158
Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Embed Size (px)

Citation preview

Page 1: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Algorithmic Game Theoryand Internet Computing

Vijay V. Vazirani

Markets and

the Primal-Dual Paradigm

Page 2: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

The new face of computing

Page 3: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A paradigm shift inthe notion of a “market”!

Page 4: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 5: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 6: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 7: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Historically, the study of markets

has been of central importance,

especially in the West

Page 8: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Historically, the study of markets

has been of central importance,

especially in the West

General Equilibrium TheoryOccupied center stage in Mathematical

Economics for over a century

Page 9: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

General Equilibrium Theory

Also gave us some algorithmic resultsConvex programs, whose optimal solutions capture

equilibrium allocations,

e.g., Eisenberg & Gale, 1959

Nenakov & Primak, 1983

Page 10: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

General Equilibrium Theory

Also gave us some algorithmic resultsConvex programs, whose optimal solutions capture equilibrium allocations,

e.g., Eisenberg & Gale, 1959 Nenakov & Primak, 1983

Scarf, 1973: Algorithms for approximately computing fixed points

Page 11: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

New markets defined by Internet companies, e.g., Google Yahoo! Amazon eBay

Massive computing power available for running markets in a distributed or centralized manner

A deep theory of algorithms with many powerful techniques

Today’s reality

Page 12: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

What is needed today?

An inherently-algorithmic theory of

markets and market equilibria

Page 13: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

What is needed today?

An inherently-algorithmic theory of

markets and market equilibria

Beginnings of such a theory, within

Algorithmic Game Theory

Page 14: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

What is needed today?

An inherently-algorithmic theory of

markets and market equilibria

Beginnings of such a theory, within

Algorithmic Game Theory

Natural starting point:

algorithms for traditional market models

Page 15: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

What is needed today?

An inherently-algorithmic theory of markets and market equilibria

Beginnings of such a theory, within Algorithmic Game Theory

Natural starting point: algorithms for traditional market models

New market models emerging!

Page 16: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Theory of algorithms

Interestingly enough, recent study of

markets has contributed handsomely to

this theory!

Page 17: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A central tenet

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

equilibrium prices.

Page 18: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A central tenet

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

equilibrium prices.

Easy if only one good

Page 19: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Supply-demand curves

Page 20: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Irving Fisher, 1891

Defined a fundamental

market model

Page 21: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 22: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 23: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Utility function

amount of milk

Page 24: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Utility function

amount of bread

Page 25: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Utility function

amount of cheese

Page 26: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Total utility of a bundle of goods

= Sum of utilities of individual goods

Page 27: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

For given prices,

1p 2p3p

Page 28: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

For given prices,find optimal bundle of goods

1p 2p3p

Page 29: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Fisher market

Several goods, fixed amount of each good

Several buyers,

with individual money and utilities

Find equilibrium prices of goods, i.e., prices s.t., Each buyer gets an optimal bundle No deficiency or surplus of any good

Page 30: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 31: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Combinatorial Algorithm for Linear Case of Fisher’s Model

Devanur, Papadimitriou, Saberi & V., 2002

Using the primal-dual schema

Page 32: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Primal-Dual Schema

Highly successful algorithm design

technique from exact and

approximation algorithms

Page 33: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Exact Algorithms for Cornerstone Problems in P:

Matching (general graph) Network flow Shortest paths Minimum spanning tree Minimum branching

Page 34: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Approximation Algorithms

set cover facility location

Steiner tree k-median

Steiner network multicut

k-MST feedback vertex set

scheduling . . .

Page 35: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 36: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

No LP’s known for capturing equilibrium allocations for Fisher’s model

Page 37: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

No LP’s known for capturing equilibrium allocations for Fisher’s model

Eisenberg-Gale convex program, 1959

Page 38: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

No LP’s known for capturing equilibrium allocations for Fisher’s model

Eisenberg-Gale convex program, 1959

DPSV: Extended primal-dual schema to

solving a nonlinear convex program

Page 39: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Fisher’s Model

n buyers, money m(i) for buyer i k goods (unit amount of each good) : utility derived by i

on obtaining one unit of j Total utility of i,

i ij ijj

U u xiju

]1,0[

x

xuuij

ijj iji

Page 40: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Fisher’s Model

n buyers, money m(i) for buyer i k goods (unit amount of each good) : utility derived by i

on obtaining one unit of j Total utility of i,

Find market clearing prices

i ij ijj

U u xiju

]1,0[

x

xuuij

ijj iji

Page 41: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

At prices p, buyer i’s most

desirable goods, S =

Any goods from S worth m(i)

constitute i’s optimal bundle

arg max ijj

j

u

p

Bang-per-buck

Page 42: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A convex program

whose optimal solution is equilibrium allocations.

Page 43: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A convex program

whose optimal solution is equilibrium allocations.

Constraints: packing constraints on the xij’s

Page 44: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A convex program

whose optimal solution is equilibrium allocations.

Constraints: packing constraints on the xxij’s

Objective fn: max utilities derived.

Page 45: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A convex program

whose optimal solution is equilibrium allocations.

Constraints: packing constraints on the xxij’s

Objective fn: max utilities derived. Must satisfy

If utilities of a buyer are scaled by a constant,

optimal allocations remain unchangedIf money of buyer b is split among two new buyers,

whose utility fns same as b, then union of optimal

allocations to new buyers = optimal allocation for b

Page 46: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Money-weighed geometric mean of utilities

1/ ( )( )( ) im im i

i iu

Page 47: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

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 48: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

KKT conditions

1. : 0

2. : 0 1

3. , :( )

4. , : 0( )

j

j iji

ij i

j

ij iij

j

j p

j p x

u ui j

p m i

u ui j x

p m i

Page 49: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Therefore, buyer i buys from

only,

i.e., gets an optimal bundle

arg max ijj

j

u

p

Page 50: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Therefore, buyer i buys from

only,

i.e., gets an optimal bundle

Can prove that equilibrium prices

are unique!

arg max ijj

j

u

p

Page 51: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Idea of algorithm

“primal” variables: allocations

“dual” variables: prices of goods

Approach equilibrium prices from below:start with very low prices; buyers have surplus money iteratively keep raising prices and decreasing surplus

Page 52: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Idea of algorithm

Iterations:

execute primal & dual improvements

Allocations Prices

Page 53: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Will relax KKT conditions

e(i): money currently spent by i

w.r.t. a special allocation

surplus money of i( ) ( )i m i e i

Page 54: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

KKT conditions

1. : 0

2. : 0 1

3. , :( )

4. , : 0( )

j

j iji

ij i

j

ij iij

j

j p

j p x

u ui j

p m i

u ui j x

p m i

e(i)

e(i)

Page 55: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Potential function

2 2 21 2 ... n

Will show that potential drops by an inverse polynomial

factor in each phase (polynomial time).

Page 56: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Potential function

2 2 21 2 ... n

Will show that potential drops by an inverse polynomial

factor in each phase (polynomial time).( ( ))

ipoly m i

Page 57: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Point of departure

KKT conditions are satisfied via a

continuous process Normally: in discrete steps

Page 58: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Point of departure

KKT conditions are satisfied via a

continuous process Normally: in discrete steps

Open question: strongly polynomial algorithm??

Page 59: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

An easier question

Given prices p, are they equilibrium prices?

If so, find equilibrium allocations.

Page 60: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

An easier question

Given prices p, are they equilibrium prices?

If so, find equilibrium allocations.

Equilibrium prices are unique!

Page 61: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

m(1)

m(2)

m(3)

m(4)

p(1)

p(2)

p(3)

p(4)

For each buyer, most desirable goods, i.e.

arg max ijj

j

u

p

Page 62: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Max flow

m(1)

m(2)

m(3)

m(4)

p(1)

p(2)

p(3)

p(4)

infinite capacities

Page 63: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Max flow

m(1)

m(2)

m(3)

m(4)

p(1)

p(2)

p(3)

p(4)

p: equilibrium prices iff both cuts saturated

Page 64: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Two important considerations

The price of a good never exceeds

its equilibrium priceInvariant: s is a min-cut

Page 65: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Max flow

m(1)

m(2)

m(3)

m(4)

p(1)

p(2)

p(3)

p(4)

p: low prices

Page 66: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Two important considerations

The price of a good never exceeds

its equilibrium priceInvariant: s is a min-cut

Identify tight sets of goods

: ( ) ( ( ))S A p S m S

Page 67: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Two important considerations

The price of a good never exceeds

its equilibrium priceInvariant: s is a min-cutIdentify tight sets of goods

Rapid progress is madeBalanced flows

Page 68: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Network N

m p

buyers

goods

bang-per-buck edges

Page 69: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow in N

m p

W.r.t. flow f, surplus(i) = m(i) – f(i,t)

i

Page 70: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow

surplus vector: vector of surpluses w.r.t. f.

Page 71: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow

surplus vector: vector of surpluses w.r.t. f.

A flow that minimizes l2 norm of surplus vector.

Page 72: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow

surplus vector: vector of surpluses w.r.t. f.

A flow that minimizes l2 norm of surplus vector.

Must be a max-flow.

Page 73: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow

surplus vector: vector of surpluses w.r.t. f.

A flow that minimizes l2 norm of surplus vector.

Must be a max-flow.

All balanced flows have same surplus vector.

Page 74: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Balanced flow

surplus vector: vector of surpluses w.r.t. f.

A flow that minimizes l2 norm of surplus vector.

Makes surpluses as equal as possible.

Page 75: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Property 1

f: max flow in N.

R: residual graph w.r.t. f.

If surplus (i) < surplus(j) then there is no

path from i to j in R.

Page 76: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Property 1

i

surplus(i) < surplus(j)

j

R:

Page 77: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Property 1

i

surplus(i) < surplus(j)

j

R:

Page 78: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Property 1

i

Circulation gives a more balanced flow.

j

R:

Page 79: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Property 1

Theorem: A max-flow is balanced iff

it satisfies Property 1.

Page 80: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Will relax KKT conditions

e(i): money currently spent by i

w.r.t. a special allocation

surplus money of i( ) ( )i m i e i

Page 81: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Will relax KKT conditions

e(i): money currently spent by i

w.r.t. a balanced flow in N

surplus money of i( ) ( )i m i e i

Page 82: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Pieces fit just right!

Balanced flows Invariant

Bang-per-buck

edgesTight sets

Page 83: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Another point of departure

Complementary slackness conditions:

involve primal or dual variables, not both.

KKT conditions: involve primal and dual

variables simultaneously.

Page 84: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

KKT conditions

1. : 0

2. : 0 1

3. , :( )

4. , : 0( )

j

j iji

ij i

j

ij iij

j

j p

j p x

u ui j

p m i

u ui j x

p m i

Page 85: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

KKT conditions

1. : 0

2. : 0 1

3. , :( )

4. , : 0( ) ( )

j

j iji

ij i

j

ij ijij jiij

j

j p

j p x

u ui j

p m i

u xu ui j x

p m i m i

Page 86: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Primal-dual algorithms so far

Raise dual variables greedily. (Lot of effort spent

on designing more sophisticated dual processes.)

Page 87: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

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 88: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

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 89: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Our algorithm

Dual variables (prices) are raised greedily

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

Page 90: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Deficiencies of linear utility functions

Typically, a buyer spends all her money

on a single good

Do not model the fact that buyers get

satiated with goods

Page 91: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Concave utility function

amount of j

Page 92: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Concave utility functions

Do not satisfy weak gross substitutability

Page 93: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Concave utility functions

Do not satisfy weak gross substitutabilityw.g.s. = Raising the price of one good cannot lead to a

decrease in demand of another good.

Page 94: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Concave utility functions

Do not satisfy weak gross substitutabilityw.g.s. = Raising the price of one good cannot lead to a

decrease in demand of another good.

Open problem: find polynomial time algorithm!

Page 95: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Piecewise linear, concave

amount of j

Page 96: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

PTAS for concave function

amount of j

Page 97: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Piecewise linear concave utility

Does not satisfy weak gross substitutability

Page 98: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Piecewise linear, concave

amount of j

Page 99: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

rate

rate = utility/unit amount of j

amount of j

Differentiate

Page 100: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

rate

amount of j

rate = utility/unit amount of j

money spent on j

Page 101: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

rate

rate = utility/unit amount of j

money spent on j

Spending constraint utility function

$20 $40 $60

Page 102: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Spending constraint utility function

Happiness derived is

not a function of allocation only

but also of amount of money spent.

Page 103: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

$20 $40 $100

Extend model: assume buyers have utility for money

rate

Page 104: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 105: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Theorem: Polynomial time algorithm for

computing equilibrium prices and allocations for

Fisher’s model with spending constraint utilities.

Furthermore, equilibrium prices are unique.

Page 106: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Satisfies weak gross substitutability!

Page 107: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Old pieces become more complex+ there are new pieces

Page 108: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

But they still fit just right!

Page 109: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Don Patinkin, 1922-1995

Considered utility functions that are

a function of allocations and prices.

Page 110: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

An unexpected fallout!!

Page 111: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

An unexpected fallout!!

A new kind of utility functionHappiness derived is

not a function of allocation only

but also of amount of money spent.

Page 112: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

An unexpected fallout!!

A new kind of utility functionHappiness derived is

not a function of allocation only

but also of amount of money spent.

Has applications in

Google’s AdWords Market!

Page 113: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

A digression

Page 114: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords Market

Created by search engine companiesGoogleYahoo!MSN

Multi-billion dollar market – and still growing!

Totally revolutionized advertising, especially

by small companies.

Page 115: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

The view 5 years ago: Relevant Search Results

Page 116: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm
Page 117: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Business world’s view now :

(as Advertisement companies)

Page 118: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Bids for different keywords

DailyBudgets

So how does this work?

Page 119: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords Allocation Problem

Search Engine

Whose ad to put

How to maximize revenue?

LawyersRus.com

Sue.com

TaxHelper.com

asbestos

Search results

Ads

Page 120: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords Problem

Mehta, Saberi, Vazirani & Vazirani, 2005:

1-1/e algorithm, assuming budgets>>bids

Page 121: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords Problem

Mehta, Saberi, Vazirani & Vazirani, 2005:

1-1/e algorithm, assuming budgets>>bids

Optimal!

Page 122: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords Problem

Mehta, Saberi, Vazirani & Vazirani, 2005:

1-1/e algorithm, assuming budgets>>bids

Optimal!

Page 123: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Spending

constraint

utilities

AdWords

Market

Page 124: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords market

Assume that Google will determine equilibrium price/click for keywords

Page 125: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords market

Assume that Google will determine equilibrium price/click for keywords

How should advertisers specify their

utility functions?

Page 126: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Choice of utility function

Expressive enough that advertisers get

close to their ‘‘optimal’’ allocations

Page 127: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Choice of utility function

Expressive enough that advertisers get

close to their ‘‘optimal’’ allocations

Efficiently computable

Page 128: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Choice of utility function

Expressive enough that advertisers get

close to their ‘‘optimal’’ allocations

Efficiently computable

Easy to specify utilities

Page 129: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

linear utility function: a business will

typically get only one type of query

throughout the day!

Page 130: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

linear utility function: a business will

typically get only one type of query

throughout the day!

concave utility function: no efficient

algorithm known!

Page 131: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

linear utility function: a business will

typically get only one type of query

throughout the day!

concave utility function: no efficient

algorithm known!Difficult for advertisers to

define concave functions

Page 132: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Easier for a buyer

To say what are “good” allocations,

for a range of prices,

rather than how happy she is

with a given bundle.

Page 133: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Online shoe business

Interested in two keywords: men’s clog women’s clog

Advertising budget: $100/day

Expected profit:men’s clog: $2/clickwomen’s clog: $4/click

Page 134: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Considerations for long-term profit

Try to sell both goods - not just the most

profitable good

Must have a presence in the market,

even if it entails a small loss

Page 135: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

If both are profitable,better keyword is at least twice as profitable ($100, $0)otherwise ($60, $40)

If neither is profitable ($20, $0)

If only one is profitable, very profitable (at least $2/$) ($100, $0)otherwise ($60, $0)

Page 136: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

$60 $100

men’s clog

rate

2

1

rate = utility/click

Page 137: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

$60 $100

women’s clog

rate

2

4

rate = utility/click

Page 138: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

$80 $100

money

rate

0

1

rate = utility/$

Page 139: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords market

Suppose Google stays with auctions but

allows advertisers to specify bids in

the spending constraint model

Page 140: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords market

Suppose Google stays with auctions but

allows advertisers to specify bids in

the spending constraint model expressivity!

Page 141: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

AdWords market

Suppose Google stays with auctions but

allows advertisers to specify bids in

the spending constraint model expressivity!

Good online algorithm for

maximizing Google’s revenues?

Page 142: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Goel & Mehta, 2006:

A small modification to the MSVV algorithm

achieves 1 – 1/e competitive ratio!

Page 143: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Open

Is there a convex program that

captures equilibrium allocations for

spending constraint utilities?

Page 144: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Equilibrium exists (under mild conditions)

Equilibrium utilities and prices are unique

Rational

With small denominators

Spending constraint utilities satisfy

Page 145: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Equilibrium exists (under mild conditions)

Equilibrium utilities and prices are unique

Rational

With small denominators

Linear utilities also satisfy

Page 146: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Proof follows fromEisenberg-Gale Convex 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 147: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

For spending constraint utilities,proof follows from algorithm,

and not a convex program!

Page 148: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Open

Is there an LP whose optimal solutions

capture equilibrium allocations

for Fisher’s linear case?

Page 149: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Use spending constraint algorithm to solve

piecewise linear, concave utilities

Open

Page 150: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

utility

Piece-wise linear, concave

amount of j

ijf

Page 151: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

rate

rate = utility/unit amount of j

amount of j

Differentiate ijg

Page 152: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Start with arbitrary prices, adding up to

total money of buyers.

( ) ( )ij ijj

xh x g

p

Page 153: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

rate

money spent on j

rate = utility/unit amount of j

( ) ( )ij ijj

xh x g

p

Page 154: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Start with arbitrary prices, adding up to

total money of buyers.

Run algorithm on these utilities to get new prices.

( ) ( )ij ijj

xh x g

p

Page 155: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Start with arbitrary prices, adding up to

total money of buyers.

Run algorithm on these utilities to get new prices.

( ) ( )ij ijj

xh x g

p

Page 156: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Start with arbitrary prices, adding up to

total money of buyers.

Run algorithm on these utilities to get new prices.

Fixed points of this procedure are equilibrium

prices for piecewise linear, concave utilities!

( ) ( )ij ijj

xh x g

p

Page 157: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm

Algorithms & Game Theorycommon origins

von Neumann, 1928: minimax theorem for

2-person zero sum games von Neumann & Morgenstern, 1944:

Games and Economic Behavior von Neumann, 1946: Report on EDVAC

Dantzig, Gale, Kuhn, Scarf, Tucker …

Page 158: Algorithmic Game Theory and Internet Computing Vijay V. Vazirani Markets and the Primal-Dual Paradigm