1 / 30

Pure and Bayes -Nash Price of Anarchy for GSP

Pure and Bayes -Nash Price of Anarchy for GSP. Renato Paes Leme Éva Tardos. Cornell. Cornell & MSR. Keyword Auctions. sponsored search links. organic search results. Keyword Auctions. Keyword Auctions. Selling one Ad Slot. $2. $5. Prospective advertisers. $7. $3.

ledell
Download Presentation

Pure and Bayes -Nash Price of Anarchy for GSP

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. Pure and Bayes-Nash Price of Anarchy for GSP RenatoPaesLemeÉvaTardos Cornell Cornell & MSR

  2. Keyword Auctions sponsored search links organic search results

  3. Keyword Auctions

  4. Keyword Auctions

  5. Selling one Ad Slot $2 $5 Prospective advertisers $7 $3

  6. Selling one Ad Slot $2 Pays $5 per click $5 Vickrey Auction $7 • Truthful • Efficient • Simple • … $3

  7. Auction Model b1 $$$ $$$ b3 b2 $$ b4 b5 $$ $ b6 $

  8. Auction Model b1 $$$ Vickrey Auction VCG Auction $$$ b3 b2 • Truthful • Efficient • Simple (?) • … $$ Generalized Second Price Auction • Not Truthful • Not Efficient • Even Simpler • … b4 b5 $$ $ b6 $

  9. Our Results • Generalized Second Price Auction (GSP), although not optimal, has good social welfare guarantees: • 1.618 for Pure Price of Anarchy • 8 for Bayes-Nash Price of Anarchy • GSP with uncertainty

  10. (Simplified) Model • αj : click-rate of slot j • vi : value of player i • bi : bid (declared value) • Assumption: bi≤ vi Since playing bi> vi is dominated strategy. b1 α1 v1 α2 b2 v2 α3 v3 b3

  11. (Simplified) Model pays b1 per click v1 b1 α1 b2 v2 α2 α3 v3 b3

  12. (Simplified) Model σ = π-1 bi αj vi i = π(j) j = σ(i) Utility of player i : ui(b) = ασ(i) ( vi - bπ(σ(i) + 1))

  13. Model σ = π-1 bi αj vi i = π(j) j = σ(i) Utility of player i : ui(b) = ασ(i) ( vi - bπ(σ(i) + 1)) next highest bid

  14. Model σ = π-1 bi αj vi i = π(j) j = σ(i) Nash equilibrium: ui(bi,b-i) ≥ ui(b’i,b-i) Is truth-telling always Nash ?

  15. Example Non-truthful u1 = 1 (2-1) u1 = 0.9(2-0) v1 = 2 b1 = 2 b1 = 0.9 α1 = 1 v2 = 1 b1 = 1 α2 = 0.9

  16. Measuring inefficiency σ = π-1 bi αj vi i = π(j) j = σ(i) Social welfare = ∑iviασ(i) Optimal allocation = ∑iviαi

  17. Measuring inefficiency Opt Price of Anarchy = max = SW(Nash) Nash

  18. Main Theorem 1 • Thm: Pure Price of Anarchy ≤ 1.618 If bi≤ vi and (b1…bn) are bid in equilibrium, then for the allocation σ : ∑iviασ(i) ≥ 1.618-1∑iviαi • Previously known [EOS, Varian]: Price of Stability = 1

  19. GSP as a Bayesian Game Modeling uncertainty:

  20. GSP as a Bayesian Game b b ?

  21. GSP as a Bayesian Game Idea: Optimize against a distribution. b b b b b b b b b b b b b

  22. Bayes-Nash solution concept • Bayes-Nash models the uncertainty of other players about valuations • Values vi are independent random vars • Optimize against a distribution • Thm:Bayes-NashPoA≤ 8

  23. Bayesian Model v1 ~ V1 b1(v1) α1 b2(v2) v2 ~ V2 α2 α3 v3 ~ b3(v3) V3

  24. Model bi(vi) αj vi ~ Vi i = π(j) j = σ(i) Bayes-Nash equilibrium: E[ui(bi,b-i)|vi] ≥ E[ui(b’i,b-i)|vi] Expectation over v-i

  25. Bayes-Nash Equilibrium vi are random variables μ(i) = slot that player i occupies in Opt (also a random variable) E[∑iviαμ(i)] Bayes-Nash PoA = E[∑iviασ(i)] • Previously known [G-S]: Price of Stability ≠ 1

  26. Sketch of the proof v1 α1 αi α2 vπ(i) ασ(j) v2 vj + ≥ 1 vj αi α3 ασ(j) therefore: vπ(i) v3 vπ(i) ασ(j) 1 1 Simple and intuitive condition on matchings in equilibrium. ≥ ≥ or vj αi 2 2 Opt

  27. Sketch of the proof αi vj ασ(j) vπ(i) vπ(i) ασ(j) + ≥ 1 vj αi Need to show only for i < j and π(i) > π(j). It is a combination of 3 relations: ασ(j) ( vj – bπ(σ(j)+1) ) ≥αi ( vj – bπ(i) ) [ Nash ] bπ(i) ≤ vπ(i) [conservative] bπ(σ(j)+1) ≥0

  28. Sketch of the proof vπ(i) ασ(j) + 2 SW = ∑iασ(i) vi + αivπ(i) = ≥ 1 vj αi vπ(i) ασ(i) = ∑iαivi ≥ + vi αi ≥ ∑iαivi = Opt

  29. Proof idea:new structural condition Pure Nash version: αi vj vi ασ(i) +αivπ(i) ≥αivi ασ(j) vπ(i) Bayes Nash version: viE[ασ(i)|vi] + E[αμ(i) vπμ (i)|vi] ≥ ¼ viE[αμ(i)|vi]

  30. Upcoming results[Lucier-PaesLeme-Tardos] • Improved Bayes-Nash PoA to 3.164 • Valid also for correlated distributions • Future directions: • Tight Pure PoA (we think it is 1.259)

More Related