be beyond p nd plur luralit ality tr truth bias in binary
play

Be Beyond P nd Plur luralit ality: : Tr Truth-Bias in Binary - PowerPoint PPT Presentation

Be Beyond P nd Plur luralit ality: : Tr Truth-Bias in Binary Sc Scoring Rules Svetlana Obraztsova, Omer Lev, Evangelos Markakis, Zinovi Rabinovich, and Jeffrey S. Rosenschein Wh Why tr trut uth-b h-bias? s? Wh Why tr trut uth-b


  1. Be Beyond P nd Plur luralit ality: : Tr Truth-Bias in Binary Sc Scoring Rules Svetlana Obraztsova, Omer Lev, Evangelos Markakis, Zinovi Rabinovich, and Jeffrey S. Rosenschein

  2. Wh Why tr trut uth-b h-bias? s?

  3. Wh Why tr trut uth-b h-bias? s? Pinocchio Jiminy Gideon Cricket the cat 1 st preference Puppet Do what the Pleasure show blue fairy says island 2 nd preference Pleasure Puppet Puppet island show show Do what the Pleasure 3 rd preference Do what the blue fairy says island blue fairy says

  4. Wha What’s tr s trut uth-b h-bias? s? Each voter gets an ε extra utility from being truthful. The ε is small enough so that a voter would rather change the winner to someone more to its liking than to be truthful.

  5. Wh Why tr trut uth-b h-bias? s? Pinocchio Jiminy Gideon Cricket the cat 1 st preference Puppet Do what the Pleasure show blue fairy says island 2 nd preference Pleasure Puppet Puppet island show show Do what the Pleasure 3 rd preference Do what the blue fairy says island blue fairy says

  6. Wha What’s t s the he k- k- appr approval v al voting ing ru rule? Each voter gives a point to k candidates and the rest do not receive any point from the voter. The candidate with the most points, wins. When k =1, this is plurality. When k= number of candidates-1, this is veto.

  7. Ve Veto

  8. Wha What a t about t out the eq he equi uilibria? They don’t necessarily exist… a ≻ c ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ b ≻ a Lexicographic tie-breaking rule

  9. Wha What a t about t out the eq he equi uilibria? They don’t necessarily exist… a ≻ c ≻ b a ≻ b ≻ c c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ b ≻ a c ≻ b ≻ a Lexicographic tie-breaking rule

  10. Wha What a t about t out the eq he equi uilibria? They don’t necessarily exist… a ≻ c ≻ b a ≻ b ≻ c c ≻ a ≻ b a ≻ b ≻ c c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a Lexicographic tie-breaking rule

  11. Wha What a t about t out the eq he equi uilibria? They don’t necessarily exist… a ≻ c ≻ b a ≻ c ≻ b a ≻ b ≻ c c ≻ a ≻ b a ≻ b ≻ c c ≻ a ≻ b a ≻ c ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ a ≻ b c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a c ≻ b ≻ a Lexicographic tie-breaking rule

  12. Ca Can we e sa say anything g abou out t it? t? If an equilibrium is non-truthful: The winner’s score is the same as in the truthful setting. There is a threshold candidate , that would win if the winner lost a point. All non-truthful voters veto a “runner-up”, i.e., candidates one point away from winning.

  13. Ca Can we e sa say if f ca candidate e w w has has an e an equilibr uilibrium w ium whe here it it w wins ins? No. Finding if there is an equilibrium in which candidate w is the winner in a veto election with truth-biased voters is NP-complete. Furthermore, Finding if there is an equilibrium a veto election with truth-biased voters is NP- complete.

  14. But But do do no not f falt alter! The candidate following w in the tie breaking rule – t – has a truthful score at least as high as w . All voters that do not veto w prefer it to the candidate following w in the tie breaking rule ( w ≻ i t ).

  15. The The t trut uth( h(-bias bias) is is o out ut t the here! In veto elections with truth-biased voters, if the 2 conditions hold for a candidate w , determining if there is an equilibrium in which it wins can be done in polynomial time. Not true for each condition separately!

  16. Cr Crea eati ting g a graph: � po potent ntial de ial deviat viatio ions ns Nodes are source, sink, C (candidates) , V (voters) For a voter v truthfully vetoing r we add an edge ( r,v ). And for each c such that w ≻ v c ≻ v r we add an edge ( v,c ). C 1 1 = y t i c a p a c C 2 c a p a c i t y = 1 v i r C l

  17. Cr Crea eati ting g a graph: � de deviat viatio ions ns If a candidate c needs more points to beat w , there is an edge ( source,c ) with capacity of the score it needs to add to become a runner-up. If a candidate c beats w , there is an edge ( c,sink ) with capacity of the score it needs to lose to become a runner-up.

  18. Ma Maxflo xflow If maxflow<incoming to sink – not enough points changed to make w the winner. If maxflow=incoming to sink – some tweaks to flow manifestation will show the flow means voters moving veto from some candidates to others.

  19. But But w what hat abo about ut t the he conditi con tion ons? s? (1 (1) The candidate following w in the tie breaking rule – t – has a truthful score at least as high as w . Condition ensured t was the threshold candidate

  20. But But w what hat abo about ut t the he con conditi tion ons? s? (2 (2) All voters that do not veto w prefer it to the candidate following w in the tie breaking rule ( w ≻ i t ). Condition ensured no one would veto w , making t , the threshold candidate, the winner.

  21. Pl Plur urality ty

  22. Pl Plur urality ty tr trut uth-b -bias Equilibrium not ensured. Knowing if equilibrium exists is NP-complete. Winner increases score (if not-truthful) Runner-up score does not change Obraztsova et al. (SAGT 2013)

  23. k -a -approval

  24. k -a -approval tr trut uth-b h-bias Winner score can stay the same or rise. Runner-up score can increase or decrease

  25. Fu Future directions Other voting rules! (we’re not even sure what’s going on in non-binary scoring rules…) Simulation / analysis: how good are the winners? More useful conditions to make problems poly-solvable. Classes of truth-biased equilibria?

  26. bias Thanks for listening!

Recommend


More recommend