eager markov chains parosh aziz abdulla noomene ben henda richard mayr sven sandberg texpoint fonts...

48
Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg

Upload: shawna-hallum

Post on 01-Apr-2015

219 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Eager Markov Chains

Parosh Aziz Abdulla

Noomene Ben Henda

Richard Mayr

Sven Sandberg

Page 2: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Outline

Introduction Expectation Problem Algorithm Scheme Termination Conditions Subclasses of Markov Chains

Examples Conclusion

Page 3: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Introduction

Model: Infinite-state Markov chains Used to model programs with unreliable

channels, randomized algorithms…

Interest: Conditional expectations Expected execution time of a program Expected resource usage of a program

Page 4: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Introduction Infinite-state Markov

chain

Infinite set of states Target set Probability

distributions

s0

s1 s2

s3

s4

s5

Example

0.3

0.20.5

1

0.5

0.5

1

0.1

0.9

0.7 0.3

Page 5: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Introduction

Reward function

Defined over paths reaching the target set

s0

s1 s2

s3

s4

s5

0.3

0.20.5

1

0.5

0.5

1

0.1

0.9

0.7 0.3

Example

2

22

0

-3

-1

Page 6: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Expectation Problem

Instance A Markov chain A reward function

Task Compute/approximate the conditional

expectation of the reward function

Page 7: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Expectation Problem Example:

The weighted sum

The reachability probability

The conditional expectation

10.8

0.1

0.1

1

1

1

s0

2 2

0

-3

-5

0.8*4+0.1*(-5)=2.7

0.8+0.1=0.9

2.7/0.9=3

s1 s2

s3

s4

Page 8: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Expectation Problem

Remark Problem in general studied for finite-state

Markov chains

Contribution Algorithm scheme to compute it for infinite-

state Markov chains Sufficient conditions for termination

Page 9: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Algorithm Scheme At each iteration n

Compute paths up to depth n

Consider only those ending in the target set

Update the expectation accordingly

Path Exploration

Page 10: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Algorithm Scheme

Correctness The algorithm computes/approximates the

correct value

Termination Not guaranteed: lower-bounds but no upper-

bounds

Page 11: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Termination Conditions

Exponentially bounded reward function

The intuition: limit on the growth of the reward functions

Remark: The limit is reasonable: for example polynomial functions are exponentially bounded

Page 12: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Termination Conditions

n0

· k®nThe abs of the reward

k

Bound on the reward

Page 13: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Termination Conditions

Eager Markov chain

The intuition: Long paths contribute less in the expectation value

Remark: Reasonable: for example PLCS, PVASS, NTM induce all eager Markov chains

Page 14: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Termination Conditions

n0

1

k· k®n

Prob. of reaching the target in more

than n steps

Bound on the probability

Page 15: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Termination Conditions

Pf

Ws

Ce

E f

P

Wf

n0 nc

"(nc)

Page 16: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Subclasses of Markov Chains Eager Markov chains

Markov chains with finite eager attractor

Markov chains with the bounded coarseness property

NTM

PVASS

PLCS

Page 17: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager Attractor Attractor:

Almost surely reached from every state

Finite eager attractor: Almost surely reached Unlikely to stay ”too

long” outside of it

A EA

Page 18: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager Attractor

EA

0

1

n

b

Prob. to return in More than n steps

· b̄ n

Page 19: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager Attractor

Finite eager attractor implies eager Markov chain??

Reminder: Eager Markov chain:

· k®nProb. of reaching the target in more

than n steps

Page 20: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager Attractor

FEA

Paths of length n that visit the attractor t times

Page 21: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager Attractor

Proof idea: identify 2 sets of paths Paths that visit the attractor often without

going to the target set:

Paths that visit the attractor rarely without going the target set:

t · n=c

t > n=c

Page 22: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager AttractorPaths visiting the attractor rarely: t less than n/c

FEA

·P n=c

t=1 C t¡ 1n¡ 1b

t¯n¡ tPr_n

· (( cc¡ 1)(2c)t=c(1

c + b¯ )1=c¯)n

Page 23: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Finite Eager AttractorPaths visiting the attractor often: t greater than n/c

FEA

Pt ̧ ¹Pl · (1¡ ¹ )

Po_n · !(1¡ ¹ )(1¡ (1¡ ¹ )1=! ) ((1¡ ¹ )

1c! )n

Page 24: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Probabilistic Lossy Channel Systems (PLCS)

Motivation:

Finite-state processes communicating through unbounded and unreliable channels

Widely used to model systems with unreliable channels (link protocol)

Page 25: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS

ab

b

Send c!a

ab

Receive c?b

a

c?b

q0

q3 q2

q1

nop

c!a

c!b

aba

Channel c

nop1 21

1

1

Page 26: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS

c?b

q0

q3 q2

q1

nop

c!a

c!b

aba

Channel c

nop1 21

1

1

ab

b

Loss

b ba a

Page 27: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS Configuration

Control location Content of the

channel

Example [q3,”aba”]

c?b

q0

q3 q2

q1

nop

c!a

c!b

aba

Channel c

nop1 21

1

1

Page 28: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS

A PLCS induces a Markov chain:

States: Configurations

Transitions: Loss steps combined with discrete steps

Page 29: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS Example:

[q1,”abb”] [q2,”a”] By losing one of the

messages ”b” and firing the marked step.

Probability: P=Ploss*2/3

c?b

q0

q3 q2

q1

nop

c!a

c!b

aba

Channel c

nop1 21

1

1

Page 30: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PLCS

Result: Each PLCS induces a Markov chain with finite eager attractor.

Proof hint: When the size of the channels is big enough, it is more likely (with a probability greater than ½) to lose a message.

Page 31: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Bounded Coarseness

The probability of reaching the target within K steps is bounded from below by a constant b.

Page 32: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Bounded Coarseness

Boundedly coarse Markov chain implies eager Markov chain??

Reminder: Eager Markov chain:

· k®nProb. of reaching the target in more

than n steps

Page 33: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Bounded CoarsenessProb. Reach. ¸ b

Wit

hin

K s

teps

K nK steps2K

Pn · (1¡ b)nP2 · (1¡ b)2

Pn:Prob. of avoidingthe target in nK steps

P1 · (1¡ b)

Page 34: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Probabilistic Vector Addition Systems with states (PVASS)

Motivation:

PVASS are generalizations of Petri-nets.

Widely used to model parallel processes, mutual exclusion program…

Page 35: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS Configuration

Control location Values of the

variables x and y

Example:

[q1,x=2,y=0]

q0

q3 q2

q1

nop

--x --y

++x

++y1 2

++x

1

4

1

1

Page 36: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS

A PVASS induces a Markov chain:

States: Configurations

Transitions: discrete steps

Page 37: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS Example:

[q1,1,1] [q2,1,0] By taking the marked

step.

Probability: P=2/3

q0

q3 q2

q1

nop

--x --y

++x

++y1 2

++x

1

4

1

1

Page 38: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS

Result: Each PVASS induces a Markov chain which has the bounded coarseness property.

Page 39: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Noisy Turing Machines (NTM)

Motivation:

They are Turing Machines augmented with a noise parameter.

Used to model systems operating in ”hostile” environment

Page 40: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

NTM

Fully described by a Turing Machine and a noise parameter.

q1

q3q2

q4

a/b b

a/b

b #a/b

#

RRR

RR

S

S

ab# b #aa b

Page 41: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

NTMq1

q3q2

q4

a/b b

a/b

b #a/b

#

RRR

RR

S

S

ab# b #aa b

Discret Step

ab# b #aa b

bb# b #aa b

Page 42: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

NTMq1

q3q2

q4

a/b b

a/b

b #a/b

#

RRR

RR

S

S

ab# b #aa b

Noise Step

ab# b #aa b

#b# b #aa b

Page 43: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

NTM

Result: Each NTM induces a Markov chain which has the bounded coarseness property.

Page 44: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Conclusion

Summary: Algorithm scheme for approximating

expectations of reward functions

Sufficient conditions to guarantee termination: Exponentially bounded reward function Eager Markov chains

Page 45: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

Conclusion

Direction for future work

Extending the result to Markov decision processes and stochastic games

Find more concrete applications

Page 46: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Thank you

Page 47: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS Order on

configurations: <= Same control

locations Ordered values of the

variables

Example: [q0,3,4] <= [q0,3,5]

q0

q3 q2

q1

nop

--x --y

++x

++y1 2

++x

1

4

1

1

Page 48: Eager Markov Chains Parosh Aziz Abdulla Noomene Ben Henda Richard Mayr Sven Sandberg TexPoint fonts used in EMF. Read the TexPoint manual before you delete

Info

rmat

ions

tekn

olog

i

Institutionen för informationsteknologi | www.it.uu.se

PVASS

Probability of each step > 1/10

Boundedly coarse: parameters K and 1/10^K

q0

q3 q2

q1

nop

--x --y

++x

++y1 2

++x

1

4

1

1

Targetset

K iterations