Multiple Optimality Guarantees in Statistical Learning

Total Page:16

File Type:pdf, Size:1020Kb

Multiple Optimality Guarantees in Statistical Learning Multiple Optimality Guarantees in Statistical Learning John Duchi Electrical Engineering and Computer Sciences University of California at Berkeley Technical Report No. UCB/EECS-2014-79 http://www.eecs.berkeley.edu/Pubs/TechRpts/2014/EECS-2014-79.html May 15, 2014 Copyright © 2014, by the author(s). All rights reserved. Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission. Multiple Optimality Guarantees in Statistical Learning by John C Duchi A dissertation submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Computer Science and the Designated Emphasis in Communication, Computation, and Statistics in the Graduate Division of the University of California, Berkeley Committee in charge: Professor Michael I. Jordan, Co-chair Professor Martin J. Wainwright, Co-chair Professor Peter Bickel Professor Laurent El Ghaoui Spring 2014 Multiple Optimality Guarantees in Statistical Learning Copyright 2014 by John C Duchi 1 Abstract Multiple Optimality Guarantees in Statistical Learning by John C Duchi Doctor of Philosophy in Computer Science and the Designated Emphasis in Communication, Computation, and Statistics University of California, Berkeley Professor Michael I. Jordan, Co-chair Professor Martin J. Wainwright, Co-chair Classically, the performance of estimators in statistical learning problems is measured in terms of their predictive ability or estimation error as the sample size n grows. In modern statistical and machine learning applications, however, computer scientists, statisticians, and analysts have a variety of additional criteria they must balance: estimators must be efficiently computable, data providers may wish to maintain anonymity, large datasets must be stored and accessed. In this thesis, we consider the fundamental questions that arise when trading between multiple such criteria—computation, communication, privacy—while maintaining statistical performance. Can we develop lower bounds that show there must be tradeoffs? Can we develop new procedures that are both theoretically optimal and practically useful? To answer these questions, we explore examples from optimization, confidentiality pre- serving statistical inference, and distributed estimation under communication constraints. Viewing our examples through a general lens of constrained minimax theory, we prove fundamental lower bounds on the statistical performance of any algorithm subject to the constraints—computational, confidentiality, or communication—specified. These lower bounds allow us to guarantee the optimality of the new algorithms we develop addressing the addi- tional criteria we consider, and additionally, we show some of the practical benefits that a focus on multiple optimality criteria brings. In somewhat more detail, the central contributions of this thesis include the following: we develop several new stochastic optimization algorithms, applicable to general classes • of stochastic convex optimization problems, including methods that are automatically 2 adaptive to the structure of the underlying problem, parallelize naturally to attain linear speedup in the number of processors available, and may be used asynchronously, prove lower bounds demonstrating the optimality of these methods, • provide a variety of information-theoretic tools—strong data processing inequalities— • useful for proving lower bounds in privacy-preserving statistical inference, communication- constrained estimation, and optimization, develop new algorithms for private learning and estimation, guaranteeing their opti- • mality, and give simple distributed estimation algorithms and prove fundamental limits showing • that they (nearly) optimally trade off between communication (in terms of the number of bits distributed processors may send) and statistical risk. i To Emily ii Contents Contents ii List of Figures v I Introduction and background 1 1 Introduction 2 1.1 Evaluating statistical learning procedures . ........... 2 1.2 Thesisgoalsandcontributions . ..... 5 1.3 Organization of the thesis and previously published work........... 7 1.4 Notation...................................... 8 2 Minimax rates of convergence 11 2.1 Basicframeworkandminimaxrisk . ... 11 2.2 Methods for lower bounds: Le Cam, Assouad, and Fano . ...... 13 2.3 Summary ..................................... 22 2.4 Proofsofresults................................. 22 II Optimization 25 3 Stochasticoptimizationandadaptivegradientmethods 26 3.1 Stochasticoptimizationalgorithms . ....... 27 3.2 Adaptiveoptimization ............................. 32 3.3 Afewoptimalityguarantees . ... 35 3.4 Summary ..................................... 38 3.5 Proofsofconvergenceandminimaxbounds. ...... 39 4 Data sparsity, asynchrony, and faster stochastic optimization 47 4.1 Problemsetting.................................. 47 4.2 Parallel and asynchronous optimization with sparsity . ............ 48 4.3 Experiments.................................... 53 iii 4.4 Proofsofconvergence. .. 56 5 Randomizedsmoothingforstochasticoptimization 66 5.1 Introduction.................................... 66 5.2 Mainresultsandsomeconsequences . .... 68 5.3 Applicationsandexperimentalresults . ....... 74 5.4 Summary ..................................... 80 5.5 Proofsofconvergence. .. 81 5.6 Propertiesofrandomizedsmoothing. ...... 88 6 Zero-order optimization: the power of two function evaluations 97 6.1 Introduction.................................... 97 6.2 Algorithms..................................... 99 6.3 Lowerboundsonzero-orderoptimization . ....... 108 6.4 Summary ..................................... 109 6.5 Convergenceproofs ............................... 110 6.6 Proofsoflowerbounds ............................. 116 6.7 Technical results for convergence arguments . ......... 121 6.8 Technical proofs associated with lower bounds . ......... 128 III Privacy 130 7 Privacy, minimax rates of convergence, and data processing inequalities 131 7.1 Introduction.................................... 131 7.2 Backgroundandproblemformulation . ..... 135 7.3 Pairwise bounds under privacy: Le Cam and local Fano methods....... 137 7.4 Mutual information under local privacy: Fano’s method . ........... 142 7.5 Bounds on multiple pairwise divergences: Assouad’s method ......... 149 7.6 Comparisontorelatedwork . 157 7.7 Summary ..................................... 161 8 Technicalargumentsforprivateestimation 162 8.1 ProofofTheorem7.1andrelatedresults . ...... 162 8.2 ProofofTheorem7.2andrelatedresults . ...... 169 8.3 ProofofTheorem7.3............................... 172 8.4 Proofs of multi-dimensional mean-estimation results . ............. 174 8.5 Proofsofmultinomialestimationresults . ........ 180 8.6 Proofsofdensityestimationresults . ....... 182 8.7 Informationbounds............................... 187 iv IV Communication 192 9 Communication efficient algorithms 193 9.1 Introduction.................................... 194 9.2 BackgroundandProblemSet-up. 195 9.3 TheoreticalResults .............................. 197 9.4 Summary ..................................... 204 9.5 ProofofTheorem9.1............................... 204 10Optimalityguaranteesfordistributedestimation 207 10.1Introduction................................... 207 10.2Problemsetting................................. 208 10.3RelatedWork ................................... 210 10.4Mainresults.................................... 211 10.5 Consequencesforregression . ..... 218 10.6Summary ..................................... 220 10.7 Proofoutlineofmajorresults . ..... 221 10.8 Techniques,tools,andsetupforproofs . ........ 223 10.9 Proofs of lower bounds for independent protocols . .......... 228 10.10Proofs of interactive lower bounds for Gaussian observations . 236 Bibliography 242 v List of Figures 2.1 Example of a 2δ-packing............................... 15 4.1 ExperimentswithURLdata. .. 54 4.2 Stepsize sensitivity of AdaGrad .......................... 54 4.3 Click-through prediction performance of asynchronous methods ......... 55 5.1 Iterations to optimality versus gradient samples . ............. 77 5.2 Metriclearningoptimizationerror. ......... 79 5.3 Necessityofsmoothing ............................. ... 80 7.1 Graphicalstructureofprivatechannels . .......... 133 7.2 Privatesamplingstrategies. ....... 146 8.1 Densityconstructionsforlowerbounds . ......... 182 10.1 GraphicalmodelforLemma10.1 . ..... 225 vi Acknowledgments There are so many people to whom I owe credit for this thesis that I must begin with an apology: I will probably forget to mention several of you in the coming paragraphs. If I do, please forgive me, and let me know and I will be happy to buy you a beer. My acknowledgments must begin with my advisors, my two official advisors at Berkeley, and my one surrogate advisor down at Google: Michael Jordan, Martin Wainwright, and Yoram Singer. It has become clear to me that having three advisors was a necessary thing, if only for their sakes, because it kept me out of the hair of the other two while I bothered the third. More seriously, Mike and Martin have pushed me into contact with a multitude of disciplines, encouraging and exemplifying a fearlessness and healthy disrespect for academic boundaries.
Recommended publications
  • Session Types and Game Semantics a Synchronous Side and an Asynchronous Side
    Two Sides of the Same Coin: Session Types and Game Semantics A Synchronous Side and an Asynchronous Side SIMON CASTELLAN, Imperial College London, United Kingdom NOBUKO YOSHIDA, Imperial College London, United Kingdom Game semantics and session types are two formalisations of the same concept: message-passing open programs following certain protocols. Game semantics represents protocols as games, and programs as strategies; while session types specify protocols, and well-typed π-calculus processes model programs. Giving faithful models of the π-calculus and giving a precise description of strategies as a programming language are two difficult 27 problems. In this paper, we show how these two problems can be tackled at the same time by building an accurate game semantics model of the session π-calculus. Our main contribution is to fill a semantic gap between the synchrony of the (session) π-calculus and the asynchrony of game semantics, by developing an event-structure based game semantics for synchronous concurrent computation. This model supports the first truly concurrent fully abstract (for barbed congruence) interpretation of the synchronous (session) π-calculus. We further strengthen this correspondence, establishing finite definability of asynchronous strategies by the internal session π-calculus. As an application of these results, we propose a faithful encoding of synchronous strategies into asynchronous strategies by call-return protocols, which induces automatically an encoding at the level of processes. Our results bring session types and game semantics into the same picture, proposing the session calculus as a programming language for strategies, and strategies as a very accurate model of the session calculus.
    [Show full text]
  • Learning Under Limited Information∗
    Learning Under Limited Information¤ Yan Chen School of Information, University of Michigan 550 East University Avenue, Ann Arbor, MI 48109-1092 Yuri Khoroshilov Department of Finance, University of Michigan Business School 701 Tappan Street, Ann Arbor, MI 48109-1234 August 28, 2002 Running title: learning under limited information Address for manuscript correspondence: Yan Chen School of Information, University of Michigan 550 East University Avenue, Ann Arbor, MI 48109-1092 Email: [email protected] Phone: (734) 763-2285 Fax: (734) 764-2475 ¤We thank Julie Cullen, Ido Erev, Dan Friedman, Eric Friedman, Drew Fudenburg, Amy Greenwald, Teck-Hua Ho, Lutz Kilian, David Lucking-Reiley, Atanasios Mitropoulos, Tom Palfrey, Rajiv Sarin, Scott Shenker, Farshid Vahid and seminar participants at Michigan, Mississippi, Rutgers, Texas A&M, the 1998 Econometric Society meetings (Chicago), ESA 2000 (New York City) and the 2001 Econometric Society meetings (New Orleans) for discussions and comments. We are grateful to John Van Huyck for providing the data and for helping us with part of the estimation. Emre Sucu provided excellent research assistance. We are grateful to two anonymous referees for their constructive comments which significantly improved the paper. The research support provided by NSF grant SBR-9805586 and SES-0079001 to Chen and the Deutsche Forschungsgemeinschaft through SFB303 at Universit¨atBonn are gratefully acknowledged. Any remaining errors are our own. 1 Abstract We study how human subjects learn under extremely limited information. We use Chen’s (forthcoming) data on cost sharing games, and Van Huyck, Battalio and Rankin’s (1996) data on coordination games to compare three payoff-based learning models.
    [Show full text]
  • Evolutionary Game Theory: a Renaissance
    games Review Evolutionary Game Theory: A Renaissance Jonathan Newton ID Institute of Economic Research, Kyoto University, Kyoto 606-8501, Japan; [email protected] Received: 23 April 2018; Accepted: 15 May 2018; Published: 24 May 2018 Abstract: Economic agents are not always rational or farsighted and can make decisions according to simple behavioral rules that vary according to situation and can be studied using the tools of evolutionary game theory. Furthermore, such behavioral rules are themselves subject to evolutionary forces. Paying particular attention to the work of young researchers, this essay surveys the progress made over the last decade towards understanding these phenomena, and discusses open research topics of importance to economics and the broader social sciences. Keywords: evolution; game theory; dynamics; agency; assortativity; culture; distributed systems JEL Classification: C73 Table of contents 1 Introduction p. 2 Shadow of Nash Equilibrium Renaissance Structure of the Survey · · 2 Agency—Who Makes Decisions? p. 3 Methodology Implications Evolution of Collective Agency Links between Individual & · · · Collective Agency 3 Assortativity—With Whom Does Interaction Occur? p. 10 Assortativity & Preferences Evolution of Assortativity Generalized Matching Conditional · · · Dissociation Network Formation · 4 Evolution of Behavior p. 17 Traits Conventions—Culture in Society Culture in Individuals Culture in Individuals · · · & Society 5 Economic Applications p. 29 Macroeconomics, Market Selection & Finance Industrial Organization · 6 The Evolutionary Nash Program p. 30 Recontracting & Nash Demand Games TU Matching NTU Matching Bargaining Solutions & · · · Coordination Games 7 Behavioral Dynamics p. 36 Reinforcement Learning Imitation Best Experienced Payoff Dynamics Best & Better Response · · · Continuous Strategy Sets Completely Uncoupled · · 8 General Methodology p. 44 Perturbed Dynamics Further Stability Results Further Convergence Results Distributed · · · control Software and Simulations · 9 Empirics p.
    [Show full text]
  • A Distributed Auctioneer for Resource Allocation in Decentralized Systems
    A Distributed Auctioneer for Resource Allocation in Decentralized Systems Amin M. Khan Xavier Vila¸ca Universitat Polit`ecnicade Catalunya, Barcelona INESC-ID Lisboa [email protected] [email protected] Lu´ısRodrigues Felix Freitag [email protected] Universitat Polit`ecnicade Catalunya, Barcelona [email protected] September 29, 2021 Abstract In decentralized systems, nodes often need to coordinate to access shared resources in a fair manner. One approach to perform such arbitration is to rely on auction mechanisms. Although there is an extensive literature that studies auctions, most of these works assume the existence of a central, trusted auctioneer. Unfortunately, in fully decentralized systems, where the nodes that need to cooperate operate under separate spheres of control, such central trusted entity may not exist. Notable examples of such decentralized systems include community networks, clouds of clouds, cooperative nano data centres, among others. In this paper, we make theoretical and practical contributions to distribute the role of the auctioneer. From the theoretical perspective, we propose a framework of distributed simulations of the auctioneer that are Nash equilibria resilient to coalitions and asynchrony. From the practical perspective, our protocols leverage the distributed nature of the simulations to parallelise the execution. We have implemented a prototype that instantiates the framework for bandwidth allocation in community networks, and evaluated it in a real distributed setting. 1 Introduction Most distributed systems have limited resources that need to be shared by many nodes. For instance, in a network there is limited bandwidth that needs to be allocated to multiple nodes. In cloud applications, virtual machines (VMs) need to be allocated to different cloud users.
    [Show full text]
  • Advances and Open Problems in Federated Learning
    Advances and Open Problems in Federated Learning Peter Kairouz7* H. Brendan McMahan7∗ Brendan Avent21 Aurelien´ Bellet9 Mehdi Bennis19 Arjun Nitin Bhagoji13 Kallista Bonawitz7 Zachary Charles7 Graham Cormode23 Rachel Cummings6 Rafael G.L. D’Oliveira14 Hubert Eichner7 Salim El Rouayheb14 David Evans22 Josh Gardner24 Zachary Garrett7 Adria` Gascon´ 7 Badih Ghazi7 Phillip B. Gibbons2 Marco Gruteser7;14 Zaid Harchaoui24 Chaoyang He21 Lie He 4 Zhouyuan Huo 20 Ben Hutchinson7 Justin Hsu25 Martin Jaggi4 Tara Javidi17 Gauri Joshi2 Mikhail Khodak2 Jakub Konecnˇ y´7 Aleksandra Korolova21 Farinaz Koushanfar17 Sanmi Koyejo7;18 Tancrede` Lepoint7 Yang Liu12 Prateek Mittal13 Mehryar Mohri7 Richard Nock1 Ayfer Ozg¨ ur¨ 15 Rasmus Pagh7;10 Hang Qi7 Daniel Ramage7 Ramesh Raskar11 Mariana Raykova7 Dawn Song16 Weikang Song7 Sebastian U. Stich4 Ziteng Sun3 Ananda Theertha Suresh7 Florian Tramer` 15 Praneeth Vepakomma11 Jianyu Wang2 Li Xiong5 Zheng Xu7 Qiang Yang8 Felix X. Yu7 Han Yu12 Sen Zhao7 1Australian National University, 2Carnegie Mellon University, 3Cornell University, 4Ecole´ Polytechnique Fed´ erale´ de Lausanne, 5Emory University, 6Georgia Institute of Technology, 7Google Research, 8Hong Kong University of Science and Technology, 9INRIA, 10IT University of Copenhagen, 11Massachusetts Institute of Technology, 12Nanyang Technological University, 13Princeton University, 14Rutgers University, 15Stanford University, 16University of California Berkeley, 17 University of California San Diego, 18University of Illinois Urbana-Champaign, 19University of Oulu, 20University of Pittsburgh, 21University of Southern California, 22University of Virginia, 23University of Warwick, 24University of Washington, 25University of Wisconsin–Madison arXiv:1912.04977v3 [cs.LG] 9 Mar 2021 Abstract Federated learning (FL) is a machine learning setting where many clients (e.g. mobile devices or whole organizations) collaboratively train a model under the orchestration of a central server (e.g.
    [Show full text]
  • Asynchronous Best-Reply Dynamics
    Asynchronous Best-Reply Dynamics Noam Nisan1, Michael Schapira2, and Aviv Zohar2 1 Google Tel-Aviv and The School of Computer Science and Engineering, The Hebrew University of Jerusalem, Israel. 2 The School of Computer Science and Engineering, The Hebrew University of Jerusalem, Israel. fnoam,mikesch,[email protected] Abstract. In many real-world settings (e.g., interdomain routing in the Internet) strategic agents are instructed to follow best-reply dynamics in asynchronous environments. In such settings players learn of each other's actions via update messages that can be delayed or even lost. In particular, several players might update their actions simultaneously, or make choices based on outdated information. In this paper we analyze the convergence of best- (and better-)reply dynamics in asynchronous environments. We provide su±cient conditions, and necessary conditions for convergence in such settings, and also study the convergence-rate of these natural dynamics. 1 Introduction Many real-life protocols can be regarded as executions of best-reply dynamics, i.e, players (computational nodes) are instructed to repeatedly best-reply to the actions of other players. In many cases, like Internet settings, this occurs in asynchronous environments: Think of the players as residing in a computer network, where their best-replies are transmitted to other players and serve as the basis for the other players' best-replies. These update messages that players send to each other may be delayed or even lost, and so players may update their actions simultaneously, and do so based on outdated information. Perhaps the most notable example for this is the Border Gateway Protocol (BGP) that handles interdomain routing in the Internet.
    [Show full text]
  • Synchronous and Asynchronous Learning by Responsive Learning
    Synchronous and Asynchronous Learning by Resp onsive Learning Automata Eric J Friedman Department of Economics Rutgers University New Brunswick NJ friedmanfaseconrutgersedu Scott Shenker Xerox PARC Coyote Hill Road Palo Alto CA shenkerparcxeroxcom Novemb er Abstract We consider the ability of economic agents to learn in a decentralized envi ronment in which agents do not know the sto chastic payo matrix and can not observe their opp onents actions they merely know at each stage of the game their own action and the resulting payo We discuss the requirements for learning in such an environment and show that a simple probabilistic learning algorithm satises two imp ortant optimizing prop erties i When placed in an unknown but eventually stationary random environment they converge in b ounded time in a sense we make precise to strategies that maximize average payo ii They satisfy a monotonicity prop erty related to the law of the eect in which increasing the payos for a given strategy increases the probability of that strategy b eing played in the future We then study how groups of such learners interact in a general game We show that synchronous groups of these learners converge to the serially undominated set In contrast asynchronous groups do not necessarily converge to the serially undominated set highlighting the imp ortance of timing in decentralized learning However these asynchronous learners do converge to the serially unoverwhelmed set which we dene Thus the serially unoverwhelmed set can b e viewed as an appropriate solution
    [Show full text]
  • Conflict Resolution Through Strategic Vocal Negotiation and Cooperation in Nestling Barn Owls (Tytoalba)
    Unicentre CH-1015 Lausanne http://serval.unil.ch Year : 2019 CONFLICT RESOLUTION THROUGH STRATEGIC VOCAL NEGOTIATION AND COOPERATION IN NESTLING BARN OWLS (TYTOALBA) Ducouret Pauline Ducouret Pauline, 2019, CONFLICT RESOLUTION THROUGH STRATEGIC VOCAL NEGOTIATION AND COOPERATION IN NESTLING BARN OWLS (TYTOALBA) Originally published at : Thesis, University of Lausanne Posted at the University of Lausanne Open Archive http://serval.unil.ch Document URN : urn:nbn:ch:serval-BIB_1FCB6DC32ADB8 Droits d’auteur L'Université de Lausanne attire expressément l'attention des utilisateurs sur le fait que tous les documents publiés dans l'Archive SERVAL sont protégés par le droit d'auteur, conformément à la loi fédérale sur le droit d'auteur et les droits voisins (LDA). A ce titre, il est indispensable d'obtenir le consentement préalable de l'auteur et/ou de l’éditeur avant toute utilisation d'une oeuvre ou d'une partie d'une oeuvre ne relevant pas d'une utilisation à des fins personnelles au sens de la LDA (art. 19, al. 1 lettre a). A défaut, tout contrevenant s'expose aux sanctions prévues par cette loi. Nous déclinons toute responsabilité en la matière. Copyright The University of Lausanne expressly draws the attention of users to the fact that all documents published in the SERVAL Archive are protected by copyright in accordance with federal law on copyright and similar rights (LDA). Accordingly it is indispensable to obtain prior consent from the author and/or publisher before any use of a work or part of a work for purposes other than personal use within the meaning of LDA (art.
    [Show full text]
  • Distributed Computing with Adaptive Heuristics
    Distributed Computing with Adaptive Heuristics To be presented at Innovations in Computer Science 2011 Aaron D. Jaggard1;2∗ Michael Schapira3y Rebecca N. Wright1z 1Rutgers University, New Brunswick, NJ 08854, USA 2Colgate University, Hamilton, NY 13346, USA 3Princeton University, Princeton, NJ 08544,USA [email protected] [email protected] [email protected] Abstract: We use ideas from distributed computing to study dynamic environments in which com- putational nodes, or decision makers, follow adaptive heuristics [16], i.e., simple and unsophisticated rules of behavior, e.g., repeatedly “best replying” to others’ actions, and minimizing “regret”, that have been extensively studied in game theory and economics. We explore when convergence of such simple dynamics to an equilibrium is guaranteed in asynchronous computational environments, where nodes can act at any time. Our research agenda, distributed computing with adaptive heuristics, lies on the bor- derline of computer science (including distributed computing and learning) and game theory (including game dynamics and adaptive heuristics). We exhibit a general non-termination result for a broad class of heuristics with bounded recall—that is, simple rules of behavior that depend only on recent history of interaction between nodes. We consider implications of our result across a wide variety of interesting and timely applications: game theory, circuit design, social networks, routing and congestion control. We also study the computational and communication complexity of asynchronous dynamics and present some basic observations regarding the effects of asynchrony on no-regret dynamics. We believe that our work opens a new avenue for research in both distributed computing and game theory.
    [Show full text]
  • Econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible
    A Service of Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics Feldhaus, Christoph; Rockenbach, Bettina; Zeppenfeld, Christopher Conference Paper Inequality in minimum-effort coordination Beiträge zur Jahrestagung des Vereins für Socialpolitik 2020: Gender Economics Provided in Cooperation with: Verein für Socialpolitik / German Economic Association Suggested Citation: Feldhaus, Christoph; Rockenbach, Bettina; Zeppenfeld, Christopher (2020) : Inequality in minimum-effort coordination, Beiträge zur Jahrestagung des Vereins für Socialpolitik 2020: Gender Economics, ZBW - Leibniz Information Centre for Economics, Kiel, Hamburg This Version is available at: http://hdl.handle.net/10419/224650 Standard-Nutzungsbedingungen: Terms of use: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu Inequality in minimum-effort coordination∗ Christoph Feldhaus Bettina Rockenbach Christopher Zeppenfeld March 1, 2020 Abstract Successful coordination is key for economic and societal wealth.
    [Show full text]
  • 15 Dynamics at the Boundary of Game Theory and Distributed Computing
    Dynamics at the Boundary of Game Theory and Distributed Computing AARON D. JAGGARD, U.S. Naval Research Laboratory 15 NEIL LUTZ, Rutgers University MICHAEL SCHAPIRA, Hebrew University of Jerusalem REBECCA N. WRIGHT, Rutgers University We use ideas from distributed computing and game theory to study dynamic and decentralized environments in which computational nodes, or decision makers, interact strategically and with limited information. In such environments, which arise in many real-world settings, the participants act as both economic and computa- tional entities. We exhibit a general non-convergence result for a broad class of dynamics in asynchronous settings. We consider implications of our result across a wide variety of interesting and timely applications: game dynamics, circuit design, social networks, Internet routing, and congestion control. We also study the computational and communication complexity of testing the convergence of asynchronous dynamics. Our work opens a new avenue for research at the intersection of distributed computing and game theory. CCS Concepts: • Theory of computation → Distributed computing models; Convergence and learn- ing in games;•Networks → Routing protocols; Additional Key Words and Phrases: Adaptive heuristics, self stabilization, game dynamics ACM Reference format: Aaron D. Jaggard, Neil Lutz, Michael Schapira, and Rebecca N. Wright. 2017. Dynamics at the Boundary of Game Theory and Distributed Computing. ACM Trans. Econ. Comput. 5, 3, Article 15 (August 2017), 20 pages. https://doi.org/10.1145/3107182 A preliminary version of some of this work appeared in Proceedings of the Second Symposium on Innovations in Computer Science (ICS 2011) [25]. This work was partially supported by NSF grant CCF-1101690, ISF grant 420/12, the Israeli Center for Research Excellence in Algorithms (I-CORE), and the Office of Naval Research.
    [Show full text]
  • Transformational Outcomes of Civil Society Organizations
    tnmgnMda oita a rO nag i az t i no dna nemeganaM t nayeics v wheylaa sehtehT i anas l cwohsesy ivi cosl i sniagyte - o t l aA s ojnsegrla nctacaeler l inhcetsaecnave ac jdnassergorpl bo l sse DD TRANSFORMATIONAL nmopen arentluserhtworg saercednit tnemyolpmegni 812 rh hsene fiungivel-setar vae lfilufnugni ihwsdeendel erahc / s s e r eVf e s z óJ ybderevocnuerarodezitenomebotdrah 6102 OUTCOMES OF CIVIL oltkahuh A eogisahcrup rewopgni lA. tekramhguoht l go i c htrow sdnamed samehtredisnoctonseod sdnamed htrow esehtteemot elbasiyteicoslivic,gniusrup esehtteemot SOCIETY ugiamnzngo leshguorhT.sdeen f lufgninaemgnizinagro- ermnlaicstbeaeeti,krow , i coshtobsetarenegt i a tekramdnal ORGANIZATIONS egsceivla pea u av eul ipmesa, r i ac vel seggusecnedi t s . lanoitamrofsnart'seitinummoclacoL musorez-nonehtswollof scimanyd musorez-nonehtswollof fogniniagselbanehcihw,hcaorppa J feszó ssereV e nco yootanirewopme i tuagn fymono tnecmor r a l i dez hginteteuirtoda mtsys smet tnocdna, r ehtgnidnetxeotsetubi wllankotNeaeroelaicos i a agnikrowteN.esabecruoserl l swol oe anosocto aoalloc l taroba radnuobssorcotnoi fosei es ngnsoita a o citrap r t i uc l a ror nag i az t i enegdnasno r a t wense .noititepmocdnanoitarepoocfoscitcelaid ubsn bn ngo e'sreetnulovehT les' f inagro- z rbgni tuobasgni firrfoy nwr,lao taicossana anoi l gniworg, tfiorproftonyl yteicsgelok miaydteiro i omsimanyddetne lwonkf cosegde i e t -y snoitinfiedruoegnahcyllacitsardyamdna .evitcudorpgniebdnaboj afo .evitcudorpgniebdnaboj SNOITAZINAGRO YTEICOSLIVICFO SEMOCTUO LANOITAMROFSART SEMOCTUO YTEICOSLIVICFO SNOITAZINAGRO B I NBS 807-06-259-879 5-6 ( p r i n t de ) SSENISUB + B I NBS 8-5807-06-259-879 ( dp f ) E MONOC Y S I NSS - L 9971 - 4394 S I NSS 9971 - 4394 ( p r i n t de ) TRA + S I NSS 9971 - 2494 ( dp f ) NGISED + R C HCRA I T TCE ERU ytisrevinUotlaA nisufo oohcS ol uBf s i en s s ECNEICS + eeaadaoita a rO nag i az t i nemeganaMdnano t T E ONHC L GO Y www .
    [Show full text]