Download Markov Chains - Networking Research Lab @ SYSU

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

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

Document related concepts
no text concepts found
Transcript
Markov Chains
Jian He
1
Outline
• Markov Decision
Probability Review
Process
Stochastic Processes
• Hidden Markov
Definition
Chains
Classification of States • Continuous-time
Limiting Probabilities
Markov chain
Time-reversible Markov
Chain
• Branching Processes
• Random Walk
•
•
•
•
•
•
2
Probability Review
• Conditional probability • Expectation of r.v.
P[ A | B ] 
P[ AB ]
P[ B ]
E[ X ] 
• Bayes' Theorem
P[ Ai | B] 

x : p ( x ) 0
• Moment Generating Function
P[ B | Ai ]P[ Ai ]
n
i 1
 xp( x )
 ( t )  E [e ]
P[ B | Ai ]P[ Ai ]
tX
• CDF vs PDF
F (a )  P{ X  (, a]}  
a

 n ( 0)  E[ X n ]
n1
f ( x )dx
3
Stochastic Processes
• Collection of random variables
{ X ( t ), t  T }
• t is often interpreted as time,while X(t) as the
state of the process
• T is called the index set of the process
FX(t)(x)  P{X(t)  x}
• Example:
job waiting in the queue as a function of time
4
Definition
• Stochastic process {X n ,n  0 ,1,2 ,...}
• Finite or countable number of possible values
• Markov property
P{ X n1  j | X n  i , X n1  in1 ,..., X 1  i1 , X 0  i0 }
 P{ X n1  j | X n  i }  Pij
5
One-step Transition Probability
• Homogeneous Markov chain
P{X n1  j | X n  i}  P{X n  j | X n1  i}
n
• One-step transition probability
P00
P10
P 
Pi 0
P01
P11

Pi 1


P02 
P12 

Pi 2 

6
Chapman-Kolmogorov Equations
(n )
ij
• n-step transition probabilities P
( n)
ij
P
 P{X nk  j | X k  i}
nm
ij
P
P
( n)

 P P
k 0
P
n m
ik kj
n 1
n  0, i, j  0
for all n, m  0, all i, j
P  P
n
7
Classification of States
• Accessibility n  0 P  0
i j
• Communicate if i  j and j  i , then i  j
• Properties of communication
if i  j , then j  i
i i
n
ij
if i  j and j  k , then i  k
• class
states communicate with each other
• irreducible only one class
8
Classification of States(cont.)

• recurrent state
• transient state
P
n 1

n
ii

n
P
 ii  
n 1
• state i is recurrent if and only if, starting in
state i, the expected number of time periods
that the process is in state i is infinite
9
Recurrent vs Transient
• If state i is recurrent, and state i
communicates with state j, then
state j is recurrent
• At least one of the states must be
recurrent in finite-state Markov
chains
• All states of a finite irreducible
Markov chain are recurrent
10
Random Walk
• Markov chain
• State space i  0,1,2,...
• Transition probability
Pi ,i 1  p  1  Pi ,i 1
i  0,1,2,...
• irreducible
• all transient or all recurrent ?
2n
00
P
(4 p(1  p)) n
~
n
1
n1 P   if and only if p  2

n
00
11
Random Walk(cont.)
• Symmetric random walk equal probability
• Two-dimensional symmetric random walk
P( i , j ),( i 1, j )  P( i , j ),( i 1, j )  P( i , j ),( i , j 1)  P( i , j ),( i , j 1)
2n
00
P
1
~
n


2n
00
P
n 1
1

4

12
Limiting Probabilities
•
•
•
•
•
n
d

gcd{
n
|
P
period
ii  0}
aperiodic d=1
if state i has period d, and states i and j
communicate, then state j also has period d.
positive recurrent
n
nPii  
expected return time is finite
ergodic positive recurrent && aperiodic

13
Limiting Probabilities(cont.)
• Irreducible ergodic Markov chain
exists, and is independent of i
n
• Definition  j  lim Pij
n 
• Property

 j    i Pij
i 0
lim n Pijn

j0

j 0
j
1
•  j equals the long-run proportion of time that
the process will be in state j
14
Limiting Probabilities(Example)
• Transition probability matrix
0 .5 0 .4 0 .1
P  0 .3 0 .4 0 .3
0 .2 0 .3 0 .5
 0  0.5 0  0.3 1  0.2 2
 1  0.4 0  0.4 1  0.3 2
 2  0.1 0  0.3 1  0.5 2
21
23
18
 0  , 1 
, 2 
62
62
62
15
Limiting Probability(Property)
• Let { X n , n  1} be an irreducible Markov chain
with stationary probabilities  j , j  0 ,and let r be a
bounded function on the state space. Then, with
probability 1,

lim
N 
N
n 1
r( X n )
N
  j  0 r ( j ) j

• If we suppose that we earn a reward r(j)
whenever the chain is in state j, then our
average reward per unit time is  j r ( j ) j
16
Advanced Random Walk
• Markov chain with states 0,1,...,n having
P0,1  1, Pi ,i 1  p, Pi ,i 1  q  1  p,1  i  n
•
N i the number of additional transitions that it
takes the chain when it first enters state i until it
enters state i+1
• i  E[ N i ]
• Goals the number of trasitions that it takes
the chain to go from state 0 to state n
n 1
N 0,n   N i
i 0
17
Mean Number of Transitions

 1  E[# additional trasitions to reach i  1 |
i
•
chain to i  1]q
•  i  1  E[ N i*1  N i* ]q
 1  q( i 1  i )
•
  q/ p

1 n 1 i 1 j n 1 i

i 1
1
  

 i    j   i   E[ N 0 , n ]  1  p 
i 1 j  0
i 1

p j 0
18
Mean Number of Transitions(cont.)
1
• when p  , E[ N 0,n ]  n2
2
n1
n1
1
2


(
n

1
)

 n1
• when p  , E[ N 0,n ]  1 
2
(1   )2
1
• p  exponentially increasing function
2
1
• p  2 for large n, essentially linear in n
19
Branching Processes
• A population consisting of individuals able to
produce offspring of the same kind
• Each individual produces j new offsring with
probability Pj , j  0 independently,by the end of
lifetime.
• X n the size of the nth generation
• { X n , n  0,1,2,...} is a Markov chain
20
Extinction
• State 0 is a recurrent state P00  1
• All other states are transient, if P0  0

•    jP j   E[ X ]  E[ X ]   n

n
n 1
j 0
X 0  1 


•
 0  lim P{ X n  0 | X 0  1}   0    Pj
n 
j 0
j
0
21
Age-dependent Branching Processes
• Attach another random variable 'age'
 t
• CDF fT ( t )  e , t  0
22
Mean Number of Individuals
• Theorem
m ( t )  E[ Z ( t )]
There exist   0 and   0 such that
t
m( t ) ~ e
whenever m( t )  1
23
Branching Processes(Example)
• Service Capacity of
P2P system in
trasient regime
• N d (t ) the number of
peers available to
serve document d at
time t
24
Service Capacity
• Basic Braching Processes Model
25
Time Reversible Markov Chains
• Stationary ergodic Markov chain
Pij stationary probabilities  j
• { X n , X n1 , X n 2 ,..., X 0 } a Markov Chain ?
transition probabilities
• Transition probabilities
 j Pji
Qij  P{ X m  j | X m 1  i } 
i
• Theorem
Time reversible, iff Qij  Pij   i Pij   j Pji
26
Reversibility(Example)
• Arbitrary connected graph
 ij
Pij 
1
  ij
2
3
3
1
j
 ij
i
j
 j  ij

i
i
 ij
i  ji
6
2
5
4
4
1
27
Reversibility(Property)
• An ergodic Markov chain for which Pij  0
whenever Pji  0 is time reversible if and only if
starting in state i, any path back to i has the
same probability as the reversed path. That is, if
Pi ,i1 Pi1 ,i2 ... Pik ,i  Pi ,ik Pik ,ik 1 ... Pi1 ,i
for all states i , i
1
,..., ik
28
Reversibility(Extenstion)
• Irreducible Markov chain Pij
• Theorem
i  0 
  Qij : tran prob of reversed chain
i  i  1    : stationary probabilit ies (both)
i


 i Pij   j Q ji 
29
Markov Decision Processes
• After observing the state of the process, an action
must be chosen
P{ X n1  j | X 0 , a0 , X 1 , a1 ,..., X n  i , an  a }  Pij (a )
• Limiting Probabilities
 ia  0 for all i , a
   1
    
i
a
a
ja
ia
i
a
ia
Pij (a ) for all j
30
Hidden Markov Chains
• finite set of signals 
p( s | j )  1
• emitting signals
s
• observing the sequence of signals

P{ Sn  s | X 1 , S1 ,..., X n1 , Sn1 , X n  j }  p( s | j )
• conditional probability of state
n
P
{
S
 sn , X n  j }
n
P { X n  j | S  sn } 
P{ S n  sn }
P { S n  sn , X n  j }

n
P
{
S
 n, X n  i }
i
31
Continuous-Time Markov Chains
• Definition
P{ X ( t  s )  j | X ( s )  i , X ( u)  x( u),0  u  s}
 P{ X ( t  s )  j | X ( s )  i }
•
•
Ti the amount of time stay in state i
P{Ti  s  t | Ti  s}  P{Ti  t }
• memoryless  exponentially distributed
Continunous-Time Markov Chain(property)
• The amount of time the process spends in a
state before making a transition into another
state is exponentially distributed
• Transition probability
Pii  0, all i
P
ij
j
 1, all i
Limiting Probabilities
(Continuous-Time version)
• The mean time stays in one state 1 / v i
• qij  v i Pij
• Properties of Limiting Probabilities
v j Pj   qkj Pk
k j
P
j
1
j
• v j Pj rate at which the process leaves state j
•  qkj Pk rate at which the process enters j
k j
Birth and Death Process
• state the number of people
• whenever there are n persons
• successive arrival time exponentially
distributed with mean 1 /  n
• successvie departure time exponentially
distributed with mean 1 / n
• Continuous-time Markov chain
Birth and Death Process(cont.)
• Transition probabilities
i
P01  1 Pi ,i 1    
i
i
Pi ,i 1 
i
i   i
• Limiting Probabilities
0 P0  1 P1
(n  n ) Pn  n1 Pn1  n1 Pn1
0 1 ...n1
Pn 

01 ...n1
1  2 ... n (1  
)
n 1 1  2 ... n
P0 
1
0 1 ...n1
1 
n 1 1  2 ... n

M/M/1 System
• Poisson arrival(or the interarrival time is
exponential)and service time is exponentially
 t
 t
distributed. Arrival is e and service is  e
M/M/1 System(cont.)
• Limiting Probabilities

P0  1 

  n
Pn  (1  )( )
 
• E[number of customers in the system]

  iPi 
i 0

1 

   system utilization

References
• Sheldon M. Ross, Introduction to Probability
Models, 9th Edition. ISBN 978-7-115-160232/O1. 2007
• G.R. Grimmett and D. R. Stirzaker, Probability
and Random Processes,2nd Edition, Clarendon
Press, Oxford,1992.
• X. Yang and G. de Veciana. Service Capacity
of Peer to Peer Networks. INFOCOM,2004
39
40
Related documents