1 / 19

The Price of Anarchy: Some Old and New Results

The Price of Anarchy: Some Old and New Results. Tim Roughgarden Stanford University. Algorithms and Game Theory. Recent Trend: design and analysis of algorithms and systems with self-interested agents Motivation: the Internet auctions (eBay, sponsored search, etc.)

Download Presentation

The Price of Anarchy: Some Old and New Results

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. The Price of Anarchy: Some Old and New Results Tim Roughgarden Stanford University

  2. Algorithms and Game Theory Recent Trend: design and analysis of algorithms and systems with self-interested agents Motivation: the Internet • auctions (eBay, sponsored search, etc.) • competition among end users, ISPs, etc. Traditional approach: • agents classified as obedient or adversarial • examples: distributed algorithms, cryptography

  3. Inefficiency of Equilibria Obvious fact: many modern applications in CS involve autonomous, self-interested agents • motivates noncooperative games as modeling tool Unsurprising fact: equilibria of noncooperative games typically inefficient • i.e., don't optimize natural objective functions • e.g., Nash equilibrium: an outcome such that no player better off by switching strategies Price of anarchy: quantify inefficiency w.r.t some objective function.

  4. Performance Guarantees Good news: in theoretical CS, have lots of techniques for measuring inefficiency. • motivated by NP-completeness, real-time algorithms, etc. Definition: approximation ratio (w.r.t. some objective function): the closer to 1 the better protagonists's obj fn value optimal obj fn value

  5. Inefficiency of Nash Flows Note: selfish routing does not minimize average delay (observed informally by [Pigou 1920]) • Cost of equilibrium flow = 1•1 + 0•1 = 1 • Cost of optimal (min-cost) flow = ½•½ +½•1 = ¾ • Price of anarchy := equilibrium/OPT ratio = 4/3 x 1 ½ s t 1 0 ½

  6. ½ ½ x 1 s t ½ ½ x 1 Braess’s Paradox Initial Network: Cost = 1.5

  7. ½ ½ x 1 s t ½ ½ x 1 Braess’s Paradox Initial Network: Augmented Network: x 1 0 s t x 1 Cost = 1.5 Cost = 2

  8. ½ ½ x 1 s t ½ ½ x 1 Braess’s Paradox Initial Network: Augmented Network: All traffic incurs more cost![Braess 68] • also has physical analogs [Cohen/Horowitz 91] x 1 0 s t x 1 Cost = 1.5 Cost = 2

  9. Unbounded Inefficiency Example: large prop delay + small queuing delay vs. small prop delay + large queuing delay • one unit (comprising many flows) selfish traffic Delay [secs] s t Rate x

  10. Unbounded Inefficiency Example: large prop delay + small queuing delay vs. small prop delay + large queuing delay • one unit (comprising many flows) selfish traffic c(x) = xd Delay [secs] s t c(x) = 1 Rate x

  11. Unbounded Inefficiency Example: large prop delay + small queuing delay vs. small prop delay + large queuing delay • one unit (comprising many flows) selfish traffic c(x) = xd 1 Delay [secs] s t c(x) = 1 0 "selfish" routing Rate x

  12. c(x) = xd 1 1-Є s t c(x) = 1 0 Є Unbounded Inefficiency Example: large prop delay + small queuing delay vs. small prop delay + large queuing delay • one unit (comprising many flows) selfish traffic Delay [secs] "selfish" routing "optimal" routing Rate x

  13. c(x) = xd 1 1-Є s t c(x) = 1 0 Є Unbounded Inefficiency Example: large prop delay + small queuing delay vs. small prop delay + large queuing delay • one unit (comprising many flows) selfish traffic Hope: performance guarantees easier to achieve in overprovisioned network. Delay [secs] "selfish" routing "optimal" routing Rate x

  14. Benefit of Overprovisioning Suppose: network is overprovisioned by β > 0 (β fraction of each edge unused). Then: Delay of selfish routing at most ½(1+1/√β) times that of optimal. • arbitrary network size/topology, traffic matrix • special case of [Roughgarden STOC 02] Moral: Even modest (10%) over-provisioning sufficient for near-optimal routing.

  15. But Are We at Equilibrium? Since 2002:price of anarchy (i.e., worst eq/OPT ratio) analyzed in many models. Possible critique: Interpretation of a POA bound presumes players reach equilibrium. • assumes players are “rational” and also successfully coordinate on an equilibrium

  16. Example Generalization Definition: a sequence s1,s2,...,sTof outcomes is no-regret if: • for each player i, each fixed action qi: • average cost player i incurs over sequence no worse than playing action qievery time • simple hedging strategies can be used by players to enforce this (for suff large T) Interpretation: players are at least “somewhat smart”, but don’t necessarily coordinate.

  17. Intrinsic Robustness of the Price of Anarchy Informal Theorem:[Roughgarden STOC 09]in many applications, every bound on the price of anarchy (for Nash equilibria) extends automatically to (e.g.) all no-regret sequences. Example Application:selfish routing games (“nonatomic” or “atomic”) with cost functions in an arbitrary fixed set.

  18. Outline of Proof • main definition: a “canonical way” to bound the price of anarchy (for pure equilibria) • theorem 1: every POA bound proved “canonically” is automatically far stronger • e.g., even applies “out-of-equilibrium”, assuming no-regret play • theorem 2: canonical method provably yields optimal bounds in fundamental cases

  19. Connections + Challenges • dynamics in games + inefficiency bounds • e.g., how do details of dynamics affect which equilibrium is reached? • possible application in control theory: worst-case performance guarantees for distributed approximations of a centralized optimum • possible application in control theory: meaningful guarantees despite non-convergence of system

More Related