1 / 47

Minimal Preference Elicitation in Combinatorial Auctions

Minimal Preference Elicitation in Combinatorial Auctions. Wolfram Conen Tuomas Sandholm Xonar GmbH Carnegie Mellon University Computer Science Department. Outline. Combinatorial auctions for multi-item auctions “The revelation problem” Previous approaches

monte
Télécharger la présentation

Minimal Preference Elicitation in Combinatorial 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. Minimal Preference Elicitation in Combinatorial Auctions Wolfram Conen Tuomas Sandholm Xonar GmbH Carnegie Mellon University Computer Science Department

  2. Outline • Combinatorial auctions for multi-item auctions • “The revelation problem” • Previous approaches • Our approach: Elicitor “agent” • Topological observations that motivate elicitation • Different elicitation queries • Policy dependent elicitor algorithms • General policy independent elicitor framework (with data structures & assimilation algorithms) & specific elicitor algorithms • Making the elicitor incentive compatible

  3. Combinatorial auction (CA) • Can bid on combinations of items [Rassenti,Smith & Bulfin 82]... • Bidder’s perspective • Allows bidder to express what she really wants • No need for lookahead / counterspeculationing of items • Auctioneer’s perspective: • Automated optimal bundling • Binary winner determination problem: • Label bids as winning or losing so as to maximize sum of bid prices • Each item can be allocated to at most one bid • NP-complete [Rothkopf et al 98 using Karp 72] • Inapproximable [Sandholm IJCAI-99 using Hastad 99]

  4. Another complex problem in (single-shot) combinatorial auctions: “The revelation problem” • Bidders may need to bid on all 2#items combinations • Need to compute the valuation for each combination • Each valuation computation can be NP-complete • For example if a carrier company bids on trucking tasks: TRACONET [Sandholm AAAI-93] • Need to communicate the bids • Need to reveal the bids

  5. Setting Combinatorial auction: m items for sale • Private values auction, no allocative externalities • Each bidder i has value function, vi: 2m R • Unique valuations (to ease presentation) Can avoid unnecessary computation/revelation/communication of valuations!

  6. Approaches for tackling the revelation problem • Classic single-shot full revelation mechanims (Vickrey-Clarke-Groves) • Exponentially many valuations revealed • (Ascending) mechanisms with price feedback (iBundle, [Parkes et al 1999] , akBa [Wurman et al. 2000] , etc.) • Can save revelation • Need exponential revelation in worst case [Nisan 2001] • Our new approach: an elicitor “agent” • Knows things that individual bidders don’t (others’ bids so far) • Asks non-redundant questions from bidders to focus their revelation • Can save revelation • Thrm. Exponential revelation in worst case if only value and order queries are allowed (even with 1 bidder) • Can be combined with price feedback mechanisms

  7. Elicitor algorithms • Query policy dependent elicitor algorithms • Algorithm & query policy are intertwined • Based on search algorithms where each search step involves asking a bidder a question • Policy independent elicitor algorithms • General framework & specific algorithms • Can support any query policy • Use exponential memory (in worst case) • Note: Query policies are online control policies, i.e. contingency plans

  8. Terminology (X1,...,X#bidders) is a collection • Bundle Xi is earmarked for agent i • An allocation is a feasible collection (i.e., collection where Xi’s don’t overlap in items) Objectives: (1) Find Pareto efficient allocation(s) (2) Find social welfare maximizing allocation(s)

  9. Rank Lattice Rank of Bundle Ø A B AB for Agent 1 4 2 3 1 for Agent 2 4 3 2 1 [1,1] [1,2] [2,1] [1,3] [2,2] [3,1] [1,4] [2,3] [3,2] [4,1] [2,4] [3,3] [4,2] [3,4] [4,3] [4,4] Infeasible Feasible Dominated

  10. Query Policy Independent Elicitation Algorithms: Computing Pareto Optima s=(1,...,1); PAR = []; OPEN = [s]; while OPEN ≠ [] do Remove(c,OPEN); SUC = suc(c); ifFeasible(c) then PAR = PAR  {c}; Remove(SUC,OPEN); elseforeach node є SUC do if node OPEN andUndominated(node,PAR) thenAppend(node,OPEN)

  11. Value-Augmented Rank Lattice Value of Bundle Ø A B AB for Agent 1 0 4 3 8 for Agent 2 0 1 6 9 17 [1,1] 14 13 [1,2] [2,1] 10 12 9 [1,3] [2,2] [3,1] 8 9 [1,4] [2,3] [3,2] [4,1] [2,4] [3,3] [4,2] [3,4] [4,3] [4,4]

  12. Query Policy Independent Elicitation Algorithms: Computing Welfare Maxima s=(1,...,1); OPEN = {s}; CLOSED = Ø; while OPEN ≠ Ødo c = arg maxc є OPEN Σi є N vi(ci) OPEN = OPEN \ {c}; ifFeasible(c) then return(c); CLOSED = CLOSED  {c}; SUC = suc(c); foreach n є SUC do if node OPEN and node CLOSED then OPEN = OPEN  {node}

  13. Policy independent elicitor algorithms

  14. Our Query Types for Agent Interrogation • Order information: Which bundle do you prefer, A or B? • Value information: What is your valuation for bundle A? (Answer: Exact or Bounds) • Rank information: • What is the rank of bundle b? • What bundle is at rank x? • Given bundle b, what is the next lower (higher) ranked bundle?

  15. General Algorithmic Framework for Elicitation Algorithm Solve(Y,G) whilenotDone(Y,G) do o = SelectOp(Y,G)  Choose Question I = PerformOp(o,N)  Ask bidder G = Propagate(I,G)  Update Graph Y = Candidates(Y,G)  Curtail set of candidate collections / allocations Input: Y – set of collections or allocations G – partially augmented order graph Output: Y – set of optimal solutions

  16. (Partially) Augmented Order Graph ∞ ∞ ∞ ∞ ∞ ∞ Agent1 Ø B A AB A Ø 0 0 0 0 > Allocations B B 4 0 3 6 2 6 1 9 Ø A B AB Agent2 1 1 0 0 1 6 [1,1] [1,2] [2,1] Rank Upper Bound [1,3] [2,2] [3,1] 1 9 [1,4] [2,3] [3,2] [1,4] AB [2,4] [3,3] [4,2] 6 [3,4] [4,3] Lower Bound [4,4] Some interesting procedures for combining different types of info

  17. We present algorithms that use any combination of value, order & rank queries • If value queries are used, all social welfare maximizing allocations are guaranteed to be found • Otherwise, all Pareto efficient allocation are guaranteed to be found • We propose several query policies that are geared toward reducing the number of queries needed

  18. Incentive compatibility of the different approaches • Classic single-shot full revelation mechanims (Vickrey-Clarke-Groves) • Can be made dominant strategy incentive compatible • (Ascending) mechanisms with price feedback (iBundle, akBa, etc.) • Can be made incentive compatible in weaker equilibrium notions • Our new approach: an elicitor “agent” • Elicitor’s questions leak information about others’ preferences • Can be made incentive compatible in weaker equilibrium notions • Ask enough questions to determine VCG prices • Could interleave these “extra” questions with real questions • To avoid lazyness; Not necessary from an incentive perspective

  19. Conclusions • Combinatorial auctions are desirable & winner determination algorithms now scale to the large • Another problem: “The Revelation Problem” • Valuation computation / revelation / communication • Presented the design for an elicitor for combinatorial auctions that focuses revelation • Can save revelation • Provably find the welfare maximizing or Pareto efficient allocations • Policy dependent search algorithms for elicitation • Based on topological observation • Policy independent general elicitation framework • Any combination of value, order & rank queries • Several algorithm instantiations in the paper • Several query policies • Presented a way to make the elicitor incentive compatible • Elicitor can be combined with price feedback mechanisms

  20. Future research • Evaluating the elicitor • Savings in revelation (how many queries needed ?) • In general case / in cases with special preference structure • Worst / average case • Generalizing the elicitor • To (combinatorial) exchanges • To (combinatorial) markets with side constraints • To (combinatorial) markets with multiattribute features

  21. For combinatorial auctions

More Related