closed formulae for revenue maximizing mechanisms in 2 d
play

Closed formulae for revenue-maximizing mechanisms in 2-D sequencing - PowerPoint PPT Presentation

Closed formulae for revenue-maximizing mechanisms in 2-D sequencing mechanism design Ruben Hoeksma rubenh@dii.uchile.cl Joint work with Marc Uetz (University of Twente) ADGO Workshop 2016 Revenue maximizing mechanism design Selling product


  1. Closed formulae for revenue-maximizing mechanisms in 2-D sequencing mechanism design Ruben Hoeksma – rubenh@dii.uchile.cl Joint work with Marc Uetz (University of Twente) ADGO Workshop 2016

  2. Revenue maximizing mechanism design Selling product (goods/services) under incomplete information. ◮ Combinatorial optimization problem ◮ Agents ‘own’ parameters ◮ May misrepresent ◮ Mechanism = set of rules: ◮ Input: strategies of the agents ◮ Output: feasible solution + payments Example Single item auction

  3. Myerson optimal single item auctions Selling a single item to a group of agents [Meyerson, 1981]. ◮ Agents: private information on valuation ◮ Priors on the private information ◮ Mechanism outcome: allocation + payments Optimal mechanism: ◮ Strategies: revealing information ◮ Truth telling w.l.o.g. ◮ ‘Nice’ properties Focus of this talk Properties of 1-D, 1.5-D and 2-D revenue optimal mechanisms for sequencing.

  4. Sequencing jobs on a single processor p j j S j ◮ Job: unit waiting cost, w j ; processing requirement, p j ◮ Jobs must be scheduled ◮ Payments, π j , reimburse jobs for waiting cost (= w j S j ) ◮ Minimize total payment All data known: ◮ � π j = � w j S j ◮ Priorities according to w j / p j (Smith’s Rule [Smith 1956])

  5. Mechanism design problem ◮ Type t j = ( w j , p j ) ∈ T j is private to agent j (owns job j ) ◮ Probability distribution ϕ j : T j → (0 , 1] public knowledge ◮ Agents may lie to maximize utility, u j = π j − w j S j ◮ Mechanism = schedule + payments ◮ Optimal mechanism, minimizing total payment

  6. Mechanism design: example ◮ Three jobs ◮ p j = 1 for all j ◮ w 1 = 5, w 2 = 2 and w 3 = 3 or w 3 = 1 σ 1 : π 2 = 4 , π 3 = 3 w 1 = 5 w 3 = 3 w 2 = 2 σ 2 : w 1 = 5 w 2 = 2 w 3 = 1 π 2 = 2 , π 3 = 2

  7. Mechanism design: example ◮ Three jobs ◮ p j = 1 for all j ◮ w 1 = 5, w 2 = 2 and w 3 = 3 or w 3 = 1 σ 1 : π 2 = 4 , π 3 = 3 w 1 = 5 w 3 = 3 w 2 = 2 σ 2 : w 1 = 5 w 2 = 2 w 3 = 1 π 2 = 2 , π 3 = 2 ◮ π 3 ( σ 2 ) − S 3 ( σ 2 ) < π 3 ( σ 1 ) − S 3 ( σ 1 ): Job 3 prefers σ 1

  8. Mechanism design: example ◮ Three jobs ◮ p j = 1 for all j ◮ w 1 = 5, w 2 = 2 and w 3 = 3 or w 3 = 1 σ 1 : π 2 = 4 , π 3 = 3 w 1 = 5 w 3 = 3 w 2 = 2 σ 2 : w 1 = 5 w 2 = 2 w 3 = 1 π 2 = 2 , π 3 = 4 ◮ π 3 ( σ 2 ) − S 3 ( σ 2 ) < π 3 ( σ 1 ) − S 3 ( σ 1 ): Job 3 prefers σ 1 ◮ Increasing π 3 ( σ 2 ) reduces total payment

  9. Model ◮ Agents with jobs: types t j = ( w j , p j ) ∈ T j ; (partly) private ◮ Mechanism strategies: report type t ′ j ∈ T j ◮ Mechanism output: machine sequence ( ES ) + payments ◮ Truthful mechanisms ◮ Payments: individual rational (IR) & incentive compatible (BNIC) (IR) π j ( t j ) − w j ( t j ) ES j ( t j ) ≥ 0 π j ( t j ) − w j ( t j ) ES j ( t j ) ≥ π j ( t ′ j ) − w j ( t j ) ES j ( t ′ (BNIC) j )

  10. Overview Open Problem [Heydenreich et al. 2008] “Identify (closed formulae for) optimal 2-D mechanisms.” Model Comments Solution method 0-D Optimization problem Priorities: w j / p j 1-D Only w j private Priorities: w j / p j 1.5-D Reported p j ≥ true p j LP-compactification 2-D Priorities: w j / E ( p j | w j ) Lemma Priorities result in ‘nice’ properties

  11. 1-Dimensional ◮ Agents with jobs: p j known, w j private ◮ Strategies: report w ′ j ◮ Mechanism output: sequences ( ES ) + payments ◮ Truthful mechanisms: Bayes-Nash incentive compatible payments ◮ [Heydenreich et al., WINE 2008; Duives et al. 2015]

  12. Type graph Given output sequences ( ES ), construct a type graph for each agent: ◮ Complete di-graph ◮ Node for each type + dummy ◮ Length of arc ( w j , w ′ j ): gain by reporting type w ′ j if really w j l ( w j , w ′ j ) = w j ( ES j ( w ′ j ) − ES j ( w j )) · · · dummy . . . w 1 w 2 < < < w k j j j Lemma Bayes-Nash implementable ⇔ no negative cycles ⇔ monotonicity. Lemma Given ES, the minimal BNIC payment for agent j reporting w j is − Dist ( w j , dummy ) .

  13. Optimal 1-D mechanism · · · dummy . . . w 1 w 2 w k < < < j j j Lemma Shortest path from w i j to the dummy traverses ( w i j , . . . , w k j , dummy ) . Lemma j )( w h − 1 Dist ( w i j , dummy ) = − w i h > i ES j ( w h − w h j ES j ( w j ) + � j ) . j

  14. Optimal 1-D mechanism · · · dummy . . . w 1 w 2 w k < < < j j j Lemma Optimal mechanism minimizes � � � � ES j ( w i ϕ j ( w i j ) w i j + ( w i − 1 − w i � ϕ j ( w h j ) j ) j ) j j i h < i � � � = ϕ j ( w j ) w j ES j ( w j ) , ( w 1 ,..., w n ) j j � h < i ϕ j ( w h j ) j + ( w i − 1 where w i j = w i − w i j ) j ϕ j ( w i j )

  15. Optimal 1-D mechanism � � � min ϕ j ( w j ) w j ES j ( w j ) ( w 1 ,..., w n ) j j Many sequencing optimization problems → priority: w j / p j . Corollary Optimal mechanism can be implemented as dominant strategies. Corollary Optimal mechanism is deterministic. Corollary Optimal mechanism is IIA.

  16. 1.5-Dimensional ◮ Agents with jobs: t j = ( w j , p j ) private ◮ Strategies: report t ′ j with p j ( t ′ j ) ≥ p j ◮ Mechanism output: sequences ( ES ) + payments ◮ Truthful mechanisms: Bayes-Nash incentive compatible payments ◮ [H. & Uetz, IPCO 2013]

  17. Type graph p ℓ · · · j ∨ . . . . . . . . . ∨ dummy p 2 · · · j ∨ p 1 · · · j w 1 w 2 < < < w k j j j Lemma No ‘dominating’ shortest path.

  18. Optimal 1.5-D mechanism Theorem (H. & Uetz, IPCO 2013) Polynomial size LP formulation for (BNIC) 1.5-D problem. Results in randomized outcome, i.e. a lottery over sequences for each vector of types. Lemma Optimal randomized mechanism > optimal deterministic mechanism. Lemma Optimal determinist mechanism > optimal deterministic IIA mechanism. Corollary Optimal mechanism does not have priorities.

  19. 2-Dimensional ◮ Agents with jobs: t j = ( w j , p j ) private ◮ Strategies: report any t ′ j ◮ Mechanism output: sequences ( ES ) + payments ◮ Truthful mechanisms: Bayes-Nash incentive compatible payments

  20. Type graph p ℓ · · · j ∨ . . . . . . . . . ∨ dummy p 2 · · · j ∨ p 1 · · · j w 1 w 2 < < < w k j j j Lemma ES j ( w j , p j ) = ES j ( w j , p ′ j ) for all j , w j , p j , p ′ j .

  21. Proof p ℓ · · · j ∨ . . . . . . . . . ∨ dummy p 2 · · · j ∨ p 1 · · · j w 1 w 2 < < < w k j j j

  22. Proof p ℓ · · · j ∨ . . . . . . . . . dummy ∨ p 2 · · · j ∨ p 1 · · · j w 1 w 2 w k < < < j j j Equal utility for all types with equal w j .

  23. Proof p ℓ · · · j ∨ . . . . . . . . . dummy ∨ p 2 · · · j ∨ p 1 · · · j w 1 w 2 w k < < < j j j Monotonicity: w j ≥ w ′ j ⇔ ES j ( w j , p j ) ≤ ES j ( w ′ j , p ′ j ) ∀ w j , w ′ j , p j , p ′ j .

  24. Proof p ℓ · · · j ∨ . . . . . . . . . ∨ dummy p 2 · · · j ∨ p 1 · · · j w 1 w 2 < < < w k j j j For all choices of p i j , . . . , p h j : k − 1 � � � j ) − Es j ( w h +1 , p h +1 π j ( w i j , p i j ) ≥ w k j Es j ( w k j , p k w h Es j ( w h j , p h j )+ ) . j j j h = i

  25. Proof p ℓ · · · j ∨ . . . . . . . . . dummy ∨ p 2 · · · j ∨ p 1 · · · j w 1 w 2 w k < < < j j j ES j ( w j , p j ) = ES j ( w j , p ′ j ) for all j , w j , p j , p ′ j .

  26. Optimal 2-D mechanism ◮ Reduction to 1-D case with (conditional) stochastic processing requirement ◮ Solved by priorities: w j / E ( p j | w j ) [Rothkopf, 1966] ◮ Dominant strategy implementation ◮ IIA

  27. Summary ◮ 2-D sequencing mechanism design reduces to 1-D case ◮ Priority sequencing rule ◮ 1.5-D optimal mechanism has no priority sequencing rule Open problem: ◮ 2-D mechanism as an approximately optimal 1.5-D mechanism?

Recommend


More recommend