Download Slajd 1 - VideoLectures.NET

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Minimax wikipedia , lookup

Artificial intelligence in video games wikipedia , lookup

Nash equilibrium wikipedia , lookup

Prisoner's dilemma wikipedia , lookup

The Evolution of Cooperation wikipedia , lookup

Evolutionary game theory wikipedia , lookup

Chicken (game) wikipedia , lookup

Transcript
Dresden, ECCS’07
Equilibrium transitions
in stochastic evolutionary games
Jacek Miękisz
Institute of Applied Mathematics
University of Warsaw
Population dynamics
time
A and B are two possible behaviors,
fenotypes or strategies of each individual
Matching of individuals
everybody interacts with everybody
random pairing of individuals
space –structured populations
Simple model of evolution
Selection
individuals interact in pairs – play games
receive payoffs = # of offspring
Fenotypes are inherited
Offspring may mutate
Main Goals
Equilibrium selection in case of multiple Nash equilibria
Dependence of the long-run behavior of population on
--- its size
--- mutation level
Stochastic dynamics of finite unstructured populations
n - # of individuals
zt - # of individuals playing A at time t
Ω = {0,…,n} - state space
selection
zt+1 > zt
if „average payoff” of A > „average payoff” of B
mutation
each individual may mutate and switch to the other strategy
with a probability ε
Markov chain with n+1 states
and a unique stationary state μεn
Previous results
Playing against the field, Kandori-Mailath-Rob 1993
A
B
A
a
c
B
b
d
a>c, d>b, a>d,
a+b<c+d
(A,A) and (B,B) are Nash equilibria
A is an efficient strategy
B is a risk-dominant strategy
Random matching of players, Robson - Vega Redondo, 1996
pt
# of crosspairings
Our results, JM J. Theor. Biol, 2005
Theorem
(random matching model)
Spatial games with local interactions
deterministic dynamics of the best-response rule
i
Stochastic dynamics
a) perturbed best response
with the probability 1-ε, a player chooses the best response
with the probability ε a player makes a mistake
b) log-linear rule or Boltzmann updating
Example
A is a dominated strategy
without A
B is stochastically stable
with A
C is ensemble stable at intermediate noise levels
in log-linear dynamics
A
B
C
A
0
0.1
1
B
0.1
2+α
1.1
C
1.1
1.1
2
where α > 0
ε
C
B
α
Open Problem
Construct a spatial game
with a unique stationary state μεΛ
which has the following property
Real Open Problem
Construct a one-dimensional cellular automaton
model with a unique stationary state μεΛ
such that when you take the infinite lattice limit
you get two measures.