Network Models and Algorithms for Strategic and Economic Reasoning Michael Kearns Computer and Information Science University of Pennsylvania World Congress.

Slides:



Advertisements
Similar presentations
Networked Trade: Theory and Behavior Networked Life CIS 112 Spring 2009 Prof. Michael Kearns.
Advertisements

Continuation Methods for Structured Games Ben Blum Christian Shelton Daphne Koller Stanford University.
1 Some Comments on Sebastiani et al Nature Genetics 37(4)2005.
1 Greedy Forwarding in Dynamic Scale-Free Networks Embedded in Hyperbolic Metric Spaces Dmitri Krioukov CAIDA/UCSD Joint work with F. Papadopoulos, M.
A Network Formation Game for Bipartite Exchange Economies [Even-Dar, K. & Suri]
Nash Equilibria In Graphical Games On Trees Edith Elkind Leslie Ann Goldberg Paul Goldberg.
6.853: Topics in Algorithmic Game Theory Fall 2011 Constantinos Daskalakis Lecture 13.
Course: Applications of Information Theory to Computer Science CSG195, Fall 2008 CCIS Department, Northeastern University Dimitrios Kanoulas.
Exact Inference in Bayes Nets
Dynamic Bayesian Networks (DBNs)
Algorithms for solving two- player normal form games Tuomas Sandholm Carnegie Mellon University Computer Science Department.
Combinatorial Algorithms for Market Equilibria Vijay V. Vazirani.
Models of Network Formation Networked Life NETS 112 Fall 2013 Prof. Michael Kearns.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Algorithms for Convex Programs (Capturing Market Equilibria.
An Introduction to Variational Methods for Graphical Models.
Introduction to Belief Propagation and its Generalizations. Max Welling Donald Bren School of Information and Computer and Science University of California.
EE462 MLCV Lecture Introduction of Graphical Models Markov Random Fields Segmentation Tae-Kyun Kim 1.
6.896: Topics in Algorithmic Game Theory Lecture 15 Constantinos Daskalakis.
1 Evolution of Networks Notes from Lectures of J.Mendes CNR, Pisa, Italy, December 2007 Eva Jaho Advanced Networking Research Group National and Kapodistrian.
Exchange Economies on Networks Networked Life CSE 112 Spring 2006 Prof. Michael Kearns.
Approximate Counting via Correlation Decay Pinyan Lu Microsoft Research.
Temporal Action-Graph Games: A New Representation for Dynamic Games Albert Xin Jiang University of British Columbia Kevin Leyton-Brown University of British.
Economic Exchange on Networks Networked Life CSE 112 Spring 2007 Prof. Michael Kearns.
Strategic Models of Network Formation Networked Life CIS 112 Spring 2010 Prof. Michael Kearns.
Graphical Models for Strategic and Economic Reasoning Michael Kearns Computer and Information Science University of Pennsylvania BNAIC 2003 Joint work.
Dynamic Spectrum Management: Optimization, game and equilibrium Tom Luo (Yinyu Ye) December 18, WINE 2008.
Trading in Networks Networked Life CIS 112 Spring 2010 Prof. Michael Kearns.
Algorithmic Game Theory and Internet Computing Amin Saberi Stanford University Computation of Competitive Equilibria.
Graphical Models for Game Theory Undirected graph G capturing local interactions Each player represented by a vertex N_i(G) = neighbors.
Economic Models of Network Formation Networked Life CIS 112 Spring 2008 Prof. Michael Kearns.
Pure Nash Equilibria: Complete Characterization of Hard and Easy Graphical Games Albert Xin Jiang U. of British Columbia MohammadAli Safari Sharif U. of.
Graphical Models Michael Kearns Michael L. Littman Satinder Signh Presenter: Shay Cohen.
Exchange Economies and Networks Networked Life CSE 112 Spring 2005 Prof. Michael Kearns.
Graphical Games Kjartan A. Jónsson. Nash equilibrium Nash equilibrium Nash equilibrium N players playing a dominant strategy is a Nash equilibrium N players.
Computing Equilibria Christos H. Papadimitriou UC Berkeley “christos”
Economic Models of Network Formation Networked Life CSE 112 Spring 2006 Prof. Michael Kearns.
"Why We're So Nice: We're Wired to Cooperate" Natalie Angier New York Times, 23 July 2002 What feels as good as chocolate on the tongue or money in the.
Graphical Models for Game Theory by Michael Kearns, Michael L. Littman, Satinder Singh Presented by: Gedon Rosner.
News and Notes 4/8 HW3 due date delayed to Tuesday 4/13 –will hand out HW4 on 4/13 also Today: finish up NW economics Tuesday 4/13 –another mandatory class.
1 Trends in Mathematics: How could they Change Education? László Lovász Eötvös Loránd University Budapest.
The Erdös-Rényi models
Optimization Based Modeling of Social Network Yong-Yeol Ahn, Hawoong Jeong.
Computing Equilibria Christos H. Papadimitriou UC Berkeley “christos”
Computer Science, Economics, and the Effects of Network Structure
History-Dependent Graphical Multiagent Models Quang Duong Michael P. Wellman Satinder Singh Computer Science and Engineering University of Michigan, USA.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Extending General Equilibrium Theory to the Digital Economy.
Undirected Models: Markov Networks David Page, Fall 2009 CS 731: Advanced Methods in Artificial Intelligence, with Biomedical Applications.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Algorithms for Convex Programs (Capturing Market Equilibria.
October Large networks: a new language for science László Lovász Eötvös Loránd University, Budapest
Trading in Networks: I. Model Prof. Michael Kearns Networked Life MKSE 112 Fall 2012.
An Introduction to Variational Methods for Graphical Models
Approximate Inference: Decomposition Methods with Applications to Computer Vision Kyomin Jung ( KAIST ) Joint work with Pushmeet Kohli (Microsoft Research)
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Primal-Dual Algorithms for Rational Convex Programs II: Dealing with Infeasibility.
2) Combinatorial Algorithms for Traditional Market Models Vijay V. Vazirani.
Probabilistic Graphical Models seminar 15/16 ( ) Haim Kaplan Tel Aviv University.
Networked Games: Coloring, Consensus and Voting Prof. Michael Kearns Networked Life NETS 112 Fall 2013.
Exact Inference in Bayes Nets. Notation U: set of nodes in a graph X i : random variable associated with node i π i : parents of node i Joint probability:
6.4 Random Fields on Graphs 6.5 Random Fields Models In “Adaptive Cooperative Systems” Summarized by Ho-Sik Seok.
MAIN RESULT: We assume utility exhibits strategic complementarities. We show: Membership in larger k-core implies higher actions in equilibrium Higher.
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani 3) New Market Models, Resource Allocation Markets.
04/21/2005 CS673 1 Being Bayesian About Network Structure A Bayesian Approach to Structure Discovery in Bayesian Networks Nir Friedman and Daphne Koller.
Ch 6. Markov Random Fields 6.1 ~ 6.3 Adaptive Cooperative Systems, Martin Beckerman, Summarized by H.-W. Lim Biointelligence Laboratory, Seoul National.
Chapter 12. Probability Reasoning Fall 2013 Comp3710 Artificial Intelligence Computing Science Thompson Rivers University.
A Brief Introduction to Bayesian networks
Trading in Networks: I. Model
Non-additive Security Games
Structured Models for Multi-Agent Interactions
Markov Random Fields Presented by: Vladan Radosavljevic.
Normal Form (Matrix) Games
Presentation transcript:

Network Models and Algorithms for Strategic and Economic Reasoning Michael Kearns Computer and Information Science University of Pennsylvania World Congress of the Game Theory Society Marseille, July 2004 Joint work with: Sham Kakade, John Langford, Michael Littman, Luis Ortiz, Robin Pemantle, Satinder Singh, Siddarth Suri

International Trade [Krempel&Pleumper ]

Corporate Partnerships [Krebs ]

Gnutella

Internet Routers

Artist Mark Lombardi

Sources of Structured Interaction in Social and Economic Analysis Trade agreements and restrictions Social relationships between business people Reporting and organizational structure in a firm Regulatory restrictions on Wall Street Shared influences within an industry or sector Geographical dispersion of parties Physical connectivity (Internet, peer-to-peer) Want to represent such constraints in the interests of accurate modeling, but…

Computational benefits of modeling network structure? –improved algorithms for equilibrium computation –distributed, local computation and adaptation –richer kinds of “conditional equilibrium” queries How does network topology influence outcomes? –local interaction vs. global equilibrium –coalitions, correlations and network structure –price variation, wealth distribution and network structure What kinds of networks to examine? –social network theory and generative models –small worlds, preferential attachment and random graphs A rich blend of ideas and methods from economics, computer science, artificial intelligence, statistics, and mathematics

Outline Graphical Games and the NashProp Algorithm –[K., Littman & Singh 01]; [Ortiz & K. 02] Correlated Equilibria, Graphical Games, and Markov Networks –[Kakade, K. & Langford 03] Graphical Economics and Social Network Theory –[Kakade, K. & Ortiz 04]; [Kakade, K., Ortiz, Pemantle, Suri 04]

Graphical Games and NashProp

Graphical Models for Game Theory Alternative to normal form, which grows exp(n) for n players Undirected graph G capturing local (strategic) interactions Each player represented by a vertex N_i(G) : neighbors of i in G (includes i) Assume: Payoffs expressible as M_i(a) where a over only N_i(G) Graphical game: (G,{M_i}) Compact representation of game Exponential in max degree (<< # of players) Must still look for special structure for efficient computations Related models: [Koller & Milch 01] [La Mura 00]

The NashProp Algorithm Local message-passing; tables of “conditional” Nash equilibria Approximate (all NE) and exact (one NE) versions Provably efficient (polynomial in representation size) for trees NashProp: generalization to arbitrary topology Close relationship to network models for probabilistic reasoning – Bayesian networks, Markov networks –junction tree algorithm, belief propagation U1U2U3 W V T(w,v) = 1  an “upstream” Nash where V = v given W = w  u: T(v,u_i) = 1 for all i, and v is a best response to u,w

Table dimensions are probability of playing 0 Black shows T(v,u) = 1 Ms want to match, Os to unmatch Relative value modulated by parent values  =  0.01,  = 0.05

Experimental Performance number of players computation time

Correlated Equilibria, Graphical Games and Markov Networks

Advantages of Correlated Equilibria Technical: –Easier to compute: linear feasibility formulation –Efficient for 2-player, multi-action case Conceptual: –Correlated actions a fact of the real world –Allows “cooperation via correlation” –Modeling of shared exogenous influences –Enlarged solution space: all mixtures of NE, and more –New (non-Nash) outcomes emerge, often natural ones –Avoid quagmire of full cooperation and coalitions –Natural convergence notion for “greedy” learning But how do we represent an arbitrary CE? –First, only seek to find CE up to (expected) payoff equivalence –Second, look to graphical models for probabilistic reasoning!

Graphical Games and Markov Networks Let G be the graph of a graphical game (strategic structure) Consider the Markov network MN(G): –Form cliques of the local neighborhoods of G –Introduce potential function  c on each clique c –Joint distribution P(a) = (1/Z)  c  c(a) –Markov networks common in AI, statistics, physics (Ising model),… Theorem: For any game with graph G, and any CE of this game, there is a CE with the same payoffs that can be represented in MN(G) Preservation of locality! Direct link between strategic and probabilistic reasoning in CE Computation: In trees (e.g.), can compute a CE efficiently –Parsimonious LP formulation

Graphical Economics and Social Network Theory

A Network Model of Exchange Economies Begin with the classical framework: –k goods or commodities –n consumers, each with their own endowments and utility functions But now assume an undirected network dictating exchange –each vertex is a consumer –edge between i and j means they are free to engage in trade –no edge between i and j: direct exchange is forbidden Note: can “encode” network in goods and utilities –for each raw good g and consumer i, introduce virtual good (g,i) –think of (g,i) as “good g when sold by consumer i” –consumer j will have zero utility for (g,i) if no edge between i and j j’s original utility for g if there is an edge between i and j

Graphical Economics: Basic Theory Network equilibria always exist (under AD condition analogues) –does not follow from AD due to zero endowments of “foreign” goods –appeal to Debreu’s quasi-rationality: zero wealth may ignore zero prices –wealth propagation lemma: spread of capital on connected graph ADProp algorithm: –computes controlled approximation to graphical equilibrium –message-passing on conditional prices and inbound/outbound demands –efficient for tree topologies and smooth utilities Remarks: –Scarf’s algorithm –complete network, linear utility case: poly-time algorithm [DPSV 02] –complete network, smooth utility case: poly(n), exp(k) [KKO 04]

A Sample Network and Equilibrium Network is union of all edges shown 2 goods (cash and wheat) Symmetry in endowments and utility functions –buyers have $1 –sellers have 1 unit wheat Solid edges: –exchange at equilibrium Dashed edges: –competitive but unused Dotted edges: –non-competitive prices Note price variation –0.33 to 2.00 Degree does not determine price! –e.g. S5 vs. S14

Economics Properties of Social Networks How does network structure influence: –price variation in large economies? –wealth distribution (Pareto’s power law)? –computation of market equilibrium? Made possible by combination of: –social network models preferential attachment (“rich get richer”) Erdos-Renyi (random graphs) –math econ exchange models e.g. Fisher, Walras-Wald, Arrow-Debreu –advances in equilibrium computation algorithms for linear and general utilities

Sample Results Price (~wealth) distribution at equilibrium: –power law in networks generated by preferential attachment –sharply peaked in Erdos-Renyi Price variation (max/min) at equilibrium: –grows as a root of n in preferential attachment –none in Erdos-Renyi! –decays exponentially with increased connectivity –generally characterized by a (weak) expansion property of graph –connections to eigenvalues of adjacency matrix, random walks Local upper and lower bounds on equilibrium prices –subgraphs with buyer and seller frontiers –powerful algorithmic implications

Quality of Local Approximations Model: (  ) n = 50 to 250 (five plots) each plot averages 5 trials Very mild dependence on n k = 5 gives exact solution; k = 3 is 60% faster (n = 250)

An Amusing Experimental Illustration

U.N. Comtrade Data Network

USA: 4.42 Germany: 4.01 Italy: 3.67 France: 3.16 Japan: 2.27 top 3 partnerstop 10 partnersEU network sorted equilibrium prices vertex degree price EU: 7.18 USA: 4.50 Japan: 2.96

European Union Network

USA: 4.42 Germany: 4.01 Italy: 3.67 France: 3.16 Japan: 2.27 top 3 partnerstop 10 partnersEU network sorted equilibrium prices vertex degree price EU: 7.18 USA: 4.50 Japan: 2.96