CSC2556 Lecture 7 Fair Division 2: Indivisible Goods Leximin Allocation CSC2556 - Nisarg Shah 1
Cake-Cutting (contd) Indivisible Goods CSC2556 - Nisarg Shah 2
Pareto Optimality (PO) β’ Definition β’ We say that an allocation π΅ = (π΅ 1 , β¦ , π΅ π ) is PO if there is no alternative allocation πΆ = (πΆ 1 , β¦ , πΆ π ) such that 1. Every agent is at least as happy: π π πΆ π β₯ π π (π΅ π ) , βπ β π 2. Some agent is strictly happier: π π πΆ π > π π (π΅ π ) , βπ β π β’ I.e., an allocation is PO if there is no βbetterβ allocation. β’ Q: Is it PO to give the entire cake to player 1? β’ A: Not necessarily. But yes if player 1 values βevery part of the cake positivelyβ. CSC2556 - Nisarg Shah 3
PO + EF β’ Theorem [Weller β85]: β’ There always exists an allocation of the cake that is both envy-free and Pareto optimal. β’ One way to achieve PO+EF: β’ Nash-optimal allocation: argmax π΅ Ο πβπ π π π΅ π β’ Obviously, this is PO. The fact that it is EF is non-trivial. β’ This is named after John Nash. o Nash social welfare = product of utilities o Different from utilitarian social welfare = sum of utilities CSC2556 - Nisarg Shah 4
Nash-Optimal Allocation 2 3 ΰ΅ 0 1 β’ Example: 2 3 β’ Green player has value 1 distributed over 0, Ξ€ β’ Blue player has value 1 distributed over [0,1] β’ Without loss of generality (why?) suppose: o Green player gets π¦ fraction of [0, Ξ€ 2 3 ] 2 3 ] AND all of [ Ξ€ 2 3 , 1] . o Blue player gets the remaining 1 β π¦ fraction of [0, Ξ€ β’ Greenβs utility = π¦ , blueβs utility = 1 β x β 2 3 + 1 3 = 3β2π¦ 3 β’ Maximize: π¦ β 3β2π¦ 3 4 ( Ξ€ 3 4 fraction of Ξ€ 2 3 is Ξ€ 1 2 ). β π¦ = Ξ€ 3 1 2 Green has utility 3 ΰ΅ 4 0 1 Allocation Blue has utility 1 2 CSC2556 - Nisarg Shah 5
Problem with Nash Solution β’ Difficult to compute in general β’ I believe it should require an unbounded number of queries in the Robertson- Webb model. But I canβt find such a result in the literature. β’ Theorem [Aziz & Ye β14]: β’ For piecewise constant valuations, the Nash-optimal solution can be computed in polynomial time. The density function of a piecewise constant valuation looks like this 0 1 CSC2556 - Nisarg Shah 6
Interlude: Homogeneous Divisible Goods β’ Suppose there are π homogeneous divisible goods β’ Each good can be divided fractionally between the agents β’ Let π¦ π,π = fraction of good π that agent π gets β’ Homogeneous = agent doesnβt care which βpartβ o E.g., CPU or RAM β’ Special case of cake-cutting β’ Line up the goods on [0,1] β piecewise uniform valuations CSC2556 - Nisarg Shah 7
Interlude: Homogeneous Divisible Goods β’ Nash-optimal solution: Maximize Ο π log π π π π = Ξ£ π π¦ π,π β π€ π,π βπ Ξ£ π π¦ π,π = 1 βπ π¦ π,π β [0,1] βπ, π β’ Gale-Eisenberg Convex Program β’ Polynomial time solvable CSC2556 - Nisarg Shah 8
Indivisible Goods β’ Goods which cannot be shared among players β’ E.g., house, painting, car, jewelry, β¦ β’ Problem: Envy-free allocations may not exist! CSC2556 - Nisarg Shah 9
Indivisible Goods: Setting 8 7 20 5 9 11 12 8 9 10 18 3 Given such a matrix of numbers, assign each good to a player. We assume additive values. So, e.g., π , = 8 + 7 = 15 CSC2556 - Nisarg Shah 10
Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 11
Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 12
Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 13
Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 14
Indivisible Goods β’ Envy-freeness up to one good (EF1): βπ, π β π, βπ β π΅ π βΆ π π π΅ π β₯ π π π΅ π \{π} β’ Technically, we need either this or π΅ π = β . β’ βIf π envies π , there must be some good in π βs bundle such that removing it would make π envy-free of π .β β’ Does there always exist an EF1 allocation? CSC2556 - Nisarg Shah 15
EF1 β’ Yes! We can use Round Robin. β’ Agents take turns in cyclic order: 1,2, β¦ , π, 1,2, β¦ , π, β¦ β’ In her turn, an agent picks the good she likes the most among the goods still not picked by anyone. β’ Observation: This always yields an EF1 allocation. β’ Informal proof on the board. β’ Sadly, on some instances, this returns an allocation that is not Pareto optimal. CSC2556 - Nisarg Shah 16
EF1+PO? β’ Nash welfare to rescue! β’ Theorem [Caragiannis et al. β16]: β’ The allocation argmax π΅ Ο πβπ π π π΅ π is EF1 + PO. β’ Note: This maximization is over only βintegralβ allocations that assign each good to some player in whole. β’ Note: Subtle tie-breaking if all allocations have zero Nash welfare. o Step 1: Choose a subset of players π β π with largest |π| such that it is possible to give a positive utility to every player in π simultaneously. o Step 2: Choose argmax π΅ Ο πβπ π π π΅ π CSC2556 - Nisarg Shah 17
Integral Nash Allocation 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 19
20 * 8 * (9+10) = 3040 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 20
(8+7) * 8 * 18 = 2160 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 21
8 * (12+8) * 10 = 1600 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 22
20 * (11+8) * 9 = 3420 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 23
Computation β’ For indivisible goods, Nash-optimal solution is strongly NP-hard to compute β’ That is, remains NP-hard even if all values in the matrix are bounded β’ Open Question: If our goal is EF1+PO, is there a different polynomial time algorithm? β’ Not sure. But a recent paper gives a pseudo-polynomial time algorithm for EF1+PO o Time is polynomial in π , π , and max π,π π π . π CSC2556 - Nisarg Shah 24
Other Fairness Notions β’ Maximin Share Guarantee (MMS): β’ Generalization of βcut and chooseβ for π players β’ MMS value of player π = o The highest value player π can getβ¦ o If she divides the goods into π bundlesβ¦ o But receives the worst bundle for her (βworst case guaranteeβ) β’ Let π¬ π π denote the family of partitions of the set of goods π into π bundles. πππ π = max πβ 1,β¦,π π min π (πΆ π ) . πΆ 1 ,β¦,πΆ π βπ¬ π π β’ An allocation is π½ -MMS if every player π receives value at least π½ β πππ π . CSC2556 - Nisarg Shah 25
Other Fairness Notions β’ Maximin Share Guarantee (MMS) β’ [Procaccia , Wang β14]: There is an example in which no MMS allocation exists. β’ [Procaccia , Wang β14]: 2 3 - MMS allocation always exists. A Ξ€ β’ [Ghodsi et al. β17]: 3 4 - MMS allocation always exists. A Ξ€ β’ [Caragiannis et al. β16]: 2 1+ 4πβ3 β MMS, and this is The Nash-optimal solution is the best possible guarantee. CSC2556 - Nisarg Shah 26
Stronger Fairness β’ Open Question: Does there always exist an EFx allocation? β’ EF1: βπ, π β π, βπ β π΅ π βΆ π π π΅ π β₯ π π π΅ π \{π} β’ Intuitively, π doesnβt envy π if she gets to remove her most valued item from π βs bundle. β’ EFx: βπ, π β π, βπ β π΅ π βΆ π π π΅ π β₯ π π π΅ π \{π} β’ Note: Need to quantify over π such that π π > 0 . π β’ Intuitively, π doesnβt envy π even if she removes her least positively valued item from π βs bundle. CSC2556 - Nisarg Shah 27
Stronger Fairness β’ The difference between EF1 and EFx: β’ Suppose there are two players and three goods with values as follows. A B C P1 5 1 10 P2 0 1 10 β’ If you give {A} β P1 and {B,C} β P2, itβs EF1 but not EFx. o EF1 because if P1 removes C from P2βs bundle, all is fine. o Not EFx because removing B doesnβt eliminate envy. β’ Instead, {A,B} β P1 and {C} β P2 would be EFx. CSC2556 - Nisarg Shah 28
Allocation of Bads β’ Negative utilities (costs instead of values) β’ Let π π,π be the cost of player π for bad π . o π· π π = Ο πβπ π π,π β’ EF: βπ, π π· π π΅ π β€ π· π π΅ π β’ PO: There should be no alternative allocation in which no player has more cost, and some player has less cost. β’ Divisible bads β’ EF + PO allocation always exists, like for divisible goods. o One way to achieve is through βCompetitive Equilibriaβ (CE). o For divisible goods, Nash-optimal allocation is the unique CE. o For bads, exponentially many CE. CSC2556 - Nisarg Shah 29
Allocation of Bads β’ Indivisible bads β’ EF1: βπ, π βπ β π΅ π π π π΅ π \ π β€ π π π΅ π β’ EFx: βπ, π βπ β π΅ π π π π΅ π \ π β€ π π π΅ π o Note: Again, we need to restrict to π such that π π,π > 0 β’ Open Question 1: o Does an EF1 + PO allocation always exist? β’ Open Question 2: o Does an EFx allocation always exist? β’ More open questions related to relaxations of proportionality CSC2556 - Nisarg Shah 30
Recommend
More recommend