Basic Values in Europe Shalom Schwartz The Hebrew University of Jerusalem
[email protected] 05-11-2006 Center for the Study of Rationality Hebrew University of Jerusalem...
-
Upload
elinor-moody -
Category
Documents
-
view
213 -
download
0
Transcript of [email protected] 05-11-2006 Center for the Study of Rationality Hebrew University of Jerusalem...
Center for the Study of RationalityHebrew University of Jerusalem
1/39
n-Player Stochastic Gameswith Additive Transitions
Frank Thuijsman
János Flesch & Koos Vrieze
Maastricht University
European Journal of Operational Research 179 (2007) 483–497
Center for the Study of RationalityHebrew University of Jerusalem
2/39
Outline
• Model
• Brief History of Stochastic Games
• Additive Transitions
• Examples
Center for the Study of RationalityHebrew University of Jerusalem
3/39
as = (a1s , a2
s , , ans) joint action
rs(as) = (r1s(as), r2
s(as), , rns(as)) rewards
ps(as) = (ps(1|as), ps(2|as), , ps(z|as)) transitions
• Infinite horizon• Complete Information• Perfect Recall• Independent and Simultaneous Choices
Finite Stochastic Game
1 s z
rs(as)
ps(as)
Center for the Study of RationalityHebrew University of Jerusalem
4/39
3-Player Stochastic Game
0, 0, 0
1, 3, 0 0, 3, 1 3, 0, 1
1
2 3 4
(1, 0, 0, 0)
(2/3, 1/3, 0, 0)
(2/3, 0, 1/3, 0)
(2/3, 0, 0, 1/3)
(1/3, 1/3, 0, 1/3)
(1/3, 0, 1/3, 1/3)
(1/3, 1/3, 1/3, 0)
(0, 1/3, 1/3, 1/3)
(0, 1, 0, 0) (0, 0, 1, 0) (0, 0, 0, 1)
0, 0, 0
0, 0, 0
0, 0, 0
0, 0, 0 0, 0, 0
0, 0, 00, 0, 0
T
L
F
R
B
N
1
2
3
Center for the Study of RationalityHebrew University of Jerusalem
5/39
general strategy i : N × S × H → X
i
(k, s, h) → X is
Markov strategy f i : N × S → X
i
(k, s) → X is
stationary strategy x i : S → X
i
(s) → X is
opponents’ strategy i , f
i and x i
Strategies
mixed actions
Center for the Study of RationalityHebrew University of Jerusalem
6/39
-Discounted rewards (with 0 < < 1)
is() = Es((1) k k1 Ri
k)
Limiting average rewards
is() = Es(limK → K1 K
k=1 Rik)
Rewards
Center for the Study of RationalityHebrew University of Jerusalem
7/39
MinMax Values
-Discounted minmax
v isinfi sup i i
s()
Limiting average minmax
v isinfi sup i i
s()
Highest rewards player i can defend
Center for the Study of RationalityHebrew University of Jerusalem
8/39
Equilibrium
= ( i)i N is an -equilibrium if
is(
i, i) ≤ i
s() +
for all i, for all i and for all s.
Center for the Study of RationalityHebrew University of Jerusalem
9/39
Question
(0, 0, 1, 0)(0, 1, 0, 0)
(0, 0, 0, 1)
(1, 0, 0, 0)(0, 0.5, 0.5, 0)
321
0, 0
0, 0
0, 0 3, 10, 0
Any -equilibrium?
(0, 0, 0, 1)
4
2,1
Center for the Study of RationalityHebrew University of Jerusalem
10/39
Highlights from (Finite)Stochastic Games History
Shapley, 1953
0-sum, “discounted”
Everett, 1957
0-sum, recursive, undiscounted
Gillette, 1957
0-sum, big match problem
Center for the Study of RationalityHebrew University of Jerusalem
11/39
Highlights from (Finite)Stochastic Games History
Fink, 1964 & Takahashi, 1964
n-player, discounted
Blackwell & Ferguson, 1968
0-sum, big match solution
Liggett & Lippmann, 1969
0-sum, perfect inf., undiscounted
Center for the Study of RationalityHebrew University of Jerusalem
12/39
Highlights from (Finite)Stochastic Games History
Kohlberg, 1974
0-sum, absorbing, undiscounted
Mertens & Neyman, 1981
0-sum, undiscounted
Sorin, 1986
2-player, Paris Match, undiscounted
Center for the Study of RationalityHebrew University of Jerusalem
13/39
Highlights from (Finite)Stochastic Games History
Vrieze & Thuijsman, 1989
2-player, absorbing, undiscounted
Thuijsman & Raghavan, 1997
n-player, perfect inf., undiscounted
Flesch, Thuijsman, Vrieze, 1997
3-player, absorbing example, undiscounted
Center for the Study of RationalityHebrew University of Jerusalem
14/39
Highlights from (Finite)Stochastic Games History
Solan, 1999
3-player, absorbing, undiscounted
Vieille, 2000
2-player, undiscounted
Solan & Vieille, 2001
n-player, quitting, undiscounted
Center for the Study of RationalityHebrew University of Jerusalem
15/39
Additive Transitions
ps(as) = ni=1 i
s pis(ai
s)
pis(ai
s) transition probabilities controlled by player i in state s
is transition power of player i in state s
0 ≤ is ≤ 1 and
i i
s = 1 for each s
Center for the Study of RationalityHebrew University of Jerusalem
16/39
Example for 2-PlayerAdditive Transitions
ps(as) = ni=1 i
s pis(ai
s)
(1, 0, 0)
(0.7, 0, 0.3) (0, 0.7, 0.3)
(0.3, 0.7, 0)
p11(2) = (0, 0, 1)
p11(1) = (1, 0, 0)
p21(1)=(1, 0, 0) p2
1(2)=(0, 1, 0)1
1 = 0.3
21 = 0.7
1
Center for the Study of RationalityHebrew University of Jerusalem
17/39
Results
1. 0-equilibria for n-player AT games (threats!)
2. 0-opt. stationary strat. for 0-sum AT games
3. Stat. -equilibria for 2-player abs. AT games
4. Result 3 can not be strengthened,
neither to 3-player abs. AT games,
nor to 2-player non-abs. AT games,
nor to give stat. 0-equilibria
Center for the Study of RationalityHebrew University of Jerusalem
18/39
The Essential Observation
Additive Transitions
induce
a Complete Ordering of the Actions
If ais is “better” than bi
s against some strategy,
Then ais is “better” than bi
s against any strategy.
Center for the Study of RationalityHebrew University of Jerusalem
19/39
“Better”
Consider strategies ais and bi
s for player i
Iffor some strategyais we have
t S ps(t | ais , ai
s ) vit ≥ t S ps(t | bi
s , ais ) vi
t
Then for all strategies bis we have
t S ps(t | ais , bi
s ) vit ≥ t S ps(t | bi
s , bis ) vi
t
Center for the Study of RationalityHebrew University of Jerusalem
20/39
Because ….
If t S ps(t | ais , ai
s)vit ≥ t S ps(t | bi
s , ais)vi
t
Then
is t S ps(t | ai
s) vit + j i j
s t S ps(t | ajs) vi
t ≥
is t S ps(t | bi
s) vit + j i j
s t S ps(t | ajs) vi
t
Center for the Study of RationalityHebrew University of Jerusalem
21/39
which implies that ….
is t S ps(t | ai
s) vit + j i j
s t S ps(t | bjs) vi
t ≥
is t S ps(t | bi
s) vit + j i j
s t S ps(t | bjs) vi
t
And therefore
t S ps(t | ais , bi
s)vit ≥ t S ps(t | bi
s , bis)vi
t
Center for the Study of RationalityHebrew University of Jerusalem
22/39
“Best”
The “best” actions for player i in state s
are those that maximize the expression
t S ps(t | ais) vi
t
Center for the Study of RationalityHebrew University of Jerusalem
23/39
The Restricted Game
Let G be the original AT game and
let G* be the restricted AT game,
where each player is restricted
to his “best” actions.
Center for the Study of RationalityHebrew University of Jerusalem
24/39
The Restricted Game
Now v *i ≥ v i for each player i.
In G* : t S ps(t | a*s) v*it = v*i
s i, s, a*s
In G : t S ps(t | bis , a*i
s ) vit < vi
s i, s,
a*is , bi
s
Center for the Study of RationalityHebrew University of Jerusalem
25/39
The Restricted Game
If x*i yields at least v*i in G*,
then x*i yields at least v*i in G as well.
Center for the Study of RationalityHebrew University of Jerusalem
26/39
Ex. 1: 2-Player Absorbing AT Game
(0, 0, 1)(0, 1, 0)
(1, 0, 0) (0.5, 0.5, 0)
(0, 0.5, 0.5)(0.5, 0, 0.5)
321
0, 0
0, 0
0, 0
0, 0 1, 33, 1
(1, 0, 0)
(1, 0, 0)
(0, 1, 0)
(0, 0, 1)
0.5
0.5
Center for the Study of RationalityHebrew University of Jerusalem
27/39
(0, 0, 1)(0, 1, 0)
(1, 0, 0) (0.5, 0.5, 0)
(0, 0.5, 0.5)(0.5, 0, 0.5)
321
0, 0
0, 0
0, 0
0, 0 1, 33, 1
T, L B, LB, RT, R
0, 0 -1, 3-2, 2-3, 1
T, L
0, 0
NO stationary 0-equilibrium
Ex. 1: 2-Player Absorbing AT Game
Center for the Study of RationalityHebrew University of Jerusalem
28/39
(0, 0, 1)(0, 1, 0)
(1, 0, 0) (0.5, 0.5, 0)
(0, 0.5, 0.5)(0.5, 0, 0.5)
321
0, 0
0, 0
0, 0
0, 0
stationary -equilibrium
with >0
1
1
0
equilibrium rewards ≈ ((1
1, 33, 1
Ex. 1: 2-Player Absorbing AT Game
Center for the Study of RationalityHebrew University of Jerusalem
29/39
(0, 0, 1)(0, 1, 0)
(1, 0, 0) (0.5, 0.5, 0)
(0, 0.5, 0.5)(0.5, 0, 0.5)
321
0, 0
0, 0
0, 0
0, 0 1, 33, 1
non-stationary -equilibrium
Player 1: B, T, T, T, ….
Player 2: R, R, R, R, ….
equilibrium rewards ((223, 11, 3
Ex. 1: 2-Player Absorbing AT Game
Center for the Study of RationalityHebrew University of Jerusalem
30/39
Ex. 2: 2-Player Non-Absorbing AT Game
(0, 0, 1, 0)(0, 1, 0, 0)
(0, 0, 0, 1)
(1, 0, 0, 0)(0, 0.5, 0.5, 0)
321
0, 0
0, 0
0, 0 3, 10, 0
q > 0 p <
q 1 q
1p
p
q = 0 p > 1 q > 0
NO stationary -equilibrium
with >0
(0, 0, 0, 1)
4
2,1
Center for the Study of RationalityHebrew University of Jerusalem
31/39
non-stationary -equilibrium
Player 1: T, B, B, B, ….
Player 2: R, R, R, R, ….
(0, 0, 1, 0)(0, 1, 0, 0)
(0, 0, 0, 1)
(1, 0, 0, 0)(0, 0.5, 0.5, 0)
321
0, 0
0, 0
0, 0 3, 10, 0
(0, 0, 0, 1)
4
2,1
Ex. 2: 2-Player Non-Absorbing AT Game
equilibrium rewards ((2, 1), (
Center for the Study of RationalityHebrew University of Jerusalem
32/39
alternative -equilibrium
Player 1: T, T, T, T, ….
Player 2: L, R, R, R, ….
(0, 0, 1, 0)(0, 1, 0, 0)
(0, 0, 0, 1)
(1, 0, 0, 0)(0, 0.5, 0.5, 0)
321
0, 0
0, 0
0, 0 3, 10, 0
(0, 0, 0, 1)
4
2,1
Ex. 2: 2-Player Non-Absorbing AT Game
equilibrium rewards ((2, 1), (
Center for the Study of RationalityHebrew University of Jerusalem
33/39
Ex. 3: 3-Player Absorbing AT Game
0, 0, 0
1, 3, 0 0, 1, 3 3, 0, 1
1
2 3 4
(1, 0, 0, 0)
(2/3, 1/3, 0, 0)
(2/3, 0, 1/3, 0)
(2/3, 0, 0, 1/3)
(1/3, 1/3, 0, 1/3)
(1/3, 0, 1/3, 1/3)
(1/3, 1/3, 1/3, 0)
(0, 1/3, 1/3, 1/3)
(0, 1, 0, 0) (0, 0, 1, 0) (0, 0, 0, 1)
0, 0, 0
0, 0, 0
0, 0, 0
0, 0, 0 0, 0, 0
0, 0, 00, 0, 0
T
L
F
R
B
N
How to share 4 among three people
if only few solutions are allowed?
1
2
3
Center for the Study of RationalityHebrew University of Jerusalem
34/39
0, 0, 0
1, 3, 0
0, 1, 3
3, 0, 1
1, 3, 0 0, 1, 3 3, 0, 1
1
2 3 4
(1, 0, 0, 0)
(2/3, 1/3, 0, 0)
(2/3, 0, 1/3, 0)
(2/3, 0, 0, 1/3)
(1/3, 1/3, 0, 1/3)
(1/3, 0, 1/3, 1/3)
(1/3, 1/3, 1/3, 0)
(0, 1/3, 1/3, 1/3)
(0, 1, 0, 0) (0, 0, 1, 0) (0, 0, 0, 1)
1/2, 2, 3/2
2, 3/2, 1/2
3/2, 1/2, 2
4/3, 4/3, 4/3T
L
F
R
B
N
Ex. 3: 3-Player Absorbing AT Game
Center for the Study of RationalityHebrew University of Jerusalem
35/39
0, 0, 0
1, 3, 0
0, 1, 3
3, 0, 1
1/3 *
1/3 *
1/3 *
2/3 *
2/3 *
2/3 *
1 *1/2, 2, 3/2
2, 3/2, 1/2
3/2, 1/2, 2
4/3, 4/3, 4/3T
L
F
R
B
N
NO stationary -equilibrium
Ex. 3: 3-Player Absorbing AT Game
1, 3, 0
0, 1, 3
3, 0, 1
Center for the Study of RationalityHebrew University of Jerusalem
36/39
0, 0, 0
1, 3, 0
0, 1, 3
3, 0, 1
1/3 *
1/3 *
1/3 *
2/3 *
2/3 *
2/3 *
1 *1/2, 2, 3/2
2, 3/2, 1/2
3/2, 1/2, 2
4/3, 4/3, 4/3T
L
F
R
B
N
Player 1 on B: 1, ¾, 0, 0, 0, 0, 1, ¾, 0, 0, 0, 0, ….
Player 2 on R: 0, 0, 1, ¾, 0, 0, 0, 0, 1, ¾, 0, 0, ….
Player 3 on F: 0, 0, 0, 0, 1, ¾, 0, 0, 0, 0, 1, ¾, ….
non-stationary -equilibrium
Ex. 3: 3-Player Absorbing AT Game
Center for the Study of RationalityHebrew University of Jerusalem
37/39
0, 0, 0
1, 3, 0
0, 1, 3
3, 0, 1
1/3 *
1/3 *
1/3 *
2/3 *
2/3 *
2/3 *
1 *1/2, 2, 3/2
2, 3/2, 1/2
3/2, 1/2, 2
4/3, 4/3, 4/3T
L
F
R
B
N
equilibrium rewards (1, 2, 1)
Ex. 3: 3-Player Absorbing AT Game
Player 1 on B: 1, ¾, 0, 0, 0, 0, 1, ¾, 0, 0, 0, 0, ….
Player 2 on R: 0, 0, 1, ¾, 0, 0, 0, 0, 1, ¾, 0, 0, ….
Player 3 on F: 0, 0, 0, 0, 1, ¾, 0, 0, 0, 0, 1, ¾, ….
Center for the Study of RationalityHebrew University of Jerusalem
38/39
Results
1. 0-equilibria for n-player AT games (threats!)
2. 0-opt. stationary strat. for 0-sum AT games
3. Stat. -equilibria for 2-player abs. AT games
4. Result 3 can not be strengthened,
neither to 3-player abs. AT games,
nor to 2-player non-abs. AT games,
nor to give stat. 0-equilibria
Center for the Study of RationalityHebrew University of Jerusalem
39/39
?