Summer School on Graphs in Computer Graphics, Image and Signal Analysis Bornholm, Denmark, August 2011

Outline
Lecture 1: Monday (10:15 11:00) Introduction to the basic concepts of game theory Lecture 2: Tuesday (11:15 12:00) Evolutionary games and graph-based data clustering Exercises: Tuesday (13:00 14:30) Reading groups and mini-presentations Lecture 3: Thursday (09:15 10:00) Graph labeling problems and graph transduction

What is Game Theory?

The central problem of game theory was posed by von Neumann as early as 1926 in Gttingen. It is the following: If n players, P1,, Pn, play a given game , how must the ith player, Pi, play to achieve the most favorable result for himself? Harold W. Kuhn Lectures on the Theory of Games (1953) A few cornerstones in game theory
19211928: Emile Borel and John von Neumann give the rst modern formulation of a mixed strategy along with the idea of nding minimax solutions of normal-form games. 1944, 1947: John von Neumann and Oskar Morgenstern publish Theory of Games and Economic Behavior. 19501953: In four papers John Nash made seminal contributions to both non-cooperative game theory and to bargaining theory. 19721982: John Maynard Smith applies game theory to biological problems thereby founding evolutionary game theory. late 1990s : Development of algorithmic game theory

Normal-form Games
We shall focus on nite, non-cooperative, simultaneous-move games in normal form, which are characterized by: A set of players: I = {1, 2, , n} (n 2) A set of pure strategy proles: S = S1 S2 Sn where each Si = {1, 2, , mi} is the (nite) set of pure strategies (actions) available to the player i A payoff function: : S n, (s) = (1(s),,n(s)), where i(s) (i=1n) represents the payoff (or utility) that player i receives when strategy prole s is played

Each player is to choose one element from his strategy space in the absence of knowledge of the choices of the other players, and payments will be made to them according to the function i(s). Players goal is to maximize their own returns.

Two Players
In the case of two players, payoffs can be represented as two m1 x m2 matrices (say, A for player 1 and B for player 2):

A = (ahk )

ahk = 1 (h,k)

Special cases:

B = (bhk )

bhk = 2 (h,k)

Zero-sum games: A + B = 0 (ahk = bhk for all h and k) Symmetric games: BT = A Doubly-symmetric games: A = AT = BT

Example 1: Prisoners Dilemma

Prisoner 2

Confess (defect) Confess (defect) Prisoner 1 Deny (cooperate) -25 , -1 -10 , -10

Deny (cooperate) -1 , -25

-3 , -3

How to Solve the Game?

Do min

ate

ds

trat egy

Prisoner 2

Confess (defect)

Deny (cooperate) -1 , -25

Do m in

ate

Prisoner 1

str ate

Confess (defect) Deny (cooperate)

-10 , -10

gy

-25 , -1

-3 , -3

Example 2: Battle of the Sexes

Wife

Soccer

Ballet

Soccer Husband Ballet

2,1

0,0

0,0

1,2

Example 3: Rock-Scissors-Paper

You

Rock

Scissors

Paper

Rock

0,0

1 , -1

-1 , 1

Me

Scissors

-1 , 1

0,0

1 , -1

Paper

1 , -1

-1 , 1

0,0

Mixed Strategies
A mixed strategy for player i is a probability distribution over his set Si of pure strategies, which is a point in the (mi-1)-dimensional standard simplex:
mi i = x i R m i : h = 1mi : x ih 0, and x ih = 1 h=1

The set of pure strategies that is assigned positive probability by mixed

strategy xii is called the support of xi:

(x i ) = {h Si : x ih > 0}
A mixed strategy prole is a vector x = (x1,,xn) where each component xii is a mixed strategy for player iI. The mixed strategy space is the multi-simplex = 1 2 n

Standard Simplices

mi = 2

mi = 3

Note: Corners of standard simplex correspond to pure strategies.

Mixed-Strategy Payoff Functions

In the standard approach, all players randomizations are assumed to be independent. Hence, the probability that a pure strategy prole s = (s1,,sn) will be used when a mixed-strategy prole x is played is:
n

x(s) = x isi
i=1

and the expected value of the payoff to player i is:

ui (x) = x i (s) i (s)

sS

In the special case of two-players games, one gets:

T u1 (x) = x1h ahk x 2k = x1 Ax 2 h=1 k=1

m1 m 2

m1 m 2 T u2 (x) = x1h bhk x 2k = x1 Bx 2 h=1 k=1

where A and B are the payoff matrices of players 1 and 2, respectively.

Best Replies
Notational shortcut. Here, and in the sequel, if z and xii, the notation (xi,zi) stands for the strategy prole in which player iI plays strategy xi, while all other players play according to z. Player is best reply to the strategy prole xi is a mixed strategy xi*i such that ui(xi*,xi) ui(xi,xi) for all strategies xi i. The best reply is not necessarily unique. Indeed, except in the extreme case in which there is a unique best reply that is a pure strategy, the number of best replies is always innite. Indeed: When the support of a best reply includes two or more pure strategies, any mixture of these strategies must also be a best reply Similarly, if there are two pure strategies that are individually best replies, any mixture of the two is necessarily also a best reply

Nash Equilibria
The Nash equilibrium concept is motivated by the idea that a theory of rational decision-making should not be a self-destroying prophecy that creates an incentive to deviate for those who believe it. A strategy prole x is a Nash equilibrium if it is a best reply to itself, namely, if: ui(xi,xi) ui(zi,xi) for all i = 1n and all strategies zi i. If strict inequalities hold for all zi xi then x is said to be a strict Nash equilibrium. Theorem. A strategy prole x is a Nash equilibrium if and only if for every player iI, every pure strategy in the support of xi is a best reply to xi. It follows that every pure strategy in the support of any players equilibrium mixed strategy yields that player the same payoff.

Finding Pure-strategy Nash Equilibria

Player 2

Left

Middle

Right
Na sh e 10 , 2 quilib rium

Top

3,1

2,3

High Player 1 Low

4,5

3,0

6,4

2,2

5,4

12 , 3

Bottom

5,6

4,5

9,7

Multiple Equilibria in Pure Strategies

Na

! ium ilibr equ sh

Soccer

Wife
Na Ballet sh e

qui libr

ium

Soccer Husband Ballet

2,1

0,0

0,0

1,2

Nash equilibrium!

You

No

eq ash

rium uilib

! Rock

Scissors

Paper

Rock

0,0

1 , -1

-1 , 1

Me

Scissors

-1 , 1

0,0

1 , -1

Paper

1 , -1

-1 , 1

0,0

Theorem (Nash, 1951). Every nite normal-form game admits a mixedstrategy Nash equilibrium. Idea of proof. 1. 2. Dene a continuous map T on such that the xed points of T are in one-to-one correspondence with Nash equilibria. Use Brouwers theorem to prove existence of a xed point.

Together with factoring, the complexity of nding a Nash equilibrium is in my opinion the most important concrete open question on the boundary of P today. Christos Papadimitriou Algorithms, games, and the internet (2001)

Evolution and the Theory of Games

We repeat most emphatically that our theory is thoroughly static. A dynamic theory would unquestionably be more complete and therefore preferable. But there is ample evidence from other branches of science that it is futile to try to build one as long as the static side is not thoroughly understood. John von Neumann and Oskar Morgenstern Theory of Games and Economic Behavior (1944)

Paradoxically, it has turned out that game theory is more readily applied to biology than to the eld of economic behaviour for which it was originally designed. John Maynard Smith Evolution and the Theory of Games (1982)

10

Evolutionary Games
Introduced by John Maynard Smith (1973, 1974, 1982) to model the evolution of behavior in animal conicts. Assumptions: A large population of individuals belonging to the same species which compete for a particular limited resource This kind of conict is modeled as a symmetric two-player game, the players being pairs of randomly selected population members Players do not behave rationally but act according to a pre-programmed behavioral pattern Reproduction is assumed to be asexual Utility is measured in terms of Darwinian tness, or reproductive success

Interpreting Mixed Strategies

There are two ways to interpret the notion of a mixed strategy into the evolutionary framework: 1. Each individual is hard-wired to play a pure strategy, but some portion of the population plays one strategy while the rest of the population plays another. 2. Each individual is hard-wired to play a particular mixed strategy that is, they are genetically congured to choose randomly from among certain options with certain probabilities. It turns out that the two interpretations are mathematically equivalent. In dening the static notions of evolutionary game theory, it is customary to focus on the second idea; the dynamical aspects are instead more conveniently dealt with using the rst.

11

Evolutionary Stability
A strategy is evolutionary stable if it is resistant to invasion by new strategies. Formally, assume: A small group of invaders appears in a large populations of individuals, all of whom are pre-programmed to play strategy x Let y be the strategy played by the invaders Let be the share of invaders in the (post-entry) population (0 < < 1) The payoff in a match in this bimorphic population is the same as in a match with an individual playing mixed strategy: w = y + (1 )x hence, the (post-entry) payoffs got by the incumbent and the mutant strategies are u(x,w) and u(y,w), respectively.

Evolutionary Stable Strategies

Denition. A strategy x is said to be an evolutionary stable strategy (ESS) if for all y{x} there exists (0,1), such that for all (0, ) we have: u[x, y + (1 )x] > u[y, y + (1 )x]
incumbent mutant

Theorem. A strategy x is an ESS if and only if it meets the following rst- and second-order best-reply conditions: 1. u(y,x) u(x,x) for all y for all y{x}

2. u(y,x) = u(x,x) u(y,y) < u(x,y)

Note. From the conditions above, we have: ESS NE If x is a strict Nash equilibrium, then x is an ESS

12

Existence of ESSs
Unlike Nash equilibria existence of ESSs is not guaranteed. Unique Nash equilibrium is x=(1/3,1/3,1/3)T Hence, all y are best replies to x Let the mutant be y=(1,0,0)T But u(y,y) = u(x,y), hence ESS = Rock You Scissors Paper

Rock

0,0

1 , -1

-1 , 1

Me

Scissors

-1 , 1

0,0

1 , -1

Paper

1 , -1

-1 , 1

0,0

Complexity Issues
Two questions of computational complexity naturally present themselves: What is the complexity of determining whether a given game has an ESS (and of nding one)? What is the complexity of recognizing whether a given x is an ESS for a given game?

Theorem (Etessami and Lochbihler, 2004). Determining whether a given two-player symmetric game has an ESS is both NP-hard and coNP-hard.

Theorem (Nisan, 2006). Determining whether a (mixed) strategy x is an ESS of a given two-player symmetric game is coNP-hard.

13

Replicator Dynamics
Let xi(t) the population share playing pure strategy i at time t. The state of the population at time t is: x(t)= (x1(t),,xn(t)). Replicator dynamics (Taylor and Jonker, 1978) are motivated by Darwins principle of natural selection:

xi payoff of pure strategy i average population payoff xi

which yields:

x i = x i [ u(e i , x) u(x, x)] = x i [(Ax) i x T Ax ]

Notes. Invariant under positive afne transformations of payoffs (i.e., u u+, with>0) Standard simplex is invariant under replicator dynamics, namely, x(0) x(t), for all t > 0 (so is its interior and boundary)

Replicator Dynamics and ESSs

Theorem (Nachbar, 1990; Taylor and Jonker, 1978). A point x is a Nash equilibrium if and only if x is the limit point of a replicator dynamics trajectory starting from the interior of . Furthermore, if x is an ESS, then it is an asymptotically stable equilibrium point for the replicator dynamics. The opposite need not be true.
0 6 4 A = 3 0 5 1 3 0

The point m=(1/3,1/3,1/3)T is asymptotically stable (its eigenvalues have negative parts). But e1=(1,0,0)T is an ESS. Hence m cannot be an ESS (being in the interior, it would have to be the unique ESS).

14

Doubly Symmetric Games

In a doubly symmetric (or partnership) game, the payoff matrix A is symmetric (A = AT). Fundamental Theorem of Natural Selection (Losert and Akin, 1983). For any doubly symmetric game, the average population payoff (x) = xTAx is strictly increasing along any non-constant trajectory of replicator dynamics, namely, d/dt(x(t)) 0 for all t 0, with equality if and only if x(t) is a stationary point.

Characterization of ESSs (Hofbauer and Sigmund, 1988) For any doubly simmetric game with payoff matrix A, the following statements are equivalent: a) x ESS b) x is a strict local maximizer of (x) = xTAx over the standard simplex c) x is asymptotically stable in the replicator dynamics

Discrete-time Replicator Dynamics

A well-known discretization of replicator dynamics, which assumes nonoverlapping generations, is the following (assuming a non-negative A):

x i (t + 1) = x i (t)

A( x(t)) i x(t)T Ax(t)

which inherits most of the dynamical properties of its continuous-time counterpart (e.g., the fundamental theorem of natural selection).

MATLAB implementation

15

References
Texts on (classical) game theory J. von Neumann and O. Morgerstern. Theory of Games and Economic Behavior. Princeton University Press (1944, 1953). D. Fudenberg and J. Tirole. Game Theory. MIT Press (1991). M. J. Osborne and A. Rubinstein. A Course in Game Theory. MIT Press (1994). Texts on evolutionary game theory J. Weibull. Evolutionary Game Theory. MIT Press (1995). J. Hofbauer and K. Sigmund. Evolutionary Games and Population Dynamics. Cambridge University Press (1998). Computationally-oriented texts N. Nisan, T. Roughgarden, E. Tardos, and V. Vazirani (Eds.) Algorithmic Game Theory. Cambridge University Press (2007). Y. Shoham and K. Leyton-Brown. Multiagent Systems: Algorithmic, Game-Theoretic, and Logical Foundations. Cambridge University Press (2009). On-line resources http://gambit.sourceforge.net/ http://gamut.stanford.edu/ a library of game-theoretic algorithms a suite of game generators for testing game algorithms

Reading Groups on Tuesday

Matching M. Pelillo, K. Siddiqi, and S. W. Zucker. Matching hierarchical structures using association graphs. ECCV 1998 (longer version in PAMI 1999). M. Pelillo. Replicator equations, maximal cliques, and graph isomorphism. NIPS 1998 (longer version in Neural Computation 1999). M. Pelillo. Matching free trees with replicator equations. NIPS 2001 (longer version in PAMI 2002). A. Albarelli, A. Torsello, S. Rota Bul, and M. Pelillo. Matching as a non-cooperative game. ICCV 2009. Grouping M. Pavan and M. Pelillo. Dominant sets and hierarchical clustering. ICCV 2003. M. Pavan and M. Pelillo. Efcient out-of-sample extension of dominant-set clusters. NIPS 2004. A. Torsello, S. Rota Bul, and M. Pelillo. Beyond partitions: Allowing overlapping groups in pairwise clustering. ICPR 2008. S. Rota Bul and M. Pelillo. A game-theoretic approach to hypergraph clustering. NIPS 2009. Applications R. Hamid, A. Johnson, S. Batta, A. Bobick, C. Isbell, G. Coleman. Detection and explanation of anomalous activities: Representing activities as bags of event n-grams. CVPR 2005. L. Li, X. Zhang, W. Hu, W. Li, and P. Zhu Soccer video shot classicaton based on color characterization using dominant sets clustering. PCM 2009. A. Albarelli, E. Rodol, A. Cavallarin, and A. Torsello. Robust gure extraction on textured background: A game-theoretic approach. ICPR 2010. X. Yang, H. Liu, and L. J. Latecki. Contour-based object detection as dominant set computation. ACCV 2010.

16