Evaluation of election outcomes under uncertainty Noam Hazon*, Yonatan Aumann*, Sarit Kraus*, Michael Wooldridge+ *Dept. of Computer Science Bar Ilan University Israel +Dept. of Computer Science University of Liverpool United Kingdom Outline A common way to aggregate agents preferences - voting Perfect Vs. imperfect information Complexity analysis of imperfect information model Constant number of candidates: P for most of the cases Number of candidates as a parameter: #P-Hard EVALUATION Vs. CHANCE-EVALUATION Conclusion, Future work Avoid manipulation! A>B>C Me B! My “friend” Why voting? Preference aggregation - to a socially desirable decision Computational aspects when moving to agents Evaluating a voting protocol Manipulation Assuming perfect information Our model - imperfect information A probability distribution over a set of preferences What is the probability of a candidate to win? Real world motivation Avoid manipulation Reduce communication A>B>C B! Our model - results Candidates number Weights constant parameter Chance-Evaluation Evaluation equal P(p,b,c,m,i,…) P(p,b,c,m,i,…) bounded P(p,b,c,m,i,…) P(p,b,c,m,i,…) unbounded NP-Hard(b,c,m,i) NP-Hard(b,c,m,i) equal P(p) #P-Hard(p,b,c) bounded NP-Cpmplete(p) #P-Hard(p,b,c) unbounded NP-Cpmplete(p) #P-Hard(p,b,c) Social choice domain Set of voters, V = {V1,…,Vn} Corresponding weights, w1,…,wn Set of outcomes/candidates, Ω = {ω1,…, ωm} Imperfect information, k = 3 V1 , W1=1 1/2 V2 , W2=2 (ω1, ω2, 1/4 (ω3, ω1, 3/4 ω3) 1/3 ω2) 1/6 (ω2, ω1, ω3) V3 , W3=1 (ω1, ω2, 9/10 (ω1, ω2, 1/10 ω3) ω3) (ω1, ω2, ω3) (ω1, ω2, ω3) Voting systems- a review Binary Voting systems Plurality Approval Preferential voting systems Instant-runoff voting (IRV) Borda Condorcet systems Copeland (Tournament) Minimax The EVALUATION problem Given social choice domain imperfect information model of voters' preferences specific candidate ω* What is the probability that ω* will be chosen? Computational time complexity depends on n, m and k Constant number of candidates Voting scenario Vs. voting result A dynamic programming approach (for plurality): 0 1 2 V1 V2 0,0,0 1 0 1/2 1/4 ω1 ω1 1,0,0 0 1/2 0,1,0 0 1/3 1/3 3/4 ω2 ω2 0,0,1 0 1/6 2,0,0 0 0 1/2*1/4 1,1,0 0 0 1/3*1/4 + 1/2*3/4 1,0,1 0 0 1/6*1/4 0,2,0 0 0 1/3*3/4 0,1,1 0 0 1/6*3/4 0,0,2 0 0 0 1/6 ω3 (ω1,ω2,ω3) Time Complexity: O(n * # of rows * k) Representing a voting result A vector of [0,n]m Plurality Approval Borda A vector of : O(kn(mn)m) [0,n]m(m-1)/2 : 2+1 m O(kn ) All Condorcet systems A vector of [0,n]m! : O(knm+1) IRV Borda … : O(knm!+1) m is constant! Adding weights Borda, Copeland, Minimax and IRV are NP-Hard [Conitzer and Sandholm 2002] Only for Unbounded weights! Weights in Poly(n) O(Poly(n)m), Candidates number constant 2 m O(Poly(n) ), Weights Chance-Evaluation Evaluation equal P(p,b,c,m,i,…) bounded P(p,b,c,m,i,…) unbounded NP-Hard(b,c,m,i) equal parameter O(Poly(n)m!) bounded unbounded # of candidates as a parameter(1) Even without weights, EVALUATION for Plurality is #P-Hard! Voters V1 V2 0 0 Candidates 1/3 v0 C0 1/3 1/3 1/3 v1 1 1 1/3 v2 2 C1 1/3 C2 1/3 2/3 2 R0 Z0 1 R 1 Z # of candidates as a parameter(2) EVALUATION for Borda and Copeland is #P-Hard. 1/3 V1 0 V2 0 v0 (c0,g2,c1,c2,c3,z,g1) 1/3 1/3 v1 1/3 1/3 (c1,c0,g2,c2,c3,z,g1) Pv 1/3 v2 1 2 2/3 1 2 (c2,c0,c1,g2,c3,z,g1) 1/3 v3 1 z0 1 z1 z2 z3 1 1 1 (c3,c0,c1,c2,g2,z,g1) (z,c3,c2,c1,g1,g2,c0) (z,c3,c2,g1,c0,g2,c1) (z,c3,g1,c1,c0,g2,c2) (z,g1,c2,c1,c0,g2,c3) Pz # of candidates as a parameter(2) EVALUATION for Borda and Copeland is #P-Hard. Candidates number Weights constant parameter Chance-Evaluation Evaluation equal P(p,b,c,m,i,…) bounded P(p,b,c,m,i,…) unbounded NP-Hard(b,c,m,i) equal #P-Hard(p,b,c) bounded #P-Hard(p,b,c) unbounded #P-Hard(p,b,c) CHANCE-EVALUATION problem Given social choice domain imperfect information model of voters' preferences specific candidate ω* Is the probability that ω* will be chosen greater than 0 ? Weighted CHANCE-EVALUATION NP-Complete (in the strong sense) for Plurality! Vz B V0 V1 V2 V3 V4 W0 W1 W2 W3 W4 1/k B+1 1 … C0 K C1 … Ck Z Un-weighted CHANCE-EVALUATION Polynomial algorithm for Plurality: V1 V2 V3 V4 V5 V6 V7 V8 1/4 A 1/2 A 1/3 A 1/3 B 1/3 A 1/3 B 1/3 B 1/3 B 1/2 B 1/2 C 1/2 B 2/3 C 2/3 D 2/3 C 2/3 D 2/3 C 1/4 D 1/6 C V1' V2 1 1 1 V3 1 1 1 1 V4 1 1 1 s V2' 1 A 2 B 1 1 V6 V8 1 1 1 2 2 C t Conclusion Recall the previous results table: Candidates number Weights constant parameter Chance-Evaluation Evaluation equal P(p,b,c,m,i,…) P(p,b,c,m,i,…) bounded P(p,b,c,m,i,…) P(p,b,c,m,i,…) unbounded NP-Hard(b,c,m,i) NP-Hard(b,c,m,i) equal P(p) #P-Hard(p,b,c) bounded NP-Cpmplete(p) #P-Hard(p,b,c) unbounded NP-Cpmplete(p) #P-Hard(p,b,c) Future More voting protocols Approximation and/or heuristics for the hard problems {hazonn,aumann,sarit}@cs.biu.ac.il , {mjw}@csc.liv.ac.uk