+ All Categories
Home > Documents > Algorithmic Game Theory and Internet Computing

Algorithmic Game Theory and Internet Computing

Date post: 11-Jan-2016
Category:
Upload: diata
View: 36 times
Download: 0 times
Share this document with a friend
Description:
Combinatorial Approximation Algorithms for Convex Programs?!. Algorithmic Game Theory and Internet Computing. Vijay V. Vazirani Georgia Tech. Rational convex program. Always has a rational solution, using polynomially many bits, if all parameters are rational. - PowerPoint PPT Presentation
Popular Tags:
94
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Georgia Tech Combinatorial Approximation Algorithms for Convex Programs?!
Transcript
Page 1: Algorithmic Game Theory and Internet Computing

Algorithmic Game Theoryand Internet Computing

Vijay V. Vazirani

Georgia Tech

Combinatorial Approximation

Algorithms for Convex Programs?!

Page 2: Algorithmic Game Theory and Internet Computing

Rational convex program

Always has a rational solution,

using polynomially many bits,

if all parameters are rational.

Some important problems in mathematical

economics and game theory are captured by

rational (nonlinear) convex programs.

Page 3: Algorithmic Game Theory and Internet Computing

A recent development

Combinatorial exact algorithms for

these problems and hence for optimally

solving their convex programs.

Page 4: Algorithmic Game Theory and Internet Computing

General equilibrium theory

Page 5: Algorithmic Game Theory and Internet Computing

A central tenet

Prices are such that demand equals supply, i.e.,

equilibrium prices.

Easy if only one good

Page 6: Algorithmic Game Theory and Internet Computing

Supply-demand curves

Page 7: Algorithmic Game Theory and Internet Computing

Irving Fisher, 1891

Defined a fundamental

market model

Page 8: Algorithmic Game Theory and Internet Computing
Page 9: Algorithmic Game Theory and Internet Computing
Page 10: Algorithmic Game Theory and Internet Computing

utility

Concave utility function

(for good j)

amount of j

Page 11: Algorithmic Game Theory and Internet Computing

( )i ij ijj G

u f x

total utility

Page 12: Algorithmic Game Theory and Internet Computing

For given prices,find optimal bundle of goods

1p 2p3p

Page 13: Algorithmic Game Theory and Internet Computing

Several buyers with different utility functions and moneys.

Page 14: Algorithmic Game Theory and Internet Computing

Several buyers with different utility functions and moneys.

Find equilibrium prices.

1p 2p3p

Page 15: Algorithmic Game Theory and Internet Computing

Combinatorial Algorithm for Linear Case of Fisher’s Model

Devanur, Papadimitriou, Saberi & V., 2002

Using primal-dual paradigm

Page 16: Algorithmic Game Theory and Internet Computing

Combinatorial Algorithm for Linear Case of Fisher’s Model

Devanur, Papadimitriou, Saberi & V., 2002

Using primal-dual paradigm

Solves Eisenberg-Gale convex program

Page 17: Algorithmic Game Theory and Internet Computing

Eisenberg-Gale Program, 1959

max log

. .

:

: 1

: 0

i ii

i ij ijj

iji

ij

m v

s t

i v

j

ij

u xx

x

Page 18: Algorithmic Game Theory and Internet Computing

Eisenberg-Gale Program, 1959

max log

. .

:

: 1

: 0

i ii

i ij ijj

iji

ij

m v

s t

i v

j

ij

u xx

x

prices pj

Page 19: Algorithmic Game Theory and Internet Computing

Why remarkable?

Equilibrium simultaneously optimizes

for all agents.

How is this done via a single objective function?

Page 20: Algorithmic Game Theory and Internet Computing

Why seek combinatorial algorithms?

Page 21: Algorithmic Game Theory and Internet Computing

Why seek combinatorial algorithms?

Structural insightsHave led to progress on related problemsBetter understanding of solution concept

Useful in applications

Page 22: Algorithmic Game Theory and Internet Computing

Auction for Google’s TV ads

N. Nisan et. al, 2009:

Used market equilibrium based approach.

Combinatorial algorithms for linear case

provided “inspiration”.

Page 23: Algorithmic Game Theory and Internet Computing
Page 24: Algorithmic Game Theory and Internet Computing

utility

Piecewise linear, concave

amount of j

Additively separable over goods

Page 25: Algorithmic Game Theory and Internet Computing

Long-standing open problem

Complexity of finding an equilibrium for

Fisher and Arrow-Debreu models under

separable, plc utilities?

Page 26: Algorithmic Game Theory and Internet Computing

How do we build on solution to linear case?

Page 27: Algorithmic Game Theory and Internet Computing

utility

amount of j

Generalize EG program to piecewise-linear, concave utilities?

ijkl

ijkuutility/unit of j

Page 28: Algorithmic Game Theory and Internet Computing

,

,

max log

. .

:

: 1

:

: 0

i ii

i ijk ijkj k

iji k

ijk ijk

ijk

m v

s t

i v

j

ijk

ijk

u xx

x l

x

Generalization of EG program

Page 29: Algorithmic Game Theory and Internet Computing

,

,

max log

. .

:

: 1

:

: 0

i ii

i ijk ijkj k

iji k

ijk ijk

ijk

m v

s t

i v

j

ijk

ijk

u xx

x l

x

Generalization of EG program

Page 30: Algorithmic Game Theory and Internet Computing

Long-standing open problem

Complexity of finding an equilibrium for Fisher and Arrow-Debreu models under separable, plc utilities?

2009: Both PPAD-complete (using combinatorial insights from [DPSV])

Chen, Dai, Du, TengChen, TengV., Yannakakis

Page 31: Algorithmic Game Theory and Internet Computing

utility

Piecewise linear, concave

amount of j

Additively separable over goods

Page 32: Algorithmic Game Theory and Internet Computing

What makes linear utilities easy?

Weak gross substitutability:

Increasing price of one good cannot

decrease demand of another.

Piecewise-linear, concave utilities do not

satisfy this.

Page 33: Algorithmic Game Theory and Internet Computing

rate

rate = utility/unit amount of j

amount of j

Differentiate

Page 34: Algorithmic Game Theory and Internet Computing

rate

amount of j

rate = utility/unit amount of j

money spent on j

Page 35: Algorithmic Game Theory and Internet Computing

rate

rate = utility/unit amount of j

money spent on j

Spending constraint utility function

$20 $40 $60

Page 36: Algorithmic Game Theory and Internet Computing

Theorem (V., 2002): Spending constraint utilities: 1). Satisfy weak gross substitutability

2). Polynomial time algorithm for computing equilibrium.

Page 37: Algorithmic Game Theory and Internet Computing

An unexpected fallout!!

Has applications to

Google’s AdWords Market!

Page 38: Algorithmic Game Theory and Internet Computing

rate

rate = utility/click

money spent on keyword j

Application to Adwords market

$20 $40 $60

Page 39: Algorithmic Game Theory and Internet Computing

Is there a convex program for this model?

“We believe the answer to this question should be ‘yes’. In our experience, non-trivial polynomial time algorithms for problems are rare and happen for a good reason – a deep mathematical structure intimately connected to the problem.”

Page 40: Algorithmic Game Theory and Internet Computing

,

max log

. .

:

:

: 0

ijij

i j j

j iji

ij ij

ij

ub

p

s t

j p b

i b m

ij b

Devanur’s program for linear Fisher

Page 41: Algorithmic Game Theory and Internet Computing

max log

. .

:

:

:

: 0

ijkijk

ijk j

j ijkik

ijk ijk

ijk ijk

ijk

ub

p

s t

j p b

i b m

ijk b l

ijk b

C. P. for spending constraint!

Page 42: Algorithmic Game Theory and Internet Computing

EG convex program = Devanur’s program

Fisher marketwith plc utilities

Spending constraint market

Page 43: Algorithmic Game Theory and Internet Computing
Page 44: Algorithmic Game Theory and Internet Computing

Price discrimination markets

Business charges different prices from different

customers for essentially same goods or services.

Goel & V., 2009:

Perfect price discrimination market.

Business charges each consumer what

they are willing and able to pay.

Page 45: Algorithmic Game Theory and Internet Computing

plc utilities

Page 46: Algorithmic Game Theory and Internet Computing

Middleman buys all goods and sells to buyers,

charging according to utility accrued.Given p, there is a well defined rate for each buyer.

Page 47: Algorithmic Game Theory and Internet Computing

Middleman buys all goods and sells to buyers,

charging according to utility accrued.Given p, there is a well defined rate for each buyer.

Equilibrium is captured by a convex program Efficient algorithm for equilibrium

Page 48: Algorithmic Game Theory and Internet Computing

Middleman buys all goods and sells to buyers,

charging according to utility accrued.Given p, there is a well defined rate for each buyer.

Equilibrium is captured by a convex program Efficient algorithm for equilibrium

Market satisfies both welfare theorems!

Page 49: Algorithmic Game Theory and Internet Computing

,

,

max log

. .

:

: 1

:

: 0

i ii

i ijk ijkj k

iji k

ijk ijk

ijk

m v

s t

i v

j

ijk

ijk

u xx

x l

x

Generalization of EG program works!

Page 50: Algorithmic Game Theory and Internet Computing

EG convex program = Devanur’s program

Price discrimination market(plc utilities)

Spending constraint market

Page 51: Algorithmic Game Theory and Internet Computing
Page 52: Algorithmic Game Theory and Internet Computing

Nash bargaining game, 1950

Captures the main idea that both players

gain if they agree on a solution.

Else, they go back to status quo.

Page 53: Algorithmic Game Theory and Internet Computing

Example

Two players, 1 and 2, have vacation homes:

1: in the mountains

2: on the beach

Consider all possible ways of sharing.

Page 54: Algorithmic Game Theory and Internet Computing

Utilities derived jointly

1v

2v

S : convex + compact

feasible set

Page 55: Algorithmic Game Theory and Internet Computing

Disagreement point = status quo utilities

1v

2v

1c

2c

S

Disagreement point = 1 2( , )c c

Page 56: Algorithmic Game Theory and Internet Computing

Nash bargaining problem = (S, c)

1v

2v

1c

2c

S

Disagreement point = 1 2( , )c c

Page 57: Algorithmic Game Theory and Internet Computing

Nash bargaining

Q: Which solution is the “right” one?

Page 58: Algorithmic Game Theory and Internet Computing

Solution must satisfy 4 axioms:

Pareto optimality

Invariance under affine transforms

Symmetry

Independence of irrelevant alternatives

Page 59: Algorithmic Game Theory and Internet Computing

Thm: Unique solution satisfying 4 axioms

1 2( , ) 1 1 2 2( , ) max {( )( )}v v SN S c v c v c

1v

2v

1c

2c

S

Page 60: Algorithmic Game Theory and Internet Computing

Generalizes to n-players

Theorem: Unique solution

1 1( , ) max {( ) ... ( )}v S n nN S c v c v c

Page 61: Algorithmic Game Theory and Internet Computing

Nash bargaining solution is

optimal solution to convex program:

max log( )

. .

i ii

v c

s t v S

Page 62: Algorithmic Game Theory and Internet Computing

Nash bargaining solution is

optimal solution to convex program:

Polynomial time separation oracle

max log( )

. .

i ii

v c

s t v S

Page 63: Algorithmic Game Theory and Internet Computing

Q: Compute solution combinatoriallyin polynomial time?

Page 64: Algorithmic Game Theory and Internet Computing
Page 65: Algorithmic Game Theory and Internet Computing

How should they exchange their goods?

Page 66: Algorithmic Game Theory and Internet Computing

State as a Nash bargaining game

: (.,.,.)

: (.,.,.)

: (.,.,.)

f

b

m

u R

u R

u R

(1, 0,0)

(0, 1,0)

(0, 0,1)

f f

b b

m m

c u

c u

c u

S = utility vectors obtained by distributing goods among players

Page 67: Algorithmic Game Theory and Internet Computing

Special case: linear utility functions

: (.,.,.)

: (.,.,.)

: (.,.,.)

f

b

m

u R

u R

u R

(1, 0,0)

(0, 1,0)

(0, 0,1)

f f

b b

m m

c u

c u

c u

S = utility vectors obtained by distributing goods among players

Page 68: Algorithmic Game Theory and Internet Computing

ADNB

B: n players with disagreement points, ci

G: g goods, unit amount each

S = utility vectors obtained by distributing

goods among players

0i ij ij ijj G

v u x x

Page 69: Algorithmic Game Theory and Internet Computing

Convex program for ADNB

max log( )

. .

:

: 1

: 0

i ii

i ij ijj

iji

ij

v c

s t

i v

j

ij

u xx

x

Page 70: Algorithmic Game Theory and Internet Computing

Theorem (V., 2008)

Nash bargaining program is rational.

Page 71: Algorithmic Game Theory and Internet Computing

Theorem (V., 2008)

Nash bargaining solution is rational.

Combinatorial polynomial time algorithm

for finding it.

Page 72: Algorithmic Game Theory and Internet Computing

Game-theoretic properties of NB games -- “stress tests”

Chakrabarty, Goel, V. , Wang & Yu, 2008:

Efficiency (Price of bargaining)Fairness Full competitiveness

Page 73: Algorithmic Game Theory and Internet Computing

An application (Lucent)

“fair” throughput problem on a

wireless channel.

Page 74: Algorithmic Game Theory and Internet Computing

EG convex program = Devanur’s program

Price disc. market

Spending constraint market

ADNB

Page 75: Algorithmic Game Theory and Internet Computing

EG convex program = Devanur’s program

Price disc. market

Spending constraint market

Kelly, 1997: proportional fairness Jain & V., 2007: Eisenberg-Gale markets

ADNB

Page 76: Algorithmic Game Theory and Internet Computing

A new development

Orlin, 2009: Strongly polynomial algorithm

for Fisher’s linear case.

Open: For rest.

Page 77: Algorithmic Game Theory and Internet Computing

AGT’s gift to theory of algorithms!

New complexity classes: PPAD, FIXPStudy complexity of total problems

A new algorithmic directionCombinatorial algorithms for convex programs

Page 78: Algorithmic Game Theory and Internet Computing

Nonlinear programs with rational solutions!

Open

Page 79: Algorithmic Game Theory and Internet Computing

Nonlinear programs with rational solutions!

Solvable combinatorially!!

Open

Page 80: Algorithmic Game Theory and Internet Computing

Primal-Dual Paradigm

Combinatorial Optimization (1960’s & 70’s):

Integral optimal solutions to LP’s

Page 81: Algorithmic Game Theory and Internet Computing

Primal-Dual Paradigm

Combinatorial Optimization (1960’s & 70’s):

Integral optimal solutions to LP’s

Approximation Algorithms (1990’s):

Near-optimal integral solutions to LP’s

Page 82: Algorithmic Game Theory and Internet Computing

Primal-Dual Paradigm

Combinatorial Optimization (1960’s & 70’s):

Integral optimal solutions to LP’s

Approximation Algorithms (1990’s):

Near-optimal integral solutions to LP’s

Algorithmic Game Theory (New Millennium):

Rational solutions to nonlinear convex programs

Page 83: Algorithmic Game Theory and Internet Computing

Primal-Dual Paradigm

Combinatorial Optimization (1960’s & 70’s):

Integral optimal solutions to LP’s

Approximation Algorithms (1990’s):

Near-optimal integral solutions to LP’s

Algorithmic Game Theory (New Millennium):

Rational solutions to nonlinear convex programs

Approximation algorithms for convex programs?!

Page 84: Algorithmic Game Theory and Internet Computing
Page 85: Algorithmic Game Theory and Internet Computing

Extending primal-dual paradigm to framework of

convex programs and KKT conditions

Page 86: Algorithmic Game Theory and Internet Computing

Eisenberg-Gale Program, 1959

max ( ) log

. .

:

: 1

: 0

ii

i ij ijj

iji

ij

m i u

s t

i u

j

ij

u xx

x

Page 87: Algorithmic Game Theory and Internet Computing

Main point of departure

Complementary slackness conditions:

involve primal or dual variables, not both.

KKT conditions: involve primal and dual

variables simultaneously.

Page 88: Algorithmic Game Theory and Internet Computing

KKT conditions

1. : 0

2. : 0 1

3. , :( )

4. , : 0( )

j

j iji

ij i

j

ij ijij jij

j

j p

j p x

u ui j

p m i

u xui j x

p m i

Page 89: Algorithmic Game Theory and Internet Computing

Primal-dual algorithms so far

Raise dual variables greedily. (Lot of effort spent

on designing more sophisticated dual processes.)

Page 90: Algorithmic Game Theory and Internet Computing

Primal-dual algorithms so far

Raise dual variables greedily. (Lot of effort spent

on designing more sophisticated dual processes.)

Only exception: Edmonds, 1965: algorithm

for weight matching.

Page 91: Algorithmic Game Theory and Internet Computing

Primal-dual algorithms so far

Raise dual variables greedily. (Lot of effort spent

on designing more sophisticated dual processes.)

Only exception: Edmonds, 1965: algorithm

for weight matching.

Otherwise primal objects go tight and loose.

Difficult to account for these reversals

in the running time.

Page 92: Algorithmic Game Theory and Internet Computing

Our algorithms

Dual variables (prices) are raised greedily

Page 93: Algorithmic Game Theory and Internet Computing

Our algorithms

Dual variables (prices) are raised greedily

Yet, primal objects go tight and looseBecause of enhanced KKT conditions

Page 94: Algorithmic Game Theory and Internet Computing

Our algorithms

Dual variables (prices) are raised greedily

Yet, primal objects go tight and looseBecause of enhanced KKT conditions

New algorithmic ideas are needed!


Recommended