second part: algorithmic mechanism design. implementation theory imagine a “planner” who...
Post on 21-Dec-2015
222 views
TRANSCRIPT
Implementation theory Imagine a “planner” who develops criteria
for social welfare, but cannot enforce the desirable allocation directly, as he lacks information about several parameters of the situation. A mean then has to be found to “implement” such criteria.
Subfield of economic theory with an engineering perspective: Designs economic mechanisms like computer scientists design algorithms, protocols, systems, …
The implementation problem
Given: An economic system
comprising of self-interested, rational agents, which hold some secret information
A system-wide goal Question:
Does there exist a mechanism that can enforce (through suitable economic incentives) the selfish agents to behave in such a way that desired goal is implemented?
The system-wide goal Goal: to implement desired social choices
in a strategic setting Social choice: aggregation of preferences
of participants towards a joint decision Why strategic setting?
participants act rationally and selfishly Usually preferences of participants are private
and can be used to manipulate the system
Designing a Mechanism
Informally, designing a mechanism means to define a game in which a desired outcome must be reached
However, games induced by mechanisms are different from games in standard form: Players hold independent private values The payoff matrix is a function of these types each player doesn’t really know about the
other players’ payoffs, but only about its one!
Games with incomplete information
MD and network protocols Large networks (e.g. Internet) are built
and controlled by diverse and competitive entities
Entities own different components of the network and hold private information
Entities are selfish and have different preferences
MD is a useful tool to design protocols working in such an environment
An example: auctions
t1=10
t2=12
t3=7
r1=11
r2=10
r3=7
the winner should be the
guy with highest value
the mechanism decidesthe winner and the
corresponding paymentti: type of player ivalue player i is willing to pay
if player i wins and has to pay phis utility is ui=ti-p
Mechanism degree of freedom
The mechanism has to decide: The allocation of the item (social
choice) The payment by the winner
…in a way that cannot be manipulated
A simple mechanism: no payment
t1=10
t2=12
t3=7
r1=+
r2=+
r3=+
…it doesn’t work…
?!?
The highest bid winsand the price of the item
is 0
Another simple mechanism: pay your bid
t1=10
t2=12
t3=7
r1=9
r2=8
r3=6
Is it the right choice?
The highest bid winsand the winner will
pay his bid
The winner is player 1 and he’ll
pay 9
Player i may bid ri< ti (in this way he is guaranteed not to incur a negative utility) …and so the winner could be the wrong one…
…it doesn’t work…
An elegant solution: Vickrey’s second price auction
t1=10
t2=12
t3=7
r1=10
r2=12
r3=7
every player has convenience to declare the truth!
(we’ll see)
I know they are not lying
The highest bid winsand the winner will
pay the secondhighest bid
The winner is player 2 and he’ll pay 10
TheoremIn the Vickrey auction, for every player i, ri=ti is adominant strategy
prooffix i, ti, r-i, and look at strategies for player i
T= maxji {rj}
Case: ti ≥ T
ri=ti gives utility ui= ti-T ≥ 0
declaring ri ≥ T cannot do better declaring ri < T leads to ui=0
Case: ti < T
ri=ti gives utility ui= 0
declaring ri ≤ T cannot do better declaring ri > T yields ui= ti-T < 0
Vickrey auction (minimization version)
t1=10
t2=12
t3=7
r1=10
r2=12
r3=7
I want to allocate the job
to the true cheapest machine
Once again, the second price auction works:
the cheapest bid winsand the winner will
get the secondcheapest bid
The winner is machine 3 and it will receive 10
job to be
allocated to
machines
job to be
allocated to
machines
ti: cost incurred by i if he does the job
if machine i is selected and receives a payment of p its utility is p-ti
Mechanism Design Problem: ingredients
N agents; each agent has some private information tiTi (actually, the only private info) called type
A set of feasible outcomes X For each vector of types t=(t1, t2, …, tN), and for each
feasible outcome xX, a social-choice function f(t,x) measures the quality of x as a function of t (the problem is that types are unknown!). This is the function that the mechanism aims to optimize (either minimize or maximize)
Each agent has a strategy space Si and performs a strategic action; we restrict ourself to direct revelation mechanisms, in which the action is reporting a value ri from the type space (with possibly ri ti), i.e., Si = Ti
Plugging-in into the min-version of the Vickrey Auction
Assume that the system-wide goal is to allocate a job by a sealed-bid auction:
Agent’s type ti: the cost incurred for doing the job; Type space Ti= (0, +]: the agent’s cost is any positive amount
of money; Set of feasible outcomes: is the set of agents (bidders); Social-choice function: is simply the type associated with a
selected bidder:f(t,x)={ti s.t. x=ti},
and the objective is to minimize f(), i.e., allocate to the bidder with lowest true cost.
Reported type ri: Amount of money the agent i bids to the system for doing the job (not known to other agents)
Mechanism Design Problem: ingredients (2)
For each feasible outcome xX, each agent makes a valuation vi(ti,x) (in terms of some common currency), expressing its preference about that output
Vickrey Auction: If agent i wins the auction then its valuation is equal to its actual cost=ti for doing the job, otherwise it is 0
For each feasible outcome xX, each agent receives a payment pi(x) in terms of the common currency; payments are used by the system to incentive agents to be collaborative. Then, the utility of outcome x will be:
ui(ti,x) = pi(x) - vi(ti,x) Vickrey Auction: For example, if agent’s cost for the job is 80,
and it gets the deal for 100 (i.e., it is paid 100), then its utility is 20
Mechanism Design Problem: the goal
Given all the above ingredients, provide a mechanism M=<g(r), p(x)>, where:
g(r) is an algorithm which computes an outcome x=x(r)X as a function of the reported types r
p(x) is a payment scheme specifying a payment w.r.t. an output x
which optimally implements the social-choice function in equilibrium (according to a given solution concept, e.g., dominant strategy equilibrium, Nash equilibrium, etc.), i.e., such that there exists a reported type vector r* for which f(t,x(r*)) is optimal (either minimum or maximum), and players’ utilities are in equilibrium.
Mechanism Design: a picture
Agent 1
Agent N
Mechanism
p1
pN
tN
t
1r
1
r
N
Private “types” Reported types
Payments
Output which should implement the social choice function
Each agent reports strategically to maximize its well-being…
…in response to a payment which is a function of the output!
Mechanism Design: Economics Issues
QUESTION: How to design a mechanism? Or, in other words:
1. How to design g(r), and2. How to define the payment functions
in such a way that the underlying social-choice function is implemented? Under which conditions can this be done?
Implementation with dominant strategies
Def.: A mechanism is an implementation with dominant strategies if there exists a reported type vector r*=(r1
*, r2*, …, rN
*) such that f() is implemented in dominant strategy equilibrium, i.e., for each agent i and for each reported type vector r =(r1, r2, …, rN), it holds:
ui(ti,x(r-i,ri*)) ≥ ui(ti,x(r))
where x(r-i,ri*)=x(r1, …, ri-1, ri
*, ri+1,…, rN).
Strategy-Proof Mechanisms
If truth telling is the dominant strategy in a mechanism then it is called Strategy-Proof or truthful r*=t.Agents report their true types instead of
strategically manipulating itThe algorithm of the mechanism runs on the
true input
Utilitarian Problems: A problem is utilitarian if its social-choice function is such that f(t,x) = i vi(ti,x)
notice: the auction problem is utilitarian, and so they are all problems where valuation functions are
separately-additive, as many network optimization problems…
Good news: for utilitarian problems there is a class
of truthful mechanisms
Vickrey-Clarke-Groves (VCG) Mechanisms
A VCG-mechanism is (the only) strategy-proof mechanism for utilitarian problems: Algorithm g(r) computes (for minimization
problems):
x = arg minyX i vi(ri,y) Payment function:
pi (x) = hi(r-i) - j≠i vj(rj,x)
where hi(r-i) is an arbitrary function of the reported types of players other than player i.
What about non-utilitarian problems? Strategy-proof mechanisms are known only when the type is a single parameter.
TheoremVCG-mechanisms are truthful for utilitarian problems
prooffix i, r-i, ti and consider a strategy riti
x= g(r-i,ti) x’=g(r-i,ri) t’=(r-i,ti)
ui(ti,x) =ui(ti,x’) =
[hi(r-i) - jivj(rj,x)] - vi(ti,x)
[hi(r-i) - jivj(rj,x’)] - vi(ti,x’)
= hi(r-
i)= hi(r-i)
- jvj(t’j,x)
- jvj(t’j,x’)
but g( ) minimizes the sum of valuations x is an optimal solution w.r.t. t’=(r-i,ti)
jvj(t’j,x) jvj(t’j,x’) ui(ti,x) ui(ti,x’).
How to define hi(r-i)?
Remark: not all functions make sense. For instance, what happens if we set hi(r-i)=0 in the Vickrey auction (min-version)?Answer: It happens that agents’ utilities become negative! This is undesirable in reality, since with such perspective agents would not partecipate to the auction!
Clarke payments
This is a special VCG-mechanism in which
hi(r-i)=j≠i vj(rj,x(r-i))
pi =j≠i vj(rj,x(r-i)) -j≠i vj(rj,x) With Clarke payments, agents’ utility are
always non-negative agents are interested in playing the
game
solution minimizing the sumof valuations when i doesn’t play
Clarke payments for the Vickrey auction (minimization version)
The VCG-mechanism is: x=arg minyX i vi(ri,y)
allocate to the bidder with lowest reported cost (i.e., the true one, since it is strategy-proof)
pi = j≠i vj(rj,x(r-i)) -j≠i vj(rj,x) …pay the winner the second lowest offer,
and pay 0 the losersRemark: the difference between the second lowest offer and the lowest offer is unbounded (frugality issue)
VCG-Mechanisms: Advantages
For System Designer: The goal, i.e., the optimization of the
social-choice function, is achieved with certainty.
For Agents: Agents have truth telling as the
dominant strategy, so they need not require any computational systems to deliberate about other agents strategies
VCG-Mechanisms: Disadvantages
For System Designer: The payments may be sub-optimal (frugality) System has to calculate N+1 functions
Once with all agents (for g(r)) and once for every agent (for the associated payment)
If the problem is hard to solve then the computational cost may be very heavy
For Agents: Agents may not like to tell the truth to the
system designer as it can be used in other ways.
Mechanism Design: Algorithmic Issues
QUESTION: What is the time complexity of the mechanism? Or, in other words:
What is the time complexity of g(r)? What is the time complexity to calculate the N
payment functions? What does it happen if it is NP-hard to
implement the underlying social-choice function?
Question: What is the time complexity of the Vickrey auction?Answer: Θ(N), where N is the number of players. Indeed, it suffices to check all the offers, by keeping track of the lowest and second lowest one.
Algorithmic mechanism design for graph problems
Simplifying the Internet model, we assume that each agent owns a single edge of a graph G=(V,E), and establishes the cost for using it
The agent’s type is the true weight of the edge
Classic optimization problems on G become mechanism design optimization problems!
Many basic network design problems have been faced: shortest path (SP), single-source shortest paths tree (SPT), minimum spanning tree (MST), and many others