Mechanism Design and Fair Allocation Problems
Transcript of Mechanism Design and Fair Allocation Problems
![Page 1: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/1.jpg)
COST IC1205 Summer School on Fair Division
Grenoble, France, 13-17 July, 2015
Mechanism Design and Fair Allocation Problems
Gianluigi Greco
![Page 2: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/2.jpg)
Social Choice Theory
A > C > B
B > A > C
C > B > A
A > B > CRule for breaking ties: Alternatives
{ A, B, C }
Social Choice Function:
Compute the alternative that is
top-ranked by the majority
![Page 3: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/3.jpg)
Social Choice Theory Mechanism Design
A > C > B
B > A > C
C > B > A
A > B > CRule for breaking ties: Alternatives
{ A, B, C }
Social Choice Function:
Compute the alternative that is
top-ranked by the majority
Strategic issues!
B
![Page 4: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/4.jpg)
Mechanism Design
Social Choice Theory is non-strategic
In practice, agents declare their preferences
They are self interested
They might not reveal their true preferences
We want to find optimal outcomes w.r.t. true preferences
Optimizing w.r.t. the declared preferences might not
achieve the goal
![Page 5: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/5.jpg)
Outline
Mechanism Design
Mechanisms with Verification
Game Theory
Mechanisms and Allocation Problems
Complexity Analysis
![Page 6: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/6.jpg)
Basic Concepts (1/2)
Each agent is associated with a type
Each agent has a strategy
private knowledge, preferences,…
C > B > A
the action manifested
C > B > A
![Page 7: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/7.jpg)
Basic Concepts (2/2)
Consider the vector of the joint strategies
Each agent gets some utility
(A, B, C)
C > B > A
(A, B, C)
A
A
3 2 1
1
![Page 8: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/8.jpg)
Game Theory (by Example)
Consider the utility function of agent
Let us reason on the case where
selects A
selects B
A B A A
C > B > A3 2 1
1
A B B B
A B C A 1
2will select B
![Page 9: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/9.jpg)
Game Theory (by Example)
A > C > B
B > A > C
C > B > A2
3
1
![Page 10: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/10.jpg)
Solution Concepts
A Nash equilbrium is a strategy profile
such that, for every agent and for every ,
The strategies of the other agents are fixed…
![Page 11: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/11.jpg)
Solution Concepts
A Nash equilbrium is a strategy profile
such that, for every agent and for every ,
2 0
0 1
out
John goes outBob
home
John stays at home
1 1
0 0
out
Bob goes outJohn
home
Bob stays at home
![Page 12: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/12.jpg)
A Closer Look
To play a Nash equilibrium,
every agent must have perfect information
rationality is common knowledge
all agents must select the same Nash equilibrium
2 0
0 1
out
John goes outBob
home
John stays at home
1 1
0 0
out
Bob goes outJohn
home
Bob stays at home
Dominant strategy
![Page 13: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/13.jpg)
Dominant Strategies (by Example)
A > C > B
B > A > C
C > B > A
![Page 14: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/14.jpg)
Solution Concepts
A Nash equilbrium is a strategy profile
A strategy is dominant for agent , if for every
such that, for every agent and for every ,
and for every ,
Independently on the other agents…
![Page 15: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/15.jpg)
Outline
Mechanism Design
Mechanisms with Verification
Game Theory
Mechanisms and Allocation Problems
Complexity Analysis
![Page 16: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/16.jpg)
Social Choice Functions
A social choice function
given a type vector
selects an outcome
A > C > B
B > A > C
C > B > A
type vector
A
Social Choice Function:
Compute the alternative that is
top-ranked by the majority
Break ties: A > B > C
outcome
![Page 17: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/17.jpg)
Mechanism Design
Outcome Rule
Utility
strategy profile
outcome in
Social Choice Function
equilibrium
![Page 18: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/18.jpg)
Mechanism Design
Outcome Rule
Utility
strategy profile
outcome in
Social Choice Function
equilibrium
![Page 19: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/19.jpg)
Mechanism and Implementation
type vector 1 outcome 1 type vector 2 outcome 2
social choice function
![Page 20: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/20.jpg)
Mechanism and Implementation
type vector 1 type vector 2 outcome 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
For a given type vector, all startegy profiles are in principle admissible
A
![Page 21: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/21.jpg)
Mechanism and Implementation
type vector 1 type vector 2 outcome 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
For a given type vector, all startegy profiles are in principle admissible
An outcome rule is applied
A A CA B
A
![Page 22: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/22.jpg)
Mechanism and Implementation
type vector 1 type vector 2 outcome 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
For a given type vector, all startegy profiles are in principle admissible
An outcome rule is applied
So, utilities can be computed and equilibria can be selected
A A CA B (3,3,2)
A
![Page 23: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/23.jpg)
Mechanism and Implementation
type vector 1 type vector 2 outcome 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
A A CA B
A
(3,3,2)
![Page 24: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/24.jpg)
Mechanism and Implementation
type vector 1 type vector 2 outcome 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
A A AA A
A
![Page 25: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/25.jpg)
Mechanism and Implementation
type vector 1 type vector 2
social choice function
(A, A, A) (A, A, B) (A, B, A) (A, B, B) (C, C, C)
strategy profiles
A A AA A
A C
![Page 26: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/26.jpg)
Mechanism and Implementation
A mechanism is a tuple , where
for each agent , is the set of available strategies
is an outcome rule that
given a strategy profile
selects an outcome
implements in dominant strategy the social choice function if,
for each type vector ,
where is a dominant strategy.
![Page 27: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/27.jpg)
Types VS Strategies
In a direct revelation mechanism, each strategy is
restricted to a declaration about the private type
C > B > A
type
strategy
![Page 28: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/28.jpg)
Types VS Strategies
In a direct revelation mechanism, each strategy is
restricted to a declaration about the private type
Outcome Rule
true types
declared types
Utility
![Page 29: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/29.jpg)
Types VS Strategies
Outcome Rule
true types
declared types
Utility
DEFINITION. A direct-revelation mechanism is strategy-proof (dominant-strategy
incentive-compatible) if truth-revelation is a dominant strategy for each agent.
If the mechanism implements a function , then =
![Page 30: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/30.jpg)
Revelation Principle
It is a central theoretical tool in mechanism design
[Gibbard, 1973]
[Green and Laffont, 1977]
[Mayerson, 1979]
THEOREM. If a social choice function can be implemented in dominant strategies,
then it can be implemented by a strategy-proof direct-revelation mechanism.
C > B > A
type
strategy
![Page 31: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/31.jpg)
Impossibility Result
A social choice function is dictatorial if one agent
always receives one of its most preferred alternatives
![Page 32: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/32.jpg)
Impossibility Result
A social choice function is dictatorial if one agent
always receives one of its most preferred alternatives
A preference relation is general when it defines a
complete and transitive ordering over the alternatives
![Page 33: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/33.jpg)
Impossibility Result
Very bad news...
[Gibbard, 1973] and [Satterthwaite, 1975]
…, but must be interpreted with care
THEOREM. Assume general preferences, at least two agents, and at least three
optimal outcomes. A social choice function can be implemented in dominant
strategies if, and only if, it is dictatorial.
The result does not necessarily hold in restricted environments
![Page 34: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/34.jpg)
Payments
A utility is quasi-linear if it has the following form
Monetary compensation to induce truthfulness
valuation function
cardinal preferences
payment by the agent
![Page 35: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/35.jpg)
Payments
A utility is quasi-linear if it has the following form
Payments are defined by the mechanism
Monetary compensation to induce truthfulness
valuation function
cardinal preferences
payment by the agent
![Page 36: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/36.jpg)
Direct Mechanisms with Payments
Social Choice Function
Utility
Actually, agents might directly declare their valuations
![Page 37: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/37.jpg)
Direct Mechanisms with Payments
Social Choice Function =
![Page 38: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/38.jpg)
Vickrey-Clarke-Groves (VCG) Mechanisms
Consider quasi-linear utilities:
Consider social choice functions that are efficient:
Given , maximizes the sum of the valuations
The mechanism selects the outcome maximizing
Payments are such that
Family of mechanisms (e.g., the value of the optimal outcome without the agent)
![Page 39: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/39.jpg)
Vickrey-Clarke-Groves (VCG) Mechanisms
The mechanism selects the outcome maximizing
Payments are such that
Family of mechanisms (e.g., the value of the optimal outcome without the agent)
An auction with one item
We have bids: b1 > b2 > > bn
Agent 1 receives the item
Agent 1 pays b2
![Page 40: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/40.jpg)
Payment Rules (Again…)
Monetary compensation to induce truthfulness
The algebraic sum of the monetary transfers is zero
In particular, mechanisms cannot run into deficit
Monetary compensation to induce fairness
For instance, it is desirable that no agent envies the
allocation of any another agent, or that
The outcome is Pareto efficient, i.e., there is no
different allocation such that every agent gets at
least the same utility and one of them improves.
![Page 41: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/41.jpg)
Fairness vs Efficiency
25 26
2526
Two optimal allocations
Is there any fair allocation?
![Page 42: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/42.jpg)
(A Few…) Impossibility Results
Efficiency + Truthfulness + Budget Balance
Fairness + Truthfulness + Budget Balance
[Green, Laffont; 1977]
[Hurwicz; 1975]
[Tadenuma, Thomson;1995]
[Alcalde, Barberà; 1994]
[Andersson, Svensson, Ehlers; 2010]
Allocation
AlgorithmMechanismAllocation
declarations
payments
![Page 43: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/43.jpg)
Outline
Mechanism Design
Mechanisms with Verification
Game Theory
Mechanisms and Allocation Problems
Complexity Analysis
![Page 44: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/44.jpg)
(A Few…) Impossibility Results
Allocation
AlgorithmMechanismAllocation
declarations
Efficiency + Truthfulness + Budget Balance
Fairness + Truthfulness + Budget Balance
payments
![Page 45: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/45.jpg)
(A Few…) Impossibility Results
Verification on «selected» declarations
Allocation
AlgorithmMechanismAllocation
declarations
Verifier
Efficiency + Truthfulness + Budget Balance
Fairness + Truthfulness + Budget Balance
payments
![Page 46: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/46.jpg)
Approaches to Verification
![Page 47: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/47.jpg)
Approaches to Verification
[Green, Laffont; 1986]
[Nisan, Ronen; 2001]
![Page 48: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/48.jpg)
Approaches to Verification
[Auletta, De Prisco, Ferrante, Krysta, Parlato, Penna,
Persiano, Sorrentino, Ventre]
![Page 49: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/49.jpg)
Approaches to Verification
[Auletta, De Prisco, Ferrante, Krysta, Parlato, Penna,
Persiano, Sorrentino, Ventre]
[Caragiannis, Elkind, Szegedy, Yu; 2012]
![Page 50: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/50.jpg)
Approaches to Verification
Punishments are
used to enforce
truthfulness
![Page 51: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/51.jpg)
Approaches to Verification
Punishments are
used to enforce
truthfulness
Verification is performed via sensing
Hence, it is subject to errors; for instance, because of the limited precision of the measurement instruments.
It might be problematic to decide whether an observed discrepancy between verified values and declared ones is due to a strategic behavior or to such sensing errors.
[Greco, Scarcello; 2014]
![Page 52: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/52.jpg)
Approaches to Verification
Verification is performed via sensing
Hence, it is subject to errors; for instance, because of the limited precision of the measurement instruments.
It might be problematic to decide whether an observed discrepancy between verified values and declared ones is due to a strategic behavior or to such sensing errors.
3 Verifier 3.01
![Page 53: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/53.jpg)
Approaches to Verification (bis)
Agents might be uncertain of their private
features; for instance, due to limited
computational resources
There might be no strategic issues
3 Verifier 3.01
![Page 54: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/54.jpg)
Approaches to Verification (ter)
Punishments enforce truthfulness
They might be disproportional to the harm done by misreporting
Inappropriate in real life situations in which uncertainty is inherent due to measurements errors or uncertain inputs.
3 Verifier 3.01
[Feige, Tennenholtz; 2011]
100.000EUR
![Page 55: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/55.jpg)
Approaches to Verification
Punishments are
used to enforce
truthfulness
The verifier returns a value.
![Page 56: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/56.jpg)
Approaches to Verification
Punishments are
used to enforce
truthfulness
The verifier returns a value. But,…
no punishment
payments are always computed under the presumption of innocence, where incorrect declared values do not mean manipulation attempts by the agents
error tolerance
the consequences of errors in the declarations produce a linear “distorting effect” on the various properties of the mechanism
![Page 57: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/57.jpg)
Payment Rules
Monetary compensation to induce truthfulness
The algebraic sum of the monetary transfers is zero
In particular, mechanisms cannot run into deficit
Monetary compensation to induce fairness
For instance, it is desirable that no agent envies the
allocation of any another agent, or that
The outcome is Pareto efficient, i.e., there is no
different allocation such that every agent gets at
least the same utility and one of them improves.
![Page 58: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/58.jpg)
Payment Rules & Full Verification
Monetary compensation to induce truthfulness
The algebraic sum of the monetary transfers is zero
In particular, mechanisms cannot run into deficit
Monetary compensation to induce fairness
For instance, it is desirable that no agent envies the
allocation of any another agent, or that
The outcome is Pareto efficient, i.e., there is no
different allocation such that every agent gets at
least the same utility and one of them improves.
![Page 59: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/59.jpg)
Outline
Mechanisms and Allocation Problems
Mechanism Design
Mechanisms with Verification
Game Theory
Complexity Analysis
![Page 60: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/60.jpg)
The Model
Goods are indivisible and non-sharable
Constraints on the maximum number of goods to be allocated to each agent
Cardinal preferences: Utility functions
![Page 61: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/61.jpg)
The Model
Goods are indivisible and non-sharable
Constraints on the maximum number of goods to be allocated to each agent
Cardinal preferences: Utility functions
73
8 94
369
7
![Page 62: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/62.jpg)
The Model
Goods are indivisible and non-sharable
Constraints on the maximum number of goods to be allocated to each agent
Cardinal preferences: Utility functions
73
8 94 9
36
7
![Page 63: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/63.jpg)
The Model
Goods are indivisible and non-sharable
Constraints on the maximum number of goods to be allocated to each agent
Cardinal preferences: Utility functions
Social Welfare
Efficiency
73
8 94
369
7
![Page 64: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/64.jpg)
The Model
Goods are indivisible and non-sharable
Constraints on the maximum number of goods to be allocated to each agent
Cardinal preferences: Utility functions
Social Welfare
Efficiency
73
8 94
369
7
![Page 65: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/65.jpg)
Strategic Issues
Private Type
Social Welfare
Efficiency
![Page 66: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/66.jpg)
Strategic Issues
Private Type
Social Welfare
Efficiency
![Page 67: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/67.jpg)
Strategic Issues
Private Type
Social Welfare
Efficiency
![Page 68: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/68.jpg)
Strategic Issues: Example
Social Welfare
Efficiency
Before: 8+9=17
73
8 94
369
7
![Page 69: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/69.jpg)
Strategic Issues: Example
Social Welfare
Efficiency
1
Before: 8+9=17
73
8 94
369
7
1
1
![Page 70: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/70.jpg)
Strategic Issues: Example
Social Welfare
Efficiency
Before: 8+9=17
After: 9+7=16
17
38
1
1
94
369
7
![Page 71: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/71.jpg)
Strategic Issues: Verification
17
38
1
1
94
369
7
We assume full-verification.But, of course, we can verify only the goods that are selected.
![Page 72: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/72.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
![Page 73: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/73.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
![Page 74: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/74.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
![Page 75: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/75.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
![Page 76: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/76.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
Focus on an arbitrary coalition of agents
![Page 77: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/77.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
Focus on an arbitrary coalition of agents
![Page 78: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/78.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
Focus on an arbitrary coalition of agents
In this novel setting, compute an optimal allocation
![Page 79: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/79.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
Focus on an arbitrary coalition of agents
In this novel setting, compute an optimal allocation
![Page 80: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/80.jpg)
A Key Lemma
Consider an optimal allocation (w.r.t. some declared types)
Ignore the goods that are not allocated,
and hence that cannot be verified later…
Focus on an arbitrary coalition of agents
In this novel setting, compute an optimal allocation
The allocation is also optimal for that coalition, even if all
goods were actually available
![Page 81: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/81.jpg)
The Mechanism…
![Page 82: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/82.jpg)
The Mechanism…
Allocated goods are considered only
![Page 83: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/83.jpg)
The Mechanism…
Allocated goods are considered only
By the previous lemma, this is without loss of generality.
In fact, allocated goods are the only ones that we verify.
![Page 84: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/84.jpg)
The Mechanism…
«Bonus and Compensation»,
by Nisan and Ronen (2001)
Allocated goods are considered only
![Page 85: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/85.jpg)
The Mechanism…
«Bonus and Compensation»,
by Nisan and Ronen (2001)
Allocated goods are considered only
No punishments!
![Page 86: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/86.jpg)
The Mechanism…
«Bonus and Compensation»,
by Nisan and Ronen (2001)
Allocated goods are considered only
Truth-telling is a dominant strategy for each agent
![Page 87: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/87.jpg)
The Mechanism…
«Bonus and Compensation»,
by Nisan and Ronen (2001)
Allocated goods are considered only
Truth-telling is a dominant strategy for each agent
Does not depend on i
Is maximized when the declared type coincides
with the verified one
![Page 88: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/88.jpg)
The Mechanism…
«Bonus and Compensation»,
by Nisan and Ronen (2001)
Allocated goods are considered only
Truth-telling is a dominant strategy for each agent
![Page 89: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/89.jpg)
Coalitional Games
Players form coalitions
Each coalition is associated with a worth
A total worth has to be distributed
Solution Concepts characterize outcomes in terms ofFairness
Stability
![Page 90: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/90.jpg)
Coalitional Games: Shapley Value
Solution Concepts characterize outcomes in terms ofFairness
Stability
![Page 91: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/91.jpg)
Relevant Properties of the Shapley Value
Core Allocation
![Page 92: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/92.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
verified values
selected products
and
![Page 93: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/93.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
verified values
selected products
and
![Page 94: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/94.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
Each agent gets the Shapley value
verified values ( )
selected products
and
![Page 95: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/95.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
Each agent gets the Shapley value
Properties The resulting mechanism is «fair» and «buget balanced»
verified values ( )
selected products
and
![Page 96: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/96.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
Each agent gets the Shapley value
Properties The resulting mechanism is «fair» and «buget balanced»
verified values ( )
selected products
and
![Page 97: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/97.jpg)
The Mechanism
is the contribution of the coalition w.r.t.
Each agent gets the Shapley value
Properties The resulting mechanism is «fair» and «buget balanced»
verified values ( )
selected products
and
The game is supermodular;
so the Shapley value is stable
![Page 98: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/98.jpg)
Further Observations for Fairness
Let be an optimal allocation
Let be an allocation
![Page 99: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/99.jpg)
Further Observations for Fairness
Let be an optimal allocation
Let be an allocation
(best allocation for the coalition with products in )
As is optimal, then is in fact optimal even by
considering all possible products as available ≥
![Page 100: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/100.jpg)
Further Observations for Fairness
Let be an optimal allocation
Let be an allocation
(best allocation for the coalition with products in )
As is optimal, then is in fact optimal even by
considering all possible products as available ≥
By the monotonicity of the Shapley value, ≥
![Page 101: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/101.jpg)
Further Observations for Fairness
Let be an optimal allocation
Let be an allocation
Optimal allocations are always preferred by ALL agents
There is no difference between two different optimal allocations
≥
![Page 102: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/102.jpg)
Further Observations for Fairness
Let be an optimal allocation
Let be an allocation
Optimal allocations are always preferred by ALL agents
There is no difference between two different optimal allocations
≥
Fairness
![Page 103: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/103.jpg)
Other Solution Concepts?
![Page 104: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/104.jpg)
Other Solution Concepts?
How fairness/stability can be measured?
The excess is a measure of the dissatisfaction of S
![Page 105: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/105.jpg)
Other Solution Concepts?
How fairness/stability can be measured?
The excess is a measure of the dissatisfaction of S
![Page 106: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/106.jpg)
Excess…
How fairness/stability can be measured?
The excess is a measure of the dissatisfaction of S
![Page 107: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/107.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
![Page 108: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/108.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
![Page 109: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/109.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
![Page 110: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/110.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
![Page 111: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/111.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
![Page 112: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/112.jpg)
…and the Nucleolus
Arrange excess values in non-increasing order
[Schmeidler]
![Page 113: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/113.jpg)
Outline
Mechanisms and Allocation Problems
Mechanism Design
Mechanisms with Verification
Game Theory
Complexity Analysis
![Page 114: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/114.jpg)
Complexity Issues
For many classes of «compact games» (e.g., graph games),
the Shapley-value can be efficiently calculated
Here, the problem emerges to be #P-complete
![Page 115: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/115.jpg)
Complexity Issues
For many classes of «compact games» (e.g., graph games),
the Shapley-value can be efficiently calculated
Here, the problem emerges to be #P-complete
#P is the class the class of all functions that can be computed by counting
Turing machines in polynomial time.
A counting Turing machine is a standard nondeterministic Turing machine
with an auxiliary output device that prints in binary notation the number of
accepting computations induced by the input.
Prototypical problem: to count the number of truth variable assignments
that satisfy a Boolean formula.
![Page 116: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/116.jpg)
Complexity Issues
For many classes of «compact games» (e.g., graph games),
the Shapley-value can be efficiently calculated
Here, the problem emerges to be #P-complete
Reduction from the problem of counting the number of perfect matchings in certain bipartite graphs [Valiant, 1979]
#P is the class the class of all functions that can be computed by counting
Turing machines in polynomial time.
A counting Turing machine is a standard nondeterministic Turing machine
with an auxiliary output device that prints in binary notation the number of
accepting computations induced by the input.
Prototypical problem: to count the number of truth variable assignments
that satisfy a Boolean formula.
![Page 117: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/117.jpg)
Complexity Issues
#P-complete
However…
![Page 118: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/118.jpg)
Probabilistic Computation
#P-complete
However…
Always Efficient and Budget Balanced
All other properties in expectation (with high probability)
Coupling of the algorithm with a sampling strategy for the
coalitions by [Liben-Nowell,Sharp, Wexler, Woods; 2012]
![Page 119: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/119.jpg)
Probabilistic Computation
Coupling of the algorithm with a sampling strategy for the
coalitions by [Liben-Nowell,Sharp, Wexler, Woods; 2012]
Use sampling, rather than exaustive search.
![Page 120: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/120.jpg)
Back to Exact Computation: Islands of Tractability
Can we find classes of instances for
«allocation games» over which the Shapley
value can be efficiently computed?
![Page 121: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/121.jpg)
Back to Exact Computation: Islands of Tractability
Can we find classes of instances for
«allocation games» over which the Shapley
value can be efficiently computed?
Utility functions
Values taken from specific domains
For instance, use k values at most #P-complete, even for k=2
[G., Lupia and Scarcello; 2015]
![Page 122: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/122.jpg)
Back to Exact Computation: Islands of Tractability
Can we find classes of instances for
«allocation games» over which the Shapley
value can be efficiently computed?
Utility functions
Values taken from specific domains
For instance, use k values at most
Structural restrictions…
#P-complete, even for k=2
![Page 123: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/123.jpg)
Bounded Sharing Degree
Sharing degree
Maximum number of agents competing for the same good
Sharing degree = 2
![Page 124: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/124.jpg)
Bounded Sharing Degree
Sharing degree
Maximum number of agents competing for the same good
Sharing degree = 2
The Shapley value can be computed in polynomial
time whenever the sharing degree is 2 at most.
![Page 125: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/125.jpg)
Bounded Interactions
![Page 126: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/126.jpg)
Bounded Interactions
Interaction graph
There is an edge between any pair of agents competing for the same good
![Page 127: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/127.jpg)
Bounded Interactions
Interaction graph
There is an edge between any pair of agents competing for the same good
The Shapley value can be computed in polynomial
time whenever the interaction graph is a tree.
or, more generally, if it has bounded treewidth
![Page 128: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/128.jpg)
g
q
ab
f
c
d
ph
l
nm
ok
e
i
j
Tree Decompositions [Robertson & Seymour ‘86]
![Page 129: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/129.jpg)
Tree Decompositions [Robertson & Seymour ‘86]
![Page 130: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/130.jpg)
Tree Decompositions [Robertson & Seymour ‘86]
• Every edge realized in some bag
• Connectedness condition
![Page 131: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/131.jpg)
Connectedness condition for h
ah
ahq
hij abc
hkl
hkp klo
mno
bcdcefag
![Page 132: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/132.jpg)
tw(acyclic graph)=1
tw(cycle) = 2
tw(G+v) tw(G)+1
tw(G+e) tw(G)+1
tw(Kn) = n-1
tw is fixed-parameter tractable (parameter: treewidth)
Properties of Treewidth
![Page 133: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/133.jpg)
Bounded Interactions
Interaction graph
There is an edge between any pair of agents competing for the same good
The Shapley value can be computed in polynomial
time whenever the interaction graph is a tree.
or, more generally, if it has bounded treewidth
![Page 134: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/134.jpg)
Proof Idea: Ingredient 1
list the values in increasing order: w1,…,wm
, where
is the number of coalitions such that and
![Page 135: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/135.jpg)
Proof Idea: Ingredient 2
is the number of coalitions such that and
The marginal contribution can be characterized via the
existence of an allocation with certain properties
![Page 136: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/136.jpg)
Proof Idea: Ingredient 2
is the number of coalitions such that and
The marginal contribution can be characterized via the
existence of an allocation with certain properties
agent i
interaction graph
![Page 137: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/137.jpg)
Proof Idea: Ingredient 2
is the number of coalitions such that and
The marginal contribution can be characterized via the
existence of an allocation with certain properties
agent i
interaction graph
restricted w.r.t.
![Page 138: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/138.jpg)
Proof Idea: Ingredient 2
is the number of coalitions such that and
agent i
interaction graph
restricted w.r.t.
there is an allocation in the scenario induced over where each
agent gets a good with value at least
![Page 139: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/139.jpg)
Proof Idea: Ingredient 2
Keep only goods with the desired value
Focus on the induced scenario
The problem reduces to counting the number of
coalitions with size h for which each agent can get a good
is the number of coalitions such that and
there is an allocation in the scenario induced over where each
agent gets a good with value at least
![Page 140: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/140.jpg)
Proof Idea: Ingredient 3
The problem reduces to counting the number of
coalitions with size h for which each agent can get a good
![Page 141: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/141.jpg)
CSPs: Informal Definition
Variables:
A, B, C, D, and E
Domain:
RGB = {red, green, blue}
Constraints:
AB, AC,A E, A D, B C, C D, D E
E
D A
CB
E
D A
CB
![Page 142: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/142.jpg)
CSPs: Informal Definition
Variables:
A, B, C, D, and E
Domain:
D(A) = D(B) = D(C) = D(D) = D(E) = {red, green, blue}
Constraints:
AB; AC; A E; A D; B C; C D; D E
E
D A
C
Bprimal graph
![Page 143: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/143.jpg)
Example Encoding
![Page 144: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/144.jpg)
A B C
Variables:
Agent A, agent B, and agent C +
Domain:
D(A) =
D(B) =
D(C) =
Constraints:
AB; BC; X= if, and only if, INX=false
variables INA, INB, INC
boolean: {true, false}
Example Encoding
![Page 145: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/145.jpg)
A B C
INA= INB = true
INC = false
The problem reduces to counting the number of coalitions
with size h for which each agent can get a good
Example Encoding
![Page 146: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/146.jpg)
Proof Idea: Ingredient 3
The problem reduces to counting the number of
coalitions with size h for which each agent can get a good
![Page 147: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/147.jpg)
Proof Idea: Ingredient 3
The problem reduces to counting the number of
coalitions with size h for which each agent can get a good
Decision problems
Computation Problems
Counting?
in «Tractability: Practical Approaches to hard Problems»
[Gottlob, Greco, Scarcello, 2013]
![Page 148: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/148.jpg)
Proof Idea: Ingredient 3
Decision problems
Computation Problems
Counting?
Solutions projected over a set W of
output variables
Variables not in W are auxiliary ones
![Page 149: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/149.jpg)
Proof Idea: Ingredient 3
Decision problems
Computation Problems
Counting?
Solutions projected over a set W of
output variables
Variables not in W are auxiliary ones
For instance, we cannot use a variable to denote the allocation for an agent,
since its domain would be unbounded!
![Page 150: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/150.jpg)
Proof Idea: Ingredient 3
Decision problems
Computation Problems
Counting?
Solutions projected over a set W of
output variables
Variables not in W are auxiliary ones
For instance, we cannot use a variable to denote the allocation for an agent,
since its domain would be unbounded!
![Page 151: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/151.jpg)
Proof Idea: Ingredient 3
For instance, we cannot use a variable to denote the allocation for an agent,
since its domain would be unbounded!
Usually,
Build the CSP
Compute a decomposition
Use structural tractability results
Here
Compute a decomposition
Build the CSP based on the decomposition
Recompute the decomposition
Use structural tractability results
![Page 152: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/152.jpg)
Proof Idea: Ingredient 3
For instance, we cannot use a variable to denote the allocation for an agent,
since its domain would be unbounded!
(variable associated with) agent i
(variable associated with) good g
auxiliary variables encoding the roadmaps to reach the goods
![Page 153: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/153.jpg)
Proof Idea: Ingredient 3
For instance, we cannot use a variable to denote the allocation for an agent,
since its domain would be unbounded!
(variable associated with) agent i
(variable associated with) good g
auxiliary variables encoding the roadmaps to reach the goods
W.l.o.g. the tree is binary.
Hence, a few «road signs» suffices
![Page 154: Mechanism Design and Fair Allocation Problems](https://reader030.fdocuments.net/reader030/viewer/2022012711/61aaf5f56f16b550a303ff12/html5/thumbnails/154.jpg)
For references, see the bibliography of Mechanisms for Fair Allocation Problems [G. and Scarcello; JAIR 2014]