a (brief) introduction to game theoryjcohen/documents/enseignement/gametheory.pdf · the game of...
TRANSCRIPT
![Page 1: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/1.jpg)
A (Brief) Introduction to Game Theory
Johanne Cohen
PRiSM/CNRS, Versailles, France.
![Page 2: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/2.jpg)
Goal
is a Nash equilibrium.
![Page 3: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/3.jpg)
Goal
is a Nash equilibrium.
![Page 4: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/4.jpg)
Today
The game of ChickenDefinitionsNash Equilibrium
Rock-paper-scissors GameMixed strategyMixed Nash Equilibrium
Prisoner’s dilemma
Bonus : Toward learning equilibria
![Page 5: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/5.jpg)
Outline
The game of ChickenDefinitionsNash Equilibrium
Rock-paper-scissors GameMixed strategyMixed Nash Equilibrium
Prisoner’s dilemma
Bonus : Toward learning equilibria
![Page 6: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/6.jpg)
The game of Chicken (Hawk-dove game)
I A single lane bridge
I Two drivers Bob and Alice want to go crossit from opposite directions.
Each driver can Cross or Stop
I Both drivers want to minimize the time spentto reach other side.
I if both attempt to cross, the result is a fataltraffic accident.
There are 4 outcomes depending on thechoices made by each of the 2 drivers
![Page 7: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/7.jpg)
The game of Chicken (Hawk-dove game)
I A single lane bridge
I Two drivers Bob and Alice want to go crossit from opposite directions.
Each driver can Cross or Stop
I Both drivers want to minimize the time spentto reach other side.
I if both attempt to cross, the result is a fataltraffic accident.
There are 4 outcomes depending on thechoices made by each of the 2 drivers
![Page 8: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/8.jpg)
The game of Chicken (Hawk-dove game)
I A single lane bridge
I Two drivers Bob and Alice want to go crossit from opposite directions.
Each driver can Cross or Stop
I Both drivers want to minimize the time spentto reach other side.
I if both attempt to cross, the result is a fataltraffic accident.
There are 4 outcomes depending on thechoices made by each of the 2 drivers
![Page 9: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/9.jpg)
Model.
1. Who play ?
Alice and Bob
2. Which actions/strategies ?
Cross or Stop
3. Which payoff according to strategy profile ?
cost = transport time
BobCross Stop
Cross
(60, 60)
(1, 2)
Alic
e
Stop (2, 1)
(5, 5)
![Page 10: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/10.jpg)
Model.
1. Who play ? Alice and Bob
2. Which actions/strategies ?Cross or Stop
3. Which payoff according to strategy profile ?cost = transport time
BobCross Stop
Cross
(60, 60)
(1, 2)
Alic
e
Stop (2, 1)
(5, 5)
![Page 11: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/11.jpg)
Model.
1. Who play ? Alice and Bob
2. Which actions/strategies ?Cross or Stop
3. Which payoff according to strategy profile ?cost = transport time
BobCross Stop
Cross
(60, 60)
(1, 2)
Alic
e
Stop (2, 1)
(5, 5)
![Page 12: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/12.jpg)
Model.
1. Who play ? Alice and Bob
2. Which actions/strategies ?Cross or Stop
3. Which payoff according to strategy profile ?cost = transport time
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1)
(5, 5)
![Page 13: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/13.jpg)
Model.
1. Who play ? Alice and Bob
2. Which actions/strategies ?Cross or Stop
3. Which payoff according to strategy profile ?cost = transport time
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
![Page 14: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/14.jpg)
Games in standard form
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
I The two strategies of Bob correspond to the two columns.
I The entries of the matrix are the outcomes incurred by theplayers in each situation.
![Page 15: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/15.jpg)
Rational behavior
rational behavior of player : select strategy which minimizes itscost.
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
For example :
1. If Bob selects to Cross, then Alice would select to Stop.2. If Bob selects to Stop, then Alice would select to Cross.
Alice has a rational behavior.
![Page 16: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/16.jpg)
Rational behavior
rational behavior of player : select strategy which minimizes itscost.
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
For example :
1. If Bob selects to Cross, then Alice would select to Stop.2. If Bob selects to Stop, then Alice would select to Cross.
Alice has a rational behavior.
![Page 17: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/17.jpg)
Best response
Best responses of player i :
= the strategies which produce the most favorableoutcome for a player
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
Equilibrium = mutual best responses
![Page 18: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/18.jpg)
Best response
Best responses of player i :
= the strategies which produce the most favorableoutcome for a player
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
Equilibrium = mutual best responses
![Page 19: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/19.jpg)
Nash Equilibrium
Consider a game with a set of n players {1, . . . , n}I Each player has a set of possible strategies Si
I s = (s1, · · · , sn) is a vector of strategies selected by the players
A pure strategy Nash Equilibrium (NE) is a vector of strategiess = (s1, · · · , sn) such that
∀i , ∀s ′i , we have ci (s1, · · · , si , · · · , sn) ≤ ci (s1, · · · , s ′i , · · · , sn).
![Page 20: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/20.jpg)
Nash Equilibrium
Consider a game with a set of n players {1, . . . , n}I Each player has a set of possible strategies Si
I s = (s1, · · · , sn) is a vector of strategies selected by the players
A pure strategy Nash Equilibrium (NE) is a vector of strategiess = (s1, · · · , sn) such that
∀i , ∀s ′i we have ci (si , s−i ) ≤ ci (s′i , s−i ).
![Page 21: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/21.jpg)
Back to example
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
I 2 pure strategy Nash Equilibria :
(Cross, Stop ) and (Stop, Cross )
I But which equilibrium should be selected ? Which one will beselected by the system if its converges ?
One solution
![Page 22: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/22.jpg)
Back to example
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
I 2 pure strategy Nash Equilibria :
(Cross, Stop ) and (Stop, Cross )
I But which equilibrium should be selected ? Which one will beselected by the system if its converges ?
One solution
![Page 23: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/23.jpg)
Back to example
BobCross Stop
Cross (60, 60) (1, 2)
Alic
e
Stop (2, 1) (5, 5)
I 2 pure strategy Nash Equilibria :
(Cross, Stop ) and (Stop, Cross )
I But which equilibrium should be selected ? Which one will beselected by the system if its converges ?
One solution
![Page 24: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/24.jpg)
Outline
The game of ChickenDefinitionsNash Equilibrium
Rock-paper-scissors GameMixed strategyMixed Nash Equilibrium
Prisoner’s dilemma
Bonus : Toward learning equilibria
![Page 25: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/25.jpg)
Rock-paper-scissors Game
Rules : 2 players select one strategy fromRock/Paper/scissors.
Standard form :
@wikipedia
Scissors Paper Rock
Scissors (0, 0) (−1, 1) (1,−1)
Paper (1,−1) (0, 0) (−1, 1)
Rock (−1, 1) (1,−1) (0, 0)
I No pure strategy Nash equilibrium.
I But if players select strategies at random,pi (Rock) + pi (Paper) + pi (Scissors) = 1
Nash equilibrium of mixed strategies.
![Page 26: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/26.jpg)
Rock-paper-scissors Game
Rules : 2 players select one strategy fromRock/Paper/scissors.
Standard form :
@wikipedia
Scissors Paper Rock
Scissors (0, 0) (−1, 1) (1,−1)
Paper (1,−1) (0, 0) (−1, 1)
Rock (−1, 1) (1,−1) (0, 0)
I No pure strategy Nash equilibrium.
I But if players select strategies at random,pi (Rock) + pi (Paper) + pi (Scissors) = 1
Nash equilibrium of mixed strategies.
![Page 27: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/27.jpg)
Rock-paper-scissors GameRules : 2 players select one strategy from
Rock/Paper/scissors.
Standard form :
@wikipedia
Scissors Paper Rock
Scissors (0, 0) (−1, 1) (1,−1)
Paper (1,−1) (0, 0) (−1, 1)
Rock (−1, 1) (1,−1) (0, 0)
I No pure strategy Nash equilibrium.I But if players select strategies at random,
pi (Rock) + pi (Paper) + pi (Scissors) = 1
And if one player prefers one strategy (Paper) to the othersthen his opponent prefers the corresponding winning strategy
(Scissors) :
Nash equilibrium of mixed strategies.
![Page 28: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/28.jpg)
Rock-paper-scissors GameRules : 2 players select one strategy from
Rock/Paper/scissors.
Standard form :
@wikipedia
Scissors Paper Rock
Scissors (0, 0) (−1, 1) (1,−1)
Paper (1,−1) (0, 0) (−1, 1)
Rock (−1, 1) (1,−1) (0, 0)
I No pure strategy Nash equilibrium.I But if players select strategies at random,
pi (Rock) + pi (Paper) + pi (Scissors) = 1I If each player picks each of his 3 strategies with probability
1/3,then nobody can improve its payoff.
Nash equilibrium of mixed strategies.
![Page 29: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/29.jpg)
Using the random selection method.
Consider a game with a set of n players {1, . . . , n}I Each player has a set of possible pure strategies Si
I a cost function ci : S1 × · · · × Sn → N
A mixed strategy is a probability distribution pi
over his set of possible pure strategies (actions).
∀i ,∑
s∈Sipi (s) = 1
A mixed profile p is a vector of n elements (p1, . . . , pn)such that player i selects actions using probability pi .
![Page 30: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/30.jpg)
Expected cost
The expected cost Ci of player i with the game profile p isCi (p) = E [ci (p)]
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
Assume
that player Bob decides to pick Cross with probability 1/3 ,
that player Alice decides to pick Cross with probability 1/2
CBob(pBob, pAlice) =
![Page 31: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/31.jpg)
Expected cost
The expected cost Ci of player i with the game profile p isCi (p) = E [ci (p)]
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
Assume
that player Bob decides to pick Cross with probability 1/3 ,
that player Alice decides to pick Cross with probability 1/2
CBob(pBob, pAlice) =13 ×
12 × 60+
![Page 32: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/32.jpg)
Expected cost
The expected cost Ci of player i with the game profile p isCi (p) = E [ci (p)]
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
Assume
that player Bob decides to pick Cross with probability 1/3 ,
that player Alice decides to pick Cross with probability 1/2
CBob(pBob, pAlice) =13 ×
12 × 60+ 1
6 × 1 + 13 × 2 + 1
3 × 5
![Page 33: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/33.jpg)
Expected cost
The expected cost Ci of player i with the game profile p isCi (p) = E [ci (p)]
BobCross Stop
Cross (60, 60) (1, 2)A
lice
Stop (2, 1) (5, 5)
Assume
that player Bob decides to pick Cross with probability 1/3 ,
that player Alice decides to pick Cross with probability 1/2
CBob(pBob, pAlice) = 756
![Page 34: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/34.jpg)
Best response of a mixed strategy
BobCross Stop
Cross (60, 60) (1, 2)
Alic
eStop (2, 1) (5, 5)
Bob picks
{Cross with probability qStop with probability 1− q
What happens for Alice ?
if Alice selects the action Cross, thenexpected cost = 60q + 1(1− q)
if Alice selects the action Stop, thenexpected cost = 2q + 5(1− q)
When does Alice select the action Cross ?
if 60q + 1(1− q) < 2q + 5(1− q), in others words if q < 2/31
![Page 35: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/35.jpg)
Best response of a mixed strategy
BobCross Stop
Cross (60, 60) (1, 2)
Alic
eStop (2, 1) (5, 5)
Bob picks
{Cross with probability qStop with probability 1− q
What happens for Alice ?
if Alice selects the action Cross, thenexpected cost = 60q + 1(1− q)
if Alice selects the action Stop, thenexpected cost = 2q + 5(1− q)
When does Alice select the action Cross ?
if 60q + 1(1− q) < 2q + 5(1− q), in others words if q < 2/31
![Page 36: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/36.jpg)
Best response of a mixed strategy
BobCross Stop
Cross (60, 60) (1, 2)
Alic
eStop (2, 1) (5, 5)
Bob picks
{Cross with probability qStop with probability 1− q
What happens for Alice ?
if Alice selects the action Cross, thenexpected cost = 60q + 1(1− q)
if Alice selects the action Stop, thenexpected cost = 2q + 5(1− q)
When does Alice select the action Cross ?
if 60q + 1(1− q) < 2q + 5(1− q), in others words
if q < 2/31
![Page 37: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/37.jpg)
Best response of a mixed strategy
BobCross Stop
Cross (60, 60) (1, 2)
Alic
eStop (2, 1) (5, 5)
Bob picks
{Cross with probability qStop with probability 1− q
What happens for Alice ?
if Alice selects the action Cross, thenexpected cost = 60q + 1(1− q)
if Alice selects the action Stop, thenexpected cost = 2q + 5(1− q)
When does Alice select the action Cross ?
if 60q + 1(1− q) < 2q + 5(1− q), in others words
if q < 2/31
![Page 38: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/38.jpg)
Using the random selection method...
6
-Stop
Cross
0 1q
2/31
2/31
p
stop
Cross
d
dd
1. Alice selects Cross if q < 2/31
2. Using the same argument as previously :Assume that Alice selects{
Cross with probability pStop with probability 1− p
,
Bob selects Cross if p < 2/31
3. Nash Equilibrium = intersection of the bothlines.
Two Nash Equilibria of pure strategies
One Nash Equilibrium of mixed strategies
![Page 39: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/39.jpg)
Using the random selection method...
6
-Stop
Cross
0 1q
2/31
2/31
p
stop
Cross
d
dd
1. Alice selects Cross if q < 2/31
2. Using the same argument as previously :Assume that Alice selects{
Cross with probability pStop with probability 1− p
,
Bob selects Cross if p < 2/31
3. Nash Equilibrium = intersection of the bothlines.
Two Nash Equilibria of pure strategies
One Nash Equilibrium of mixed strategies
![Page 40: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/40.jpg)
Using the random selection method...
6
-Stop
Cross
0 1q
2/31
2/31
p
stop
Cross
d
dd
1. Alice selects Cross if q < 2/31
2. Using the same argument as previously :Assume that Alice selects{
Cross with probability pStop with probability 1− p
,
Bob selects Cross if p < 2/31
3. Nash Equilibrium = intersection of the bothlines.
Two Nash Equilibria of pure strategies
One Nash Equilibrium of mixed strategies
![Page 41: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/41.jpg)
Using the random selection method...
6
-Stop
Cross
0 1q
2/31
2/31
p
stop
Cross
d
d
d
1. Alice selects Cross if q < 2/31
2. Using the same argument as previously :Assume that Alice selects{
Cross with probability pStop with probability 1− p
,
Bob selects Cross if p < 2/31
3. Nash Equilibrium = intersection of the bothlines.
Two Nash Equilibria of pure strategies
One Nash Equilibrium of mixed strategies
![Page 42: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/42.jpg)
Using the random selection method...
6
-Stop
Cross
0 1q
2/31
2/31
p
stop
Cross
d
dd
1. Alice selects Cross if q < 2/31
2. Using the same argument as previously :Assume that Alice selects{
Cross with probability pStop with probability 1− p
,
Bob selects Cross if p < 2/31
3. Nash Equilibrium = intersection of the bothlines.
Two Nash Equilibria of pure strategies
One Nash Equilibrium of mixed strategies
![Page 43: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/43.jpg)
Mixed Nash equilibrium
Consider a game with a set of n players {1, . . . , n}I Each player has a set of possible pure strategies Si
I a cost function ci : S1 × · · · × Sn → N
A mixed Nash equilibrium is a profile p∗ = (p∗1 , · · · , p∗n) such that
∀i , ∀p′i ∈ Pi we have Ci (p∗i , p
∗−i ) ≤ Ci (p
′i , p
∗−i ).
Pi the set of mixed strategies of i .
back to example
![Page 44: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/44.jpg)
Nash’s Theorem
Theoreme [Nash51]
Every finite game (with a finite set of players andfinite set of strategies) has a mixed strategy Nashequilibrium
Recall : there is a game withoutpure strategy Nash equilibrium.
![Page 45: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/45.jpg)
Outline
The game of ChickenDefinitionsNash Equilibrium
Rock-paper-scissors GameMixed strategyMixed Nash Equilibrium
Prisoner’s dilemma
Bonus : Toward learning equilibria
![Page 46: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/46.jpg)
Prisoner’s dilemma : statement
Bob and Alice that committed a crimeare interviewed separately by the police.
The offer of the police is the following :
1. If only one of them confess, then he/she will be relaxedand the other will get a sentense of 10 years.
2. If they both remain silent, then they both will have to serveprison sentences of 1 year.
3. if they both confess then they both will get a sentense of 8years.
They have two strategies : Confess or Silent.
![Page 47: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/47.jpg)
Standard form
Two strategies : Confess or Silent.
BobConfess Silent
Confess (8, 8) (0, 10)A
lice
Silent (10, 0) (1, 1)
I The strategy Confess dominates strategy Silent.
∀s ∈ Si ci (s, s−i ) ≥ ci (Confess, s−i )
I (Confess, Confess) is a Nash equilibrium
I (Silent, Silent) is more favorable than (Confess,Confess)
![Page 48: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/48.jpg)
Standard form
Two strategies : Confess or Silent.
BobConfess Silent
Confess (8, 8) (0, 10)A
lice
Silent (10, 0) (1, 1)
I The strategy Confess dominates strategy Silent.
∀s ∈ Si ci (s, s−i ) ≥ ci (Confess, s−i )
I (Confess, Confess) is a Nash equilibrium
I (Silent, Silent) is more favorable than (Confess,Confess)
![Page 49: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/49.jpg)
Domination in the sense of Pareto
BobConfess Silent
Confess (8, 8) (0, 10)
Alic
eSilent (10, 0) (1, 1)
Definition : Profile s Pareto-dominates profile s if
1. ∀i , ci(s) ≤ ci(s),
2. ∃j , cj(s) < cj(s),
Remark : (Silent, Silent) Pareto-dominates(Confess, Confess).
![Page 50: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/50.jpg)
Domination in the sense of Pareto
Notion of cooperation
Definition : Profile s Pareto-dominates profile s if
1. ∀i , ci(s) ≤ ci(s),
2. ∃j , cj(s) < cj(s),
Remark : (Silent, Silent) Pareto-dominates(Confess, Confess).
![Page 51: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/51.jpg)
And if games are repeated ?
I a fixed number k of times,I Confess dominates Silent at step k of the repeated game ;
the two players hence play Confess.I same reasoning for the last but one step.I players play Confess at time k, k − 1, · · · , 1
Introduction of a probability δthat the game continues for one more step
I in a infinite number of steps,I Strategies = mixed strategies in the static game.
I Construction of a strategy of behaviors that correspondto a simulation of mixed strategy S
and if a player i deviates, then it is punished
![Page 52: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/52.jpg)
And if games are repeated ?
I a fixed number k of times,I Confess dominates Silent at step k of the repeated game ;
the two players hence play Confess.I same reasoning for the last but one step.I players play Confess at time k, k − 1, · · · , 1
Introduction of a probability δthat the game continues for one more step
I in a infinite number of steps,I Strategies = mixed strategies in the static game.
I Construction of a strategy of behaviors that correspondto a simulation of mixed strategy S
and if a player i deviates, then it is punished
![Page 53: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/53.jpg)
And if games are repeated ?
I a fixed number k of times,I Confess dominates Silent at step k of the repeated game ;
the two players hence play Confess.I same reasoning for the last but one step.I players play Confess at time k, k − 1, · · · , 1
Introduction of a probability δthat the game continues for one more step
I in a infinite number of steps,I Strategies = mixed strategies in the static game.
I Construction of a strategy of behaviors that correspondto a simulation of mixed strategy S
and if a player i deviates, then it is punished
![Page 54: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/54.jpg)
Outline
The game of ChickenDefinitionsNash Equilibrium
Rock-paper-scissors GameMixed strategyMixed Nash Equilibrium
Prisoner’s dilemma
Bonus : Toward learning equilibria
![Page 55: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/55.jpg)
Toward learning equilibria
I The same game is repeated at each step.
I At every step t, player i has to solve the following problem :
Which action to play at time t, given the past history of thegame ?
that is to say
for all players i, vi (t) = f (Q),
where fi is a function that gives the behavior of i in functionof history Q.
![Page 56: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/56.jpg)
A dynamic : fictitious player
A player is a fictitious player if the player has the followingbehavior :
the player will play a best responsein function of the past statistic ofof strategies of his/her adversary,
That is to say
If player 2 used nj times the action j between step 1 andt − 1, then player 1 will estimate
that player 2 will play the action iwith probability q2,j(t) =
nj
t−1 at time t.
![Page 57: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/57.jpg)
A dynamic : fictitious player
player 21 2
player 11 (3,1) (0,3)2 (1,2) (2,0)
I Going from a discrete time to a continuous time
I The system = the couple (q1,1, q2,1)with qi ,1 = probability that player i plays strategy 1.
![Page 58: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/58.jpg)
A dynamic : fictitious player
A B
CD
q1,1
q2,1
(1, 0)(1, 1)
(1, 0)
(0, 0)
A B
CD
q1,1
q2,1
(1, 0)(1, 1)
(1, 0)
(0, 0)
direction of the dynamic example of behaviorin zone A of the dynamic
For zone A :I player 1 will be willing to use pure strategy 2,
and player 2 pure strategy 1.I the dynamic (q1,1, q2,1) will stay in A up to time t + τ
for small τ > 0.So q2,1(t + τ) =
tq2,1(t)t+τ . By making converging τ → 0, we
obtain
q′2,1(t) =q2,1(t)
t.
![Page 59: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/59.jpg)
Questions ?
![Page 60: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/60.jpg)
Prisoner’s dilemma (other interpretation)
J 2transmit transmit
J1 transmit (1-c , 1-c) (-c , 1)transmit (1 , -c) (0 , 0)
I c > 0 is the cost of traffic,
I “1” represents the fact that packets reach the destination.
![Page 61: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/61.jpg)
Domination in the sense of Pareto
J 2transmit transmit
J1 transmit (1-c , 1-c) (-c , 1)
transmit (1 , -c) (0 , 0)
Remark :
(transmit,transmit) is more favorable than(transmit,transmit).
Definition : The profile s Pareto-domine the profil ssi
1. ∀i , ui(s) ≥ ui(s),
2. ∃j , uj(s) > uj(s),
![Page 62: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/62.jpg)
Domination in the sense of Pareto
Notion of cooperation
Remark :
(transmit,transmit) is more favorable than(transmit,transmit).
Definition : The profile s Pareto-domine the profil ssi
1. ∀i , ui(s) ≥ ui(s),
2. ∃j , uj(s) > uj(s),
![Page 63: A (Brief) Introduction to Game Theoryjcohen/documents/enseignement/GameTheory.pdf · The game of Chicken (Hawk-dove game) I A single lane bridge I Two drivers Bob and Alice want to](https://reader034.vdocument.in/reader034/viewer/2022042211/5eb0ecbc00b1622d1e67002e/html5/thumbnails/63.jpg)
Questions ?