1 / 107

Cooperative Games, Mechanism Design, and Auctions

Cooperative Games, Mechanism Design, and Auctions. Onn Shehory March 9-13 2009 Politecnico di Milano. The Glove Game. Players set: {1,2,3} 1,2 have right-had gloves 3 has left-had gloves Players may stay alone and profit 0 Or join together: {1,3}, {2,3}, {1,2,3} to gain 1

fauve
Download Presentation

Cooperative Games, Mechanism Design, and Auctions

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Cooperative Games, Mechanism Design, and Auctions Onn Shehory March 9-13 2009 Politecnico di Milano

  2. The Glove Game • Players set: {1,2,3} • 1,2 have right-had gloves • 3 has left-had gloves • Players may stay alone and profit 0 • Or join together: {1,3}, {2,3}, {1,2,3} to gain 1 • What mechanism should they follow? • How do they behave given such mechanism? • How is profit divided?

  3. The English Auction • An item is placed for sale • Players free to bid • New bid must be higher than current • Winner: highest bid

  4. Outline • Protocols and strategies • Attitudes and rationality • Stability and equilibrium • Nash revisited • Cooperative games • Solution concepts • Coalition Formation • RFP coalitions • Mechanism design • Auctions • Auctions – field results

  5. What is a Mechanism/Protocol? • A protocol (aka mechanism): • provides a set of rules and behaviors to be followed by its participants • following the rules of a protocol is to a player’s discretion, though deviation may leave her “out of the game” • examples: auctions, negotiation, voting • Desired properties: • maximize payoffs • not manipulable/enforceable • simple to implement and execute

  6. What are Strategies? • A strategy: • is one of the possible actions a player can select given the protocol • is not dictated (or provided) by the protocol • is usually the result of the player’s reasoning and decisions, based on local algorithms and information • examples: • in an auction – bid as low as possible • in elections – vote for your faction • A good strategy: • should maximize the player’s payoff given the protocol and the behavior of other players • should be difficult or impossible to manipulate • should be computationally feasible • may depend on the strategies of other players

  7. Players Attitudes • Self-interest: a self-interested player is attempting to maximize its own personal payoff • Benevolence/altruism: a benevolent player is attempting to increase others’ payoffs and the cumulative payoff of the society

  8. Risk Attitudes • Risk prone: a player that has a preference for risk • Risk averse: a player that has a preference for avoiding risk • Risk neutral: a player that has no risk preference • Players do not need to be strictly prone, neutral or averse – they may mix these • Human players tend to have alternating and context dependent risk attitudes

  9. Example Risk Attitudes • You are offered two options: • Get 1000 Euro, cash • Get a lottery certificate, with a prize value of 10,000 Euro, and a 10% chance of winning • What should you choose? • Select 2, for a chance for getting 10,000? • Select 1 to avoid risk? • What should you choose if chance is 20%? • Select 2, to maximize expected utility (2000)?

  10. Rationality • A rational behavior is such that prefers a greater payoff over a smaller one • A rational player should always behave rationally. That is, from among several options available, he should select the one that results in maximum payoff • The problem: • in may cases the number of options is overwhelming • there may be no algorithm for finding the best

  11. Bounded Rationality • To overcome problems of rationality, bounded rationality: • limits the time/computation for option consideration • prunes the search space • imposes restrictions on the types of options • Results in fewer possibilities, hence • computationally feasible • may be too restrictive, far from optimal • strategically inferior to rational

  12. “Good Enough” Behavior • Make the bounded rationality rational: • modify linear payoff functions to incorporate computational costs • put a cap on payoff • add a small-amounts’ indifference • The payoff of an option is good enough if • too much additional computation to find other good options, or • other options do not provide a significant payoff increase, or • the player is indifferent w.r.t. the increase

  13. Stability and Equilibria

  14. Protocol Evaluation • Payoff maximization: can refer to individual payoffs, group payoffs, or social welfare - the sum of individual payoffs • Pareto-optimality: a payoff vector p(x1,x2,…,xn) is Pareto-optimal if there is no other feasible payoff vector p' such that at least one payoff is better in p' and no payoff is worse in p • Stability: a protocol is stable if once the players arrived at a solution they do not deviate from it

  15. Stability and Equilibria • There are multiple stability concepts. In game theory, the notion of equilibrium is used: • dominant strategies: the agents have some strategies that, regardless of what others do, maximize payoff • Nash equilibrium: the agents have strategies that, as long as other stick to theirs, maximize payoff • Mixed Nash: the agents each have a set of strategies from among which they select one with some probability • Bayes-Nash: adds types (e.g. history) to the previous one

  16. The Prisoner’s Dilemma

  17. No Pure Nash Equilibrium

  18. No Pure Nash Equilibrium

  19. Mixed Nash • Player 1 will cooperate with probability pc and defect with probability pd • Player 2 will cooperate with probability qc and defect with probability qd • Expected utility of an agent is the utility from a strategy times the probability of this strategy being selected • When there are multiple possibilities, the expected utility is a sum over these possibilities

  20. Computing the Probabilities • The expected utility of an player x when the other player y follows strategy s is denoted by Ux(s) • In the case of equilibrium (mixed Nash), the expected utility of x should be the same for all of the possible strategies of y • In our case we have players 1,2 and strategies c,d • We require that Ux(c) = Ux(d), which means that for each of the two players, the expected utility from the other cooperating should be equal to the expected utility from the other defecting

  21. Computation Details • For player 1 we have: • U1(c)=6 pc+ 0 pd, U1(d)=0 pc+ 5 pd • For player 2 we have: • U2(c)=2 qc+ 3 qd, U2(d)=1 qc+ 0 qd • The requirement that Ux(c) = Ux(d) results in: • qc= 0.642, qd = 0.358 • pc= 0.317, pd = 0.683

  22. Tragedy of Common Goods • Information on the web is (mostly) free • Web agents that seek up to date information may query web site as frequently as desired • If all agents will do so, the network will be overly congested, and some servers will crash • So, is it undesirable to behave this way? • If all (or most) of the agents prevent congestion, it is in the best interest of each individual agent to increase network use ...

  23. Cooperative Games

  24. Cooperative Games • A cooperative game (aka coalitional game): • Cooperation within groups is enforceable • Groups compete, and not individuals • Each group (=coalition) has a value v • Characteristic function: v : 2N, from coalitions to payments • Players decide which coalitions to form (to maximize payoff)

  25. Coalitional Games • Coalitional game (N,v) • A set of players N • A coalitionS is a group of players, subset of N, which cooperate • Value (or utility) of a coalition v • v(S)is a real, represents the gain of coalition S in the game (N,v) • v(N)is the value of forming the grand coalition, coalition of all players • Player payoff xi • The portion of v(S) received by a player i in coalition S • Characteristic function form implies: • vdepends only on the internal structure of the coalition • Transferable utility • The value of a coalition can be distributed arbitrarily among its players

  26. Coalitional Games: Example • Example: Majority Vote • Prime minister is elected by majority vote • A coalition consisting of a majority of players has a worth of 1 since it is a decision maker • Value of a coalition does not depend on the external strategies of the users => characteristic function form

  27. Super-additive Games • Super-additive game • (N,v)is super-additive if • Here, cooperation is always beneficial • Unification of two coalitions increases overall payoff • Monotonicity: larger coalitions gain more • Not all games are super-additive!

  28. Coalitional Stability • Stability of a coalition • Depends on how the value visdistributed among the players • How to divide v? • Improper payoff division => players may leave the coalition, unhappy with their share • Multiple solution concepts address this point

  29. Coordination Game • For 1, A>C, D>B • For 2, a>c, d>b • Red circles are pure Nash • For driving side, all benefit if all adopt the same side, but two equilibria points exist

  30. Solution Concepts

  31. Solution Concepts • Assumption: the grand coalition will form • Even when the solution includes multiple coalitions • Solve for sub-games • Each players gets a payoff xi • Major issue: payoff distribution • A solution concept is a payoff allocation vector x N • Efficiency: ∑ xi = v(N) • Individual rationality: xi ≥ v(i) • Group rationality: v(S)≤ ∑ xi S • Imputation: a payoff allocation vector which is efficient and individually rational (and group rational for N) • Many solution concepts are imputations • Players prefer coalitions based on their respective payoffs

  32. Example • Socks selling • Sold in pairs for 3euro a pair • 2 sellers, each holds 5 socks • Each can get 6euro, but together they get 15euro. • Imputations: (6, 9), (7, 8), (7.5, 7.5)

  33. Some Properties • Null player: when added to a coalition, contributes nothing to its value • Existence (of a solution concept): determines whether the solution concept exists for every game • Examples: Kernel exists, Core does not • Symmetry: symmetric players receive equal payoffs • Uniqueness: the solution concept is unique

  34. The Shapley Value (1953) • Unique, efficient, symmetric, allocates zero to null players • Individually rational for super-additive games • The payoff allocated to player i is φi(v) = 1/n! ΣSN\i (s!-(n-s-1)!) (v(S∪i)-v(S)) • s = |S| • This is considered a fair allocation

  35. The Glove Game Example • N={1,2,3} • 1,2 have right-had gloves • 3 has left-had gloves • v(S) = 1 for {1,3}, {2,3}, {1,2,3}, otherwise 0 • φ1(v) = 1/6 • φ2(v) = 1/6 • φ3(v) = 4/6

  36. The Stable Setvon Neumann & Morgenstern (1944) • Given a game v and imputations x,y • x dominates y if • there is a nonempty coalition S, such that • the members of S prefer payoff from x over payoff from y • v(S) ≥ ∑ xi S • S players can threaten to quit the grand coalition if x not implemented

  37. Stable Set Example • x=(1,2,3), y=(3,2,1) • For players 1,2, y is better than x • Does y dominate x? • Depends on v({1,2}) • If v({1,2}) ≥ 5, y dominates x via {1,2}

  38. Stable Set Example 2 • x=(50,50,0), y=(0,60,40), z=(15,0,85) • v({1,2}) = v({1,3}) = v({2,3}) =100 • y dominates x via {2,3} • z dominates y via {1,3} • x dominates z via {1,2}

  39. Some Properties • Existence: the stable set may or may not exist • Uniqueness: when exists, it is usually not unique • Internal stability: no imputation within the stable set dominate one another • External stability: all payoff vectors outside the stable set are dominated by at least on member of the set • Interpretation: the stable set represents conflicts, but excludes inferior behaviors

  40. The Core • The coreis a set of imputations(x1, . . .,xN)satisfying two conditions • No coalition has a value greater than the sum of its members’ payoffs (coalition rationality) • The core can be empty • A non-empty core in a super-additive game => stable grand coalition • No coalition has an incentive to leave (and receive a greater payoff)

  41. Properties of the Core • May be empty (rather common) • Not unique • Subset of the stable set • Has several variants • E.g., epsilon-core • Least-core

  42. Shoes Example • A pair: a left and a right shoe. Sold for €50 • 21 players: 10 have 1 left shoe, 11 have 1 right shoe • The core: a single imputation, gives 10 to left shoe owners, 0 to right shoe owners • Any left-right pair can form a coalition and sell for €50 • Any such pair getting less than that will block the imputation • For imputation in the core, any of these pairs gets exactly 50 (we can only sell 10 pairs, totaling to 500) • One right-shoe owner gets 0 payment • Examine the pairs: if any left-shoe owner gets less than 50, say 40, it can join this player, sell their shoes, give her 5, and keep 45 to herself. Both are better off • But such a left-shoe owner cannot exist: all left shoe owners get already 50. • This holds for any unequal partition • The core is very sensitive to oversupply.

  43. The KernelDavis & Maschler (1965) • We do not assume imputations • That’s it – no group rationality • Efficiency still required • Result: we are interested no only in payoff distribution, but in the coalitions that form • Implicitly assumes bargaining (but in practice computes its result)

  44. Surplus • Maximum surplus • Given an efficient payoff vector x and players i,j, the maximum surplus of i over j is: sij(x)= max(v(S)- ∑kSxk : iS, jS) • the maximal amount i can gain without the cooperation of j by withdrawing from the grand coalition N (where x applies), other players in i's withdrawing coalition are satisfied with their x payoffs • The maximum surplus measures a player's bargaining power over another

  45. The Kernel Defined • The kernel is the set of payoff vectors x that satisfy, for all pairs i,j: (sij(x)-sji(x))·(xj - v(j)) ≤ 0 and (sji(x)-sij(x))·(xi - v(i)) ≤ 0 • If sij(x) > sji(x) then ioutweighsj – it has more bargaining power • But ifxj = v(j), j can obtain xj on his own, thus i’s threat is invalidated • For vectors in the kernel, no player can outweigh another – no valid threat  stability

  46. Kernel Properties • Equilibrium of agents‘ surpluses: In each coalition no player can outweigh another, thus getting a better payoff (surplus) in an alternative coalition excluding the opponent „I can get more without you, than you can without me.“ • Exists, Pareto-optimal, not unique • A subset of the bargaining set • Exponentially hard to compute • Computational solution: Stearns (1968) • May converge very slowly • A single point out of many • Polynomial variants (Shehory/Kraus 1996; Klusch/Shehory 1996)

  47. Example • Three players • v(i)=0, v(12)=2, v(13)=3, v(23)=4, v(123)=8 • Kernel: 2, 2.5, 3.5 • Shapley: 13/6, 16/6, 19/6 • v(i)=0, v(12)=2, v(13)=3, v(23)=4, v(123)=5 • Kernel: 2/3, 5/3, 8/3 • Shapley: 7/6, 10/6, 13/6

  48. More Concepts • The Bargaining set • The Nucleolus • Both based on excess: v(S)- ∑kSxk • Possible gain of players when quitting a coalition

  49. Coalition Formation

  50. Coalitions and Complexity • Given N players, there are 2N-1 different possible coalitions • If there are k tasks, may need to multiply by exp(k) • The number of configurations is O(N(N/2)) • Hence, exhaustive search is infeasible • Additionally, players may have conflicting preferences over the possible configurations • Nevertheless, coalitions are beneficial

More Related