CSCI 3210: Computational Game Theory Inefficiency of Equilibria - - PDF document

csci 3210 computational game theory inefficiency of
SMART_READER_LITE
LIVE PREVIEW

CSCI 3210: Computational Game Theory Inefficiency of Equilibria - - PDF document

12/5/19 CSCI 3210: Computational Game Theory Inefficiency of Equilibria & Routing Games Ref: Ch 17, 18 [AGT] Mohammad T . Irfan Email: mirfan@bowdoin.edu Web: www.bowdoin.edu/~mirfan 1 Split or steal game u NE outcome vs. socially


slide-1
SLIDE 1

12/5/19 1

CSCI 3210: Computational Game Theory

Mohammad T . Irfan Email: mirfan@bowdoin.edu Web: www.bowdoin.edu/~mirfan

Inefficiency of Equilibria & Routing Games

Ref: Ch 17, 18 [AGT]

1

Split or steal game

u NE outcome vs. socially best/optimal outcome Payoff matrix

Split Steal Split

$33K, $33K $0+fr., $66K

Steal

$66K, $0+fr. $0, $0

Lucy Tony

2

slide-2
SLIDE 2

12/5/19 2

Prisoner's "dilemma" game

u Again: NE outcome vs socially optimal outcome Payoff matrix

Not Confess Confess Not Confess

1, 1 10, 0

Confess

0, 10 5, 5

Suspect 1 Suspect 2 Costs (negative of payoffs)

3

Measuring the inefficiency of NE

u What is the objective function to compare

different outcomes?

u Utilitarian u Egalitarian

u How to deal with multiplicity of NE?

u Inefficiency of which NE? u Price of anarchy vs. price of stability

4

slide-3
SLIDE 3

12/5/19 3

Price of Anarchy (PoA)

u PoA =

Worst objective function value among all NE Objective function value of optimal outcome

5

Price of Stability (PoS)

u PoS =

Best objective function value among all NE Objective function value of optimal outcome

6

slide-4
SLIDE 4

12/5/19 4

Example

u Calculate PoA and PoS Payoff matrix

Not Confess Confess Not Confess

1, 1 10, 0

Confess

0, 10 5, 5

Suspect 1 Suspect 2 Costs (negative of payoffs)

7

Example

u Calculate PoA and PoS Payoff matrix

L R U

21, -1 10, 0

D

100, 10 7, 8

Row player Column player Costs (negative of payoffs)

8

slide-5
SLIDE 5

12/5/19 5

PoA vs. PoS

u Consider costs u PoA and PoS will be >= 1 u PoA = PoS when all NE have the same cost

(e.g., unique NE)

u In general, PoA >= PoS

9

PoA vs. PoS

u PoA: worst case guarantee in a system of

independent agents

u PoS: measures benefit of a protocol or

proposed outcome

10

slide-6
SLIDE 6

12/5/19 6

Pigou’s Example

PoA and PoS

11

Routing Games

13

slide-7
SLIDE 7

12/5/19 7

Model: nonatomic selfish routing

u Multicommodity flow network

u Directed network with multiple (source, sink) pairs u Each (source, sink) pair is called a commodity u ri amount of traffic for each commodity i

u Each edge e has a delay or cost function ce

u Every car going through an edge gets same delay

u Cost of a path = sum of edge costs

u Note: cost doesn't depend on identity of players u Congestion games

14

Equilibrium flow

u Let f be a feasible flow (combining all

commodities)

u f is equilibrium flow if

u All detours have higher (or equal) delay

equilibrium flow

15

slide-8
SLIDE 8

12/5/19 8

More complex graphs

u Cost of the whole flow (all red and green)

u Total delay on each edge

s1 s2 t1 t2

16

Surprise!

u Price of anarchy for any nonatomic routing

game with linear costs <= 4/3

19

slide-9
SLIDE 9

12/5/19 9

Example: nonlinear Pigou

u Consider large number p; 1 unit of traffic u Equilibrium cost = ? Optimal cost = ? PoA = ?

equilibrium flow PoA à +∞

20

Example: Braess' paradox

u 1 unit of traffic

equilibrium flow and cost? PoA = ?

21

slide-10
SLIDE 10

12/5/19 10

Braess' paradox

u New super highway between v and w

PoA = 4/3

22