fair division 2
play

Fair Division 2: Indivisible Goods Leximin Allocation CSC2556 - - PowerPoint PPT Presentation

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) Definition We say that an allocation =


  1. CSC2556 Lecture 7 Fair Division 2: Indivisible Goods Leximin Allocation CSC2556 - Nisarg Shah 1

  2. Cake-Cutting (contd) Indivisible Goods CSC2556 - Nisarg Shah 2

  3. Pareto Optimality (PO) β€’ Definition ➒ We say that an allocation 𝐡 = (𝐡 1 , … , 𝐡 π‘œ ) is PO if there is no alternative allocation 𝐢 = (𝐢 1 , … , 𝐢 π‘œ ) such that 1. Every agent is at least as happy: π‘Š 𝑗 𝐢 𝑗 β‰₯ π‘Š 𝑗 (𝐡 𝑗 ) , βˆ€π‘— ∈ 𝑂 2. Some agent is strictly happier: π‘Š 𝑗 𝐢 𝑗 > π‘Š 𝑗 (𝐡 𝑗 ) , βˆƒπ‘— ∈ 𝑂 ➒ I.e., an allocation is PO if there is no β€œbetter” allocation. β€’ Q: Is it PO to give the entire cake to player 1? β€’ A: Not necessarily. But yes if player 1 values β€œevery part of the cake positively”. CSC2556 - Nisarg Shah 3

  4. PO + EF β€’ Theorem [Weller β€˜85]: ➒ There always exists an allocation of the cake that is both envy-free and Pareto optimal. β€’ One way to achieve PO+EF: ➒ Nash-optimal allocation: argmax 𝐡 Ο‚ π‘—βˆˆπ‘‚ π‘Š 𝑗 𝐡 𝑗 ➒ Obviously, this is PO. The fact that it is EF is non-trivial. ➒ 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

  5. Nash-Optimal Allocation 2 3 ΰ΅— 0 1 β€’ Example: 2 3 ➒ Green player has value 1 distributed over 0, Ξ€ ➒ Blue player has value 1 distributed over [0,1] ➒ Without loss of generality (why?) suppose: o Green player gets 𝑦 fraction of [0, Ξ€ 2 3 ] 2 3 ] AND all of [ Ξ€ 2 3 , 1] . o Blue player gets the remaining 1 βˆ’ 𝑦 fraction of [0, Ξ€ ➒ Green’s utility = 𝑦 , blue’s utility = 1 βˆ’ x β‹… 2 3 + 1 3 = 3βˆ’2𝑦 3 ➒ Maximize: 𝑦 β‹… 3βˆ’2𝑦 3 4 ( Ξ€ 3 4 fraction of Ξ€ 2 3 is Ξ€ 1 2 ). β‡’ 𝑦 = Ξ€ 3 1 2 Green has utility 3 ΰ΅— 4 0 1 Allocation Blue has utility 1 2 CSC2556 - Nisarg Shah 5

  6. Problem with Nash Solution β€’ Difficult to compute in general ➒ I believe it should require an unbounded number of queries in the Robertson- Webb model. But I can’t find such a result in the literature. β€’ Theorem [Aziz & Ye β€˜14]: ➒ 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

  7. Interlude: Homogeneous Divisible Goods β€’ Suppose there are 𝑛 homogeneous divisible goods ➒ Each good can be divided fractionally between the agents β€’ Let 𝑦 𝑗,𝑕 = fraction of good 𝑕 that agent 𝑗 gets ➒ Homogeneous = agent doesn’t care which β€œpart” o E.g., CPU or RAM β€’ Special case of cake-cutting ➒ Line up the goods on [0,1] β†’ piecewise uniform valuations CSC2556 - Nisarg Shah 7

  8. Interlude: Homogeneous Divisible Goods β€’ Nash-optimal solution: Maximize Οƒ 𝑗 log 𝑉 𝑗 𝑉 𝑗 = Ξ£ 𝑕 𝑦 𝑗,𝑕 βˆ— 𝑀 𝑗,𝑕 βˆ€π‘— Ξ£ 𝑗 𝑦 𝑗,𝑕 = 1 βˆ€π‘• 𝑦 𝑗,𝑕 ∈ [0,1] βˆ€π‘—, 𝑕 β€’ Gale-Eisenberg Convex Program ➒ Polynomial time solvable CSC2556 - Nisarg Shah 8

  9. Indivisible Goods β€’ Goods which cannot be shared among players ➒ E.g., house, painting, car, jewelry, … β€’ Problem: Envy-free allocations may not exist! CSC2556 - Nisarg Shah 9

  10. 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‘Š , = 8 + 7 = 15 CSC2556 - Nisarg Shah 10

  11. Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 11

  12. Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 12

  13. Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 13

  14. Indivisible Goods 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 14

  15. Indivisible Goods β€’ Envy-freeness up to one good (EF1): βˆ€π‘—, π‘˜ ∈ 𝑂, βˆƒπ‘• ∈ 𝐡 π‘˜ ∢ π‘Š 𝑗 𝐡 𝑗 β‰₯ π‘Š 𝑗 𝐡 π‘˜ \{𝑕} ➒ Technically, we need either this or 𝐡 π‘˜ = βˆ… . ➒ β€œIf 𝑗 envies π‘˜ , there must be some good in π‘˜ ’s bundle such that removing it would make 𝑗 envy-free of π‘˜ .” β€’ Does there always exist an EF1 allocation? CSC2556 - Nisarg Shah 15

  16. EF1 β€’ Yes! We can use Round Robin. ➒ Agents take turns in cyclic order: 1,2, … , π‘œ, 1,2, … , π‘œ, … ➒ 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. ➒ Informal proof on the board. β€’ Sadly, on some instances, this returns an allocation that is not Pareto optimal. CSC2556 - Nisarg Shah 16

  17. EF1+PO? β€’ Nash welfare to rescue! β€’ Theorem [Caragiannis et al. β€˜16]: ➒ The allocation argmax 𝐡 Ο‚ π‘—βˆˆπ‘‚ π‘Š 𝑗 𝐡 𝑗 is EF1 + PO. ➒ Note: This maximization is over only β€œintegral” allocations that assign each good to some player in whole. ➒ Note: Subtle tie-breaking if all allocations have zero Nash welfare. o Step 1: Choose a subset of players 𝑇 βŠ† 𝑂 with largest |𝑇| such that it is possible to give a positive utility to every player in 𝑇 simultaneously. o Step 2: Choose argmax 𝐡 Ο‚ π‘—βˆˆπ‘‡ π‘Š 𝑗 𝐡 𝑗 CSC2556 - Nisarg Shah 17

  18. Integral Nash Allocation 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 19

  19. 20 * 8 * (9+10) = 3040 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 20

  20. (8+7) * 8 * 18 = 2160 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 21

  21. 8 * (12+8) * 10 = 1600 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 22

  22. 20 * (11+8) * 9 = 3420 8 7 20 5 9 11 12 8 9 10 18 3 CSC2556 - Nisarg Shah 23

  23. Computation β€’ For indivisible goods, Nash-optimal solution is strongly NP-hard to compute ➒ 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? ➒ Not sure. But a recent paper gives a pseudo-polynomial time algorithm for EF1+PO o Time is polynomial in π‘œ , 𝑛 , and max 𝑗,𝑕 π‘Š 𝑕 . 𝑗 CSC2556 - Nisarg Shah 24

  24. Other Fairness Notions β€’ Maximin Share Guarantee (MMS): ➒ Generalization of β€œcut and choose” for π‘œ players ➒ MMS value of player 𝑗 = o The highest value player 𝑗 can get… o If she divides the goods into π‘œ bundles… o But receives the worst bundle for her (β€œworst case guarantee”) ➒ Let 𝒬 π‘œ 𝑁 denote the family of partitions of the set of goods 𝑁 into π‘œ bundles. 𝑁𝑁𝑇 𝑗 = max π‘™βˆˆ 1,…,π‘œ π‘Š min 𝑗 (𝐢 𝑙 ) . 𝐢 1 ,…,𝐢 π‘œ βˆˆπ’¬ π‘œ 𝑁 ➒ An allocation is 𝛽 -MMS if every player 𝑗 receives value at least 𝛽 βˆ— 𝑁𝑁𝑇 𝑗 . CSC2556 - Nisarg Shah 25

  25. Other Fairness Notions β€’ Maximin Share Guarantee (MMS) ➒ [Procaccia , Wang ’14]: There is an example in which no MMS allocation exists. ➒ [Procaccia , Wang ’14]: 2 3 - MMS allocation always exists. A Ξ€ ➒ [Ghodsi et al. β€˜17]: 3 4 - MMS allocation always exists. A Ξ€ ➒ [Caragiannis et al. ’16]: 2 1+ 4π‘œβˆ’3 βˆ’ MMS, and this is The Nash-optimal solution is the best possible guarantee. CSC2556 - Nisarg Shah 26

  26. Stronger Fairness β€’ Open Question: Does there always exist an EFx allocation? β€’ EF1: βˆ€π‘—, π‘˜ ∈ 𝑂, βˆƒπ‘• ∈ 𝐡 π‘˜ ∢ π‘Š 𝑗 𝐡 𝑗 β‰₯ π‘Š 𝑗 𝐡 π‘˜ \{𝑕} ➒ Intuitively, 𝑗 doesn’t envy π‘˜ if she gets to remove her most valued item from π‘˜ ’s bundle. β€’ EFx: βˆ€π‘—, π‘˜ ∈ 𝑂, βˆ€π‘• ∈ 𝐡 π‘˜ ∢ π‘Š 𝑗 𝐡 𝑗 β‰₯ π‘Š 𝑗 𝐡 π‘˜ \{𝑕} ➒ Note: Need to quantify over 𝑕 such that π‘Š 𝑕 > 0 . 𝑗 ➒ Intuitively, 𝑗 doesn’t envy π‘˜ even if she removes her least positively valued item from π‘˜ ’s bundle. CSC2556 - Nisarg Shah 27

  27. Stronger Fairness β€’ The difference between EF1 and EFx: ➒ Suppose there are two players and three goods with values as follows. A B C P1 5 1 10 P2 0 1 10 ➒ If you give {A} β†’ P1 and {B,C} β†’ P2, it’s EF1 but not EFx. o EF1 because if P1 removes C from P2’s bundle, all is fine. o Not EFx because removing B doesn’t eliminate envy. ➒ Instead, {A,B} β†’ P1 and {C} β†’ P2 would be EFx. CSC2556 - Nisarg Shah 28

  28. Allocation of Bads β€’ Negative utilities (costs instead of values) ➒ Let 𝑑 𝑗,𝑐 be the cost of player 𝑗 for bad 𝑐 . o 𝐷 𝑗 𝑇 = Οƒ π‘βˆˆπ‘‡ 𝑑 𝑗,𝑐 ➒ EF: βˆ€π‘—, π‘˜ 𝐷 𝑗 𝐡 𝑗 ≀ 𝐷 𝑗 𝐡 π‘˜ ➒ PO: There should be no alternative allocation in which no player has more cost, and some player has less cost. β€’ Divisible bads ➒ EF + PO allocation always exists, like for divisible goods. o One way to achieve is through β€œCompetitive Equilibria” (CE). o For divisible goods, Nash-optimal allocation is the unique CE. o For bads, exponentially many CE. CSC2556 - Nisarg Shah 29

  29. Allocation of Bads β€’ Indivisible bads ➒ EF1: βˆ€π‘—, π‘˜ βˆƒπ‘ ∈ 𝐡 𝑗 𝑑 𝑗 𝐡 𝑗 \ 𝑐 ≀ 𝑑 𝑗 𝐡 π‘˜ ➒ EFx: βˆ€π‘—, π‘˜ βˆ€π‘ ∈ 𝐡 𝑗 𝑑 𝑗 𝐡 𝑗 \ 𝑐 ≀ 𝑑 𝑗 𝐡 π‘˜ o Note: Again, we need to restrict to 𝑐 such that 𝑑 𝑗,𝑐 > 0 ➒ Open Question 1: o Does an EF1 + PO allocation always exist? ➒ Open Question 2: o Does an EFx allocation always exist? ➒ More open questions related to relaxations of proportionality CSC2556 - Nisarg Shah 30

Recommend


More recommend