1 / 13

A Computational Characterization of Multiagent Games with Fallacious Rewards

A Computational Characterization of Multiagent Games with Fallacious Rewards. Ariel D. Procaccia and Jeffrey S. Rosenschein. Lecture Outline. Introduction Modeling mistakes Robust equilibria. Persistent equilibrium pairs. Errors due to lies Closing Remarks. Introduction. Mistakes.

jace
Download Presentation

A Computational Characterization of Multiagent Games with Fallacious Rewards

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. A Computational Characterization of Multiagent Games with Fallacious Rewards Ariel D. Procaccia and Jeffrey S. Rosenschein

  2. Lecture Outline • Introduction • Modeling mistakes • Robust equilibria. • Persistent equilibrium pairs. • Errors due to lies • Closing Remarks Introduction Mistakes Lies Conclusions

  3. Introduction • Interactions between selfish agents are often modeled as noncooperative games. • Nash equilibrium is the central solution concept. • Computationally hard to find equilibrium. Additionally, agents lack complete knowledge. • Much attention has been devoted to learning a Nash equilibrium in games. Introduction Mistakes Lies Conclusions

  4. Sources of Fallacious Rewards • It is usually assumed that agents can observe others’ rewards. • What are the sources of these observations? • Direct observation. • Modeling. • Preference revelation. • In the second and third cases, fallacious rewards might be obtained. Introduction Mistakes Lies Conclusions

  5. Terminology • The Explicit Gameis the real game, whereas the implicit game is a game with same players and actions but different rewards (Bowling & Veloso 2004). • Definition: Let G be an explicit game and G’ be an implicit game. G,G’ is an -perturbed system if for all players i and actions a, |Ri(a)-R’i(a)|  . Introduction Mistakes Lies Conclusions

  6. -perturbed system: example L R (-10,-10) (9,-9) (9,-11) U (-9,9) (-11,9) (10,10) (8,8) D Introduction Mistakes Lies Conclusions

  7. Robust Equilibria • Definition:  is an -robust eq. of G if it is a Nash eq. in any -perturbed G’. • Example: (U,L) is 1-robust, (D,R) isn’t -robust. (2,2) (1,1) (0,0) (1,1) (0,0) (1,1) (0,0) (-1,-1) Introduction Mistakes Lies Conclusions

  8. -Robust equilibrium: Results • Lemma: Robust equilibria consist of pure strategies. • Theorem: It is possible to find an -robust equilibrium in polynomial time. • Robust equilibria rarely exist. Introduction Mistakes Lies Conclusions

  9. Persistent Equilibrium pairs • Definition: Let <G,G’> be a perturbed system,  an eq. in G and  in G’. <,> is an -persistent eq. pair iff for all i, |Ri()-Ri’()|  . • If  is -robust, <,> is -persistent for any -perturbed G’. (1,1) (¼,¼) (0,0) (½,½) (0,0) (-1,-1) (-1,-1) (0,0) Introduction Mistakes Lies Conclusions

  10. Persistent Equilibrium Pairs: Results • Theorem: Determining whether a perturbed system has an -persistent equilibrium pair is NP-complete, even for two players. • Proposition: G and G’ are -perturbed and zero-sum  Optimal strategies are -persistent. • Proposition: G and G’ are -perturbed  coordination equilibria are -persistent. Introduction Mistakes Lies Conclusions

  11. Game Manipulation • Setting: Nash equilibrium is chosen based on agents’ valuations of outcomes. • Agents may lie to improve their utility. (0,0) (1,1) (1,1) (2,0) (-1,1) (-1,0) Introduction Mistakes Lies Conclusions

  12. Game Manipulation: Results • Game-Manipulation: can player i manipulate and get reward > k? • Theorem: Game-Manipulation with at least 3 players is NP-complete, with 2 players is in P. Introduction Mistakes Lies Conclusions

  13. Closing Remarks • Problems of finding robust or persistent equilibria is a problem of the system designer. • We consider other manipulations: • Coalitional. • Benevolent. • Strong. • Incentive-compatible mechanisms? • Results are independent of specific algorithms; also applicable to repeated games. Introduction Mistakes Lies Conclusions

More Related