshua R. Wang,Ryan Williams, Huacheng Yu CP23 Stanford University Four Terminal Planar Delta-Wye Reducibility Via [email protected], [email protected], [email protected], [email protected] Rooted K2,4 Minors

A graph with four special vertices (terminals) is wye-delta reducible if we can obtain a graph on 4 vertices by a se- CP24 quence of wye-delta operations and series-parallel reduc- Lopsidependency in the Moser-Tardos Framework: tions, none of which removes a terminal. A good charac- Beyond the Lopsided Lov´asz Local Lemma terization of wye-delta reducible 3-connected planar graphs with 4 terminals is given. The proofs yield an O(n2)al- The Lopsided Lov´asz Local Lemma (LLLL) is a powerful gorithm that either finds an obstruction to reducibility or principle in probability, along with the efficient resampling returns a sequence of reductions. The main ingredient in algorithm Moser & Tardos. We show a new criterion for the proofs is a good characterization of planar graphs with the Moser-Tardos algorithm to converge. This criterion is 4 terminals that do not admit a rooted K2,4 minor with the stronger than the LLLL criterion, and in fact can yield bet- terminals corresponding to the roots on the large side of the ter results even than the full Shearer criterion. A notewor- k bipartition of K2,4. Up to small connectivity reductions, thy application is for -SAT. As shown in Gebauer, Sz´abo, cases without the rooted minor fall into five structural cases and Tardos, a k-SAT instance in which every variable ap- k+1 that lead to a polynomial-time algorithm for recognition of L ≤ 2 pears e(k+1) times, is satisfiable. We improve it to these graphs and construction of rooted K , minors. This k+1 k 2 4 L ≤ 2 (1−1/k) − 2 result is of independent interest in structural graph theory. k−1 k which can be significantly stronger when k is small. We introduce a new parallel algorithm Lino Demasi, Bojan Mohar for the LLLL, which applies in nearly all settings as the Simon Fraser University sequential algorithm. [email protected], [email protected] David Harris University of Maryland CP23 [email protected] Minors and Dimension

Streib and Trotter proved in 2012 that posets with bounded CP24 height and with planar cover graphs have bounded dimen- Characterization of Cutoff for Reversible Markov sion. Recently, Joret et al. proved that the dimension Chains is bounded for posets with bounded height whose cover graphs have bounded tree-width. In this paper, it is proved A sequence of Markov chains is said to exhibit (total varia- that posets of bounded height whose cover graphs exclude tion) cutoff if the convergence to stationarity in total varia- a fixed (topological) minor have bounded dimension. This tion distance is abrupt. We consider reversible lazy chains. 56 DA15 Abstracts

We prove a necessary and sufficient condition for the oc- such that G has the property P even if we allow an adver- currence of the cutoff phenomena in terms of concentration sary to remove an r-fraction of (in- and out-going) edges of hitting time of “worst” (in some sense) sets of stationary touching each vertex. Lee and Sudakov (2012) proved that measure at least α,forsomeα ∈ (0, 1). As an application the local resilience of a random graph with edge prob- of our techniques we show that a sequence of lazy Markov ability p = ω(log n/n) with respect to Hamiltonicity is chains on finite trees exhibits a cutoff iff the product of 1/2 ± o(1). For random directed graphs, Hefetz, Steger their spectral gaps and their (lazy) mixing-times tends to and Sudakov√ (2014+) proved an analogue statement for ∞. p = ω(log n/ n). In this paper we improve their result to p = ω(log8 n/n), which is optimal up to the polylogarith- Riddhipratim Basu mic factor. Department of Statistics, UC Berkeley, California, USA [email protected] Asaf Ferber ETH Zurich Jonathan Hermon [email protected] 367 Evans Hall, University of California Berkeley, CA 94720-3860 Rajko Nenadov [email protected] ETH Z¨urich [email protected] Yuval Peres Microsoft Research, Redmond Andreas Noever, Ueli Peter, Nemanja Skoric [email protected] ETH Zurich [email protected], [email protected], nsko- [email protected] CP24 An Algorithmic Framework for Obtaining Lower Bounds for Random Ramsey Problems CP24 Surprise Probabilities in Markov Chains In this paper we introduce a general framework for prov- ing lower bounds for various Ramsey type problems within In a Markov chain started at a state x, the hitting time random settings. The main idea is to provide an algo- τ(y) is the first time that the chain reaches another state rithm that finds the desired colouring w.h.p. Our frame- y. We study the probability Px(τ(y)=t) that the first work allows to reduce the probabilistic problem of whether visit to y occurs precisely at a given time t. Informally the Ramsey property holds for random (hyper)graphs to speaking, the event that a new state is visited at a large a deterministic question of whether there exists a finite time t may be considered a “surprise’. We prove: graph that forms an obstruction. We apply this framework n • In any Markov chain with n states, Px(τ(y)=t) ≤ . to solve various open problems. We provide a matching t • n lower bound for the result of Friedgut, R¨odl and Schacht In a reversible√ chain with states, τ y t ≤ 2n (2010) and, independently, Conlon and Gowers (2014+) Px( ( )= ) t for the classical Ramsey problem for hypergraphs in the for t ≥ 4n +4. case of cliques. We also improve a result of Bohman, • For random walk on a simple graph with n ≥ 2ver- Frieze, Pikhurko and Smyth (2010) for bounded anti- 4e log n Px τ y t ≤ Ramsey problems in random graphs and extend it to hy- tices, ( ( )= ) t . pergraphs. Finally, we provide matching lower bounds for We construct examples showing that these bounds are close a proper-colouring version of anti-Ramsey problems intro- to optimal. The main feature of our bounds is that they duced by Kohayakawa, Konstadinidis and Mota (2014). require very little knowledge of the structure of the Markov chain. Rajko Nenadov ETH Z¨urich James Norris [email protected] University of Cambridge [email protected] Nemanja Skoric ETH Zurich Yuval Peres [email protected] Microsoft Research, Redmond [email protected] Angelika Steger Institute of Theoretical Computer Science Alex Zhai ETH, Z¨urich Stanford University [email protected] [email protected]

CP24 CP25 Robust Hamiltonicity of Random Directed Graphs Testing Poisson Binomial Distributions

In 1952 Dirac showed that Kn remains Hamiltonian even A Poisson Binomial distribution over n variablesisthedis- if we allow an adversary to remove n/2 edges touching tribution of the sum of n independent Bernoullis. We pro- each vertex. In 1960 Ghouila-Houri obtained an analogue vide a sample near-optimal algorithm for testing whether statement for digraphs. A natural way to generalize such a distribution P supported on {0,...,n} to which we have results to arbitrary graphs (digraphs) is using the notion sample access is a Poisson Binomial distribution, or far of local resilience. The local resilience of a graph (digraph) from all Poisson Binomial distributions. The sample com- G with respect to a property P is the maximum number r plexity of our algorithm is O(n1/4) to which we provide a DA15 Abstracts 57

matching lower bound. We note that our sample complex- [email protected], [email protected] ity improves quadratically upon that of the naive “learn followed by tolerant-test’ approach, while instance opti- mal identity testing [VV14] is not applicable since we are CP25 looking to simultaneously test against a whole family of Algorithmic Regularity for Polynomials and Appli- distributions. cations

Jayadev Acharya, Constantinos Daskalakis We discuss several notions of regularity for collections of MIT polynomials, which lead to algorithmic versions of various [email protected], [email protected] polynomial regularity lemmas such as the ones by Green and Tao [Contrib. Discrete Math. 2009] and by Kaufman and Lovett [FOCS 2008]. We show that our algorithmic CP25 regularity lemmas for polynomials also lead to algorithmic Testing Identity of Structured Distributions versions of many results relying on regularity, such as de- coding Reed-Muller codes beyond the list decoding radius We study the question of identity testing for structured (for certain structured errors) and worst-case to average- distributions. More precisely, given samples from a struc- case reductions for polynomials. We also provide algorith- tured distribution q over [n] and an explicit distribution p mic versions of some of the inverse theorems for Gowers over [n], we wish to distinguish whether q = p versus q norm. is at least -far from p,inL1 distance. In this work, we present a unified approach that yields new, simple testers, Arnab Bhattacharyya with sample complexity that is information-theoretically Indian Institute of Science optimal, for broad classes of structured distributions. [email protected]

Ilias Diakonikolas Pooya Hatami University of Edinburgh University of Chicago [email protected] [email protected]

Daniel Kane Madhur Tulsiani UC San Diego TTI Chicago [email protected] [email protected]

Vladimir Nikishkin University of Edinburgh CP25 [email protected] The Complexity of Estimating R´enyi Entropy

It was recently shown that estimating the Shannon en- CP25 tropy H(p)ofadiscretek-symbol distribution p requires Θ(k/log k) samples, a number that grows near-linearly Property Testing on Product Distributions: Opti- H p mal Testers for Bounded Derivative Properties in the support size. In many applications ( )canbe replaced by the more general R´enyi entropy of order α, H p We study algorithms that, given access to a small num- α( ). We determine the number of samples needed to H p α α< ber of samples from a function over large discrete domain, estimate α( ) for all , showing that 1requires 1/α approximately determine whether the dataset satisfies a super-linear, roughly k samples, noninteger α>1re- desired property. The distance to having the property is quires near-linear, roughly k samples, but integer α>1 − /α measured with respect to a known or an unknown distribu- requires only Θ(k1 1 ) samples. In particular, estimating tion on the datasets. We focus on functions over domains H2(p), which arises in security, DNA reconstruction, close- d d √ of the form [n] =1, 2, .., n (d-dimensional hypergrids). ness testing, and other applications, requires only Θ( k) We give an optimal tester for a general class of proper- samples. The estimators achieving these bounds are simple ties of such functions, called bounded derivative proper- and run in time linear in the number of samples. ties (BDP), for the case when the distance to the property is measured with respect to a product distribution. The Jayadev Acharya BDPs include monotonicity and the Lipschitz property. MIT We use our novel and optimum (upto constant factors) [email protected] dimension reduction, which reduces testing properties of d functions over [n] to testing functions over [n]. Our work Alon Orlitsky, Ananda Theertha Suresh resolves two open problems given by Ailon and Chazelle University of California, San Diego (Information and Computation, 2006) and improves their [email protected], [email protected] line-tester for monotonicity. Himanshu Tyagi Deeparnab Chakrabarty UC San Diego Microsoft Research India [email protected] [email protected]

Kashyap Dixit CP26 Penn State university A Dynamic Model of Barter Exchange [email protected] We consider the problem of efficient operation of a barter Madhav Jha, C. Seshadhri exchange platform for indivisible goods. We introduce a Sandia National Labs dynamic model where in each period one agent arrives 58 DA15 Abstracts

with a single item she wants to exchange for a different [email protected] item. An agent is interested in the item possessed by an- other agent with probability p, independently for all pairs of agents. We consider three settings with respect to the CP26 types of allowed exchanges: a) Only two-way cycles, b) The Structure of the Core in Assignment Markets Two or three-way cycles, c) (unbounded) chains initiated by altruistic donors. The goal of the platform is to mini- Assignment markets (Shapley & Shubik 1971) involve mize the average waiting time of an agent. We find that matching with transfers, as in labor and housing markets. in each of these settings, a policy that conducts exchanges We consider a two-sided assignment market with agent in a greedy fashion is near optimal, among a large class of types and stochastic structure similar to models used in policies that includes batching policies. Further, for small empirical studies. Each agent has a randomly drawn ”pro- p, allowing three-cycles can greatly improve the waiting ductivity” associated with each type on the other side. We time over the two-cycles only setting, and the presence of characterize how the structure of the core, i.e., the set of altruistic donors can lead to a further large improvement. stable outcomes, is determined by market characteristics. Yash Kanoria Ross Anderson Columbia Business School MIT ORC [email protected] [email protected] Daniela Saban, Jay Sethuraman Itai Ashlagi Columbia University MIT Sloan School of Management [email protected], [email protected] [email protected] CP26 David Gamarnik Robust Randomized Matchings MIT [email protected] The following zero-sum game is played on a weighted graph G: Alice selects a matching M in G and Bob selects a Yash Kanoria number k. Alice receives a payoff equal to the ratio of the Columbia Business School weight of the top k edges of M to the maximum weight of a [email protected] matching of size at most k in G.IfM guarantees a payoff of at least α then it is called α-robust. In 2002, Hassin and Rubinstein gave an algorithm that returns a 1/2-robust matching, which is best possible for this setting. We show CP26 that Alice can improve on the guarantee of 1/2when allowing her to play a randomized strategy. We devise a Approximately Stable, School Optimal, and simple algorithm that returns a 1/ ln(4)-robust randomized Student-Truthful Many-to-One Matchings (via Dif- matching. We further show that our robustness results for ferential Privacy) randomized matchings translate to an asymptotic robust- ness guarantee for deterministic matchings and we give a new simple LP-based proof of Hassin and Rubinstein’s orig- We present a mechanism for computing asymptotically sta- inal result. ble school optimal matchings, while guaranteeing that it is Jannik Matuschke, Martin Skutella an asymptotic dominant strategy for every student to re- TU Berlin port their true preferences to the mechanism. Our main [email protected], tool in this endeavor is differential privacy: we give an [email protected] algorithm that coordinates a stable matching using dif- ferentially private signals, which lead to our truthfulness guarantee. This is the first setting in which it is known Jos´eSoto how to achieve nontrivial truthfulness guarantees for stu- Universidad de Chile dents when computing school optimal matchings, assuming [email protected] worst-case preferences (for schools and students) in large markets. CP26 Bayesian Truthful Mechanisms for Job Scheduling Sampath Kannan from Bi-Criterion Approximation Algorithms University of Pennsylvania We provide polynomial-time approximately optimal [email protected] Bayesian mechanisms for makespan minimization on un- related machines as well as for max-min fair allocations Jamie Morgenstern of indivisible goods,√ with approximation factors of 2 and Carnegie Mellon university min{m − k +1, O˜( k)} respectively, matching the ap- [email protected] proximation ratios of best known polynomial-time algo- rithms. Our mechanisms are obtained by establishing Aaron Roth a polynomial-time approximation-sensitive reduction from University of Pennsylvania, USA the problem of designing approximately optimal mecha- [email protected] nisms for some arbitrary objective O to that of designing bi-criterion approximation algorithms for the same objec- Zhiwei Steven Wu tive O plus a linear allocation cost term. Our reduction is University of Pennsylvania itself enabled by extending the celebrated “equivalence of DA15 Abstracts 59

separation and optimization’[GLS81,KP80] to also accom- [email protected] modate bi-criterion approximations. Giuseppe Italiano Constantinos Daskalakis, Matt Weinberg Universita di Roma “Tor Vergata’ MIT [email protected] [email protected], [email protected] Luigi Laura CP27 “Sapienza’ Universita di Roma [email protected] Tight Bounds on Vertex Connectivity Under Ver- tex Sampling Nikos Parotsidis A fundamental result by Karger (SODA 1994) states that University of Ioannina, Greece for any λ-edge-connected graph with n nodes, indepen- [email protected] dently sampling each edge with probability p =Ω(logn/λ) results in a graph that has edge connectivity Ω(λp), with high probability. This paper proves the analogous result CP27 for vertex connectivity, when sampling vertices. We show Contagious Sets in Expanders that for any k-vertex-connected graph G with n nodes, if each node is independently sampled with probability We consider the following activation process in undirected p =Ω( log n/k), then the subgraph induced by the sam- graphs, often referred to as bootstrap percolation:aver- tex is active either if it belongs to a set of initially acti- pled nodes has vertex connectivity Ω(kp2), with high prob- vated vertices or if at some point it has at least r active ability. This bound improves upon the recent results of neighbors, where r>1 is the activation threshold. A con- Censor-Hillel et al. (SODA 2014), and is existentially op- tagious set is a set whose activation results with the en- timal. tire graph being active. Given a graph G,letm(G, r)be Keren Censor-Hillel the minimal size of a contagious set. We consider graphs Technion, Israel with expansion properties, parameterized by the spectral [email protected] gap and/or the girth. We prove that sufficiently strong m G, ≤ O n expansion properties imply that ( 2) ( d2 )(and more generally, m(G, r) ≤ O( n )). In addition, we Mohsen Ghaffari dr/(r−1) MIT demonstrate that weaker assumptions suffice in order to m G, ≤ O n log d ghaff[email protected] imply that ( 2) ( d2 ). For example, we show this for graphs of girth at least 7, and for graphs with λ(G) < (1 − )d, provided the graph has no 4-cycles. George Giakkoupis INRIA Rennes Amin Coja-Oghlan [email protected] Goethe University, Germany [email protected] Bernhard Haeupler Carnegie Mellon U. Uriel Feige [email protected] Weizmann Institute, Israel [email protected] Fabian Kuhn University of Freiburg, Germany Michael Krivelevich [email protected] Tel Aviv University [email protected] CP27 Daniel Reichman 2-Edge Connectivity in Directed Graphs Cornell University [email protected] Edge and vertex connectivity are fundamental concepts in graph theory. While they have been thoroughly studied in the case of undirected graphs, surprisingly not much CP27 has been investigated for directed graphs. In this paper we study 2-edge connectivity problems in directed graphs and, Fast Generation of Random Spanning Trees and in particular, we consider the computation of the following the Effective Resistance Metric natural relation: We say that two vertices v and w are 2- edge-connected if there are two edge-disjoint paths from v We present a new algorithm for generating a uniformly to w and two edge-disjoint paths from w to v.Thisrelation random spanning tree in an undirected graph. Our algo- 4 o partitions the vertices into blocks such that all vertices in rithm samples such a tree in expected O(m 3 + (1))time. the same block are 2-edge-connected. Differently from the This improves over the best previously known bound of √ ω undirected case, those blocks do not correspond to the 2- min(O(m n),O(n )) – that follows from the work of Kel- edge-connected components of the graph. Our main result ner and Madry [FOCS’09] and of Colbourn et al. [J. Algo- is an algorithm for computing the 2-edge-connected blocks rithms’96] – whenever the input graph is sufficiently sparse. of a directed graph in linear time. Additionally, we also At a high level, our result stems from carefully exploiting show how to compute in linear time a sparse certificate for the interplay of random spanning trees, random walks, and this relation. the notion of effective resistance, as well as from devising a way to algorithmically relate these concepts to the combi- Loukas Georgiadis natorial structure of the graph. This involves, in particular, University of Ioannina, Greece establishing a new connection between the effective resis- 60 DA15 Abstracts

tance metric and the cut structure of the underlying graph.

Aleksander Madry Microsoft Research New England aleksander.madry@epfl.ch

Damian Straszak, Jakub Tarnawski University of Wroclaw [email protected], [email protected]

CP27 Connectivity in Random Forests and Credit Net- works

In this paper, we study uniformly random forests of a graph. We define the RF-connectivity between a pair of nodes in a graph G as the probability that the two nodes belong to the same connected component in a uniformly random forest of G. Our main result is that for an arbi- trary subset S of nodes in G, the average RF-connectivity between pairs of nodes in S is at least 1 − 2/h(GS ), where h(GS ) is the edge expansion of the subgraph GS induced by S. This work is motivated by the study of credit net- works (Dandekar et al. ’11) and the negative correlation conjecture in the study of uniformly random forests.

Ashish Goel Stanford University [email protected]

Sanjeev Khanna University of Pennsylvania [email protected]

Sharath Raghvendra Virginia Tech [email protected]

Hongyang Zhang Stanford University [email protected] Conference Program SODA, ALENEX and ANALCO 2015 61

Speaker Index

62 SODA, ALENEX and ANALCO 2015 Conference Program

A E K Abboud, Amir, CP23, 2:25 Tue Eden, Alon, CP2, 9:25 Sun Kapralov, Michael, CP17, 5:45 Mon Acharya, Jayadev, CP25, 4:55 Tue Eisenbrand, Friedrich, CP5, 2:50 Sun Kaski, Petteri, ALENEX, 3:40 Mon Agrawal, Shipra, CP19, 9:50 Tue Elkin, Michael, CP11, 10:40 Mon Kim, Anthony, CP2, 9:00 Sun Akhremtsev, Yaroslav, ALENEX, 10:15 Eppstein, David, CP3, 9:00 Sun Kiwi, Marcos, ANALCO, 9:50 Sun Sun Esfandiari, Hossein, CP17, 4:30 Mon Ko, Young Kun, CP14, 2:00 Mon Allen-Zhu, Zeyuan, CP19, 10:40 Tue Kociumaka, Tomasz, CP8, 4:30 Sun Annamalai, Chidambaram, CP18, 6:10 F Kontogiannis, Spyros, ALENEX, 5:20 Fekete, Sándor P., ALENEX, 2:00 Mon Mon Mon Feldman, Vitaly, CP7, 5:45 Sun Ashlagi, Itai, CP26, 5:45 Tue Krokhin, Andrei, CP12, 9:50 Mon Fu, Hu, CP2, 10:15 Sun Kulkarni, Janardhan, CP14, 3:40 Mon B Fukunaga, Takuro, CP1, 9:25 Sun Babenko, Maxim, CP8, 5:45 Sun L Bannai, Hideo, CP8, 5:20 Sun G Lam, Jenny, ALENEX, 9:00 Sun Garg, Mohit, CP11, 9:50 Mon Bansal, Nikhil, CP1, 9:00 Sun Lee, Euiwoong, CP12, 9:00 Mon Gelles, Ran, CP18, 4:30 Mon Barba, Luis, CP22, 2:00 Tue Lemiesz, Jakub, ANALCO, 4:55 Sun Georgiadis, Loukas, CP27, 4:55 Tue Barba, Luis, CP22, 3:15 Tue Li, Shi, CP10, 9:25 Mon Ghazi, Badih, CP18, 5:20 Mon Batra, Jatin, CP1, 10:40 Sun Li, Shi, CP15, 2:50 Mon Giotis, Ioannis, ANALCO, 9:25 Sun Bergamini, Elisabetta, ALENEX, 4:55 Lin, Bingkai, CP9, 4:30 Sun Mon Gog, Simon, ALENEX, 9:50 Sun Lin, Maokai, CP18, 5:45 Mon Bertrand, Pierre, ALENEX, 10:40 Sun Gonczarowski, Yannai, CP14, 2:50 Mon Liu, Jingcheng, CP21, 9:00 Tue Bhojanapalli, Srinadh, CP13, 2:25 Mon Goswami, Mayank, CP11, 9:25 Mon Löffler, Maarten, ALENEX, 3:15 Mon Blum, Avrim, IP3, 11:30 Tue Grandoni, Fabrizio, CP1, 9:50 Sun Lokshtanov, Daniel, CP9, 5:20 Sun Bodini, Olivier, ANALCO, 2:00 Sun Gravin, Nick, CP2, 10:40 Sun Lumbroso, Jérémie, ANALCO, 10:15 Boutsidis, Christos, CP13, 2:00 Mon Greenhill, Catherine, CP21, 9:50 Tue Sun Bradonjic, Milan, ANALCO, 2:50 Sun Guo, Zeyu, CP6, 3:15 Sun Buchet, Mickaël, CP3, 9:50 Sun M H Magner, Abram, ANALCO, 3:15 Sun C Harris, David, CP24, 3:40 Tue Mahabadi, Sepideh, CP5, 3:40 Sun Ceccarello, Matteo, ALENEX, 4:30 Mon Hatami, Pooya, CP25, 6:10 Tue Maria, Clément, CP3, 10:15 Sun Censor-Hillel, Keren, CP27, 5:20 Tue Hell, Pavol, CP23, 3:15 Tue Marx, Dániel, CP9, 5:45 Sun Chan, Timothy M., CP4, 2:00 Sun Hendricks, Jacob, CP3, 9:25 Sun Mathieu, Claire, IP2, 11:30 Mon Chekuri, Chandra, CP4, 3:15 Sun Hermon, Jonathan, CP24, 3:15 Tue Miracle, Sarah, CP21, 10:15 Tue Chen, Fei, CP16, 5:20 Mon Huo, Hongwei, ALENEX, 9:25 Sun Mohar, Bojan, CP23, 3:40 Tue Chonev, Ventsislav, CP13, 3:15 Mon Monemizadeh, Morteza, CP17, 4:55 Chuzhoy, Julia, CP4, 3:40 Sun I Mon Ibsen-Jensen, Rasmus, CP14, 3:15 Mon Morgenstern, Jamie, CP26, 4:30 Tue Im, Sungjin, CP15, 2:00 Mon D Moshkovitz, Guy, CP17, 6:10 Mon Dadush, Daniel, CP5, 2:25 Sun Im, Sungjin, CP15, 2:25 Mon Diakonikolas, Ilias, CP7, 5:20 Sun Italiano, Giuseppe, CP11, 10:15 Mon N Diakonikolas, Ilias, CP25, 5:20 Tue Italiano, Giuseppe, ALENEX, 6:10 Mon Natale, Emanuele, CP6, 2:25 Sun Ding, Hu, CP20, 9:25 Tue Iwata, Yoichi, ALENEX, 2:25 Mon Nenadov, Rajko, CP24, 2:00 Tue Disser, Yann, CP12, 10:15 Mon Nikolov, Aleksandar, CP5, 3:15 Sun Dixit, Kashyap, CP25, 4:30 Tue J Norouzi-Fard, Ashkan, CP10, 9:50 Mon Jansen, Bart, CP9, 4:55 Sun Conference Program SODA, ALENEX and ANALCO 2015 63

O T Okajima, Yuzuru, ALENEX, 2:50 Mon Talvitie, Topi L., CP22, 2:25 Tue Ott, Sebastian, CP15, 3:15 Mon Tarnawski, Jakub, CP27, 5:45 Tue P U Pandurangan, Gopal, CP6, 2:50 Sun Umboh, Seeun W., CP19, 9:00 Tue Pensyl, Thomas, CP10, 10:40 Mon Pettie, Seth, CP8, 6:10 Sun V van Stee, Rob, CP19, 10:15 Tue Pokutta, Sebastian, CP12, 9:25 Mon Varadarajan, Kasturi, CP20, 9:00 Tue R Veith, David, ALENEX, 5:45 Mon Rahul, Saladi, CP3, 10:40 Sun Velingker, Ameya, CP18, 4:55 Mon Reichman, Daniel, CP27, 4:30 Tue Vigneron, Antoine, CP22, 2:50 Tue Roychoudhury, Anamitra, CP15, 3:40 Vishnoi, Nisheeth K., CP21, 10:40 Tue Mon Vondrak, Jan, CP12, 10:40 Mon Rubin, Natan, CP20, 10:15 Tue Rubinstein, Aviad, CP7, 4:30 Sun W Waggoner, Bo, CP19, 9:25 Tue Rybicki, Bartosz, CP10, 10:15 Mon Wagner, Stephan, ANALCO, 2:25 Sun S Walczak, Bartosz, CP23, 2:50 Tue Saban, Daniela, CP26, 5:20 Tue Wallner, Michael, ANALCO, 3:40 Sun Sach, Benjamin G., CP8, 4:55 Sun Walter, Michael, CP5, 2:00 Sun Saks, Michael, CP17, 5:20 Mon Wang, Joshua R., CP23, 2:00 Tue Sanita, Laura, CP10, 9:00 Mon Ward, Justin, CP16, 4:30 Mon Schwartz, Roy, CP16, 4:55 Mon Weinberg, Matt, CP26, 6:10 Tue Schwartz, Roy, CP16, 6:10 Mon Wiese, Andreas, CP20, 9:50 Tue Shun, Julian, CP6, 3:40 Sun Wild, Sebastian, ANALCO, 4:30 Sun Sivan, Balasubramanian, CP14, 2:25 Mon Williams, Ryan, CP4, 2:50 Sun Skoric, Nemanja, CP24, 2:25 Tue Winkler, Peter, IP1, 11:30 Sun Skutella, Martin, CP1, 10:15 Sun Woodruff, David, CP13, 2:50 Mon Socala, Arkadiusz, CP9, 6:10 Sun Soto, José, CP26, 4:55 Tue X Xu, Chao, CP22, 3:40 Tue Sousa Pinto, Joao, CP13, 3:40 Mon Sportiello, Andrea, ANALCO, 10:40 Sun Y Srivastava, Piyush, CP21, 9:25 Tue Yao, Andrew C., CP2, 9:50 Sun Stemmer, Uri, CP7, 4:55 Sun Yu, Huacheng, CP4, 2:25 Sun Su, Hsin-Hao, CP6, 2:00 Sun Suk, Andrew, CP20, 10:40 Tue Z Zamir, Or, CP11, 9:00 Mon Suresh, Ananda Theertha, CP25, 5:45 Tue Zhai, Alex, CP24, 2:50 Tue Svensson, Ola, CP16, 5:45 Mon Zhang, Hongyang, CP27, 6:10 Tue Swamy, Chaitanya, CP7, 6:10 Sun Zhao, Jun, ANALCO, 9:00 Sun Hotel Floor Plan