On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S....
-
Upload
warren-waitman -
Category
Documents
-
view
223 -
download
3
Transcript of On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S....
![Page 1: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/1.jpg)
On the Robustness of Preference Aggregation in Noisy EnvironmentsAriel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka
![Page 2: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/2.jpg)
Outline
• Motivation• Definition of Robustness• Results:
– About Robustness in general. – Sketch of results about specific voting rules.
• Conclusions
Motivation Definition Results Conclusions
![Page 3: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/3.jpg)
Voting in Noisy Environments
• Election: set of voters N={1,...,n}, alternatives / candidates A={x1,...,xm}.
• Voters have linear preferences Ri; winner of the election determined according to a social choice function / voting rule.
• Preferences may be faulty:– Agents may misunderstand choices. – Robots operating in an unreliable
environment.
Motivation Definition Results Conclusions
![Page 4: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/4.jpg)
Possible Informal Definitions of Robustness
• Option 1: given a uniform distribution over preference profiles, what is the probability of the outcome not changing, when the faults are adversarial?
• Reminiscent of manipulation. • Option 2 (ours): given the worst
preference profile and a uniform distribution over faults, what is the probability of the outcome not changing?
Motivation Definition Results Conclusions
![Page 5: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/5.jpg)
Formal Definition of Robustness
• Fault: a “switch” between two adjacent candidates in the preferences of one voter.– Depends on representation; consistent,
“quite good” representation.
Motivation Definition Results Conclusions
![Page 6: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/6.jpg)
Faults Illustrated
3
2
1
ran
k
x2x2
x3x3
x1x1 3
2
1
ran
k
x3x3
x1x1
x2x2
x3x3
x2x2
Motivation Definition Results Conclusions
![Page 7: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/7.jpg)
Formal Definition of Robustness
• Fault: a “switch” between two adjacent candidates in the preferences of one voter.– Depends on representation; consistent,
“quite good” representation.
• Dk(R) = prob. dist. over profiles; sample: start with R and perform k independent uniform switches.
• The k-robustness of F at R is: (F,R) = PrR1Dk(R)[F(R)=F(R1)]
Motivation Definition Results Conclusions
![Page 8: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/8.jpg)
Robustness Illustrated
F = Plurality. 1-Robustness at R is 1/3.
2
1
ran
k
x1x1
x2x2 2
1
ran
k
x1x1
x2x2 2
1
ran
k
x2x2
x1x1
x1x1
x2x2 x2x2
x1x1
Motivation Definition Results Conclusions
![Page 9: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/9.jpg)
Formal Definition of Robustness
• Fault: a “switch” between two adjacent candidates in the preferences of one voter.– Depends on representation; consistent,
“quite good” representation.
• Dk(R) = prob. dist. over profiles; sample: start with R and perform k independent uniform faults.
• The k-robustness of F at R is: (F,R) = PrR1Dk(R)[F(R)=F(R1)]
• The k-robustness of F is: (F) = minR (F,R)
Motivation Definition Results Conclusions
![Page 10: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/10.jpg)
Simple Facts about Robustness
• Theorem: k(F) (1(F))k
• Theorem: If Ran(F)>1, then 1(F) < 1.
• Proof:
RR R1R1
Motivation Definition Results Conclusions
![Page 11: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/11.jpg)
1-robustness of Scoring rules
• Scoring rules: defined by a vector =1,...,m, all i i+1. Each candidate receives i points from every voter which ranks it in the i’th place. – Plurality: =1,0,...,0– Borda: =m-1,m-2,...,0
• AF = {1 ≤ i ≤ m-1: i > i+1}; aF = |AF|
• Proposition: 1(F) (m-1-aF)/(m-1)
• Proof: – A fault only affects the outcome if i > i+1.
– There are aF such positions per voter, out of m-1.
• Proposition: 1(F) (m-aF)/m
Motivation Definition Results Conclusions
![Page 12: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/12.jpg)
Results about 1-robustness
Rule Lower Bound Upper Bound
Scoring (m-1-aF)/(m-1) (m-aF)/m
Copeland 0 1/(m-1)
Maximin 0 1/(m-1)
Bucklin (m-2)/(m-1) 1
Plurality w. Runoff
(m-5/2)/(m-1) (m-5/2)/(m-1)+5m/(2m(m-1))
Motivation Definition Results Conclusions
![Page 13: On the Robustness of Preference Aggregation in Noisy Environments Ariel D. Procaccia, Jeffrey S. Rosenschein and Gal A. Kaminka.](https://reader035.fdocuments.us/reader035/viewer/2022062313/56649c765503460f9492a157/html5/thumbnails/13.jpg)
Conclusions
• k-robustness: worst-case probability that k switches change outcome.
• Connection to 1-robustness:– High 1-robustness high k-robustness.– Low 1-robustness can expect low k-robustness.
• Tool for designers:– Robust rules: Plurality, Plurality w. Runoff, Veto,
Bucklin. – Susceptible: Borda, Copeland, Maximin.
• Future work:– Different error models. – Average-case analysis.
Motivation Definition Results Conclusions