Optimal Impartial Selection

Σχετικά έγγραφα
Nowhere-zero flows Let be a digraph, Abelian group. A Γ-circulation in is a mapping : such that, where, and : tail in X, head in

Fractional Colorings and Zykov Products of graphs

Bayesian statistics. DS GA 1002 Probability and Statistics for Data Science.

A Bonus-Malus System as a Markov Set-Chain. Małgorzata Niemiec Warsaw School of Economics Institute of Econometrics

Ordinal Arithmetic: Addition, Multiplication, Exponentiation and Limit

Other Test Constructions: Likelihood Ratio & Bayes Tests

ST5224: Advanced Statistical Theory II

Every set of first-order formulas is equivalent to an independent set

Jesse Maassen and Mark Lundstrom Purdue University November 25, 2013

Phys460.nb Solution for the t-dependent Schrodinger s equation How did we find the solution? (not required)

EE512: Error Control Coding

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 24/3/2007

Continuous Distribution Arising from the Three Gap Theorem

SCHOOL OF MATHEMATICAL SCIENCES G11LMA Linear Mathematics Examination Solutions

Statistical Inference I Locally most powerful tests

Main source: "Discrete-time systems and computer control" by Α. ΣΚΟΔΡΑΣ ΨΗΦΙΑΚΟΣ ΕΛΕΓΧΟΣ ΔΙΑΛΕΞΗ 4 ΔΙΑΦΑΝΕΙΑ 1

2 Composition. Invertible Mappings

5.4 The Poisson Distribution.

Αλγόριθμοι και πολυπλοκότητα NP-Completeness (2)

ΚΥΠΡΙΑΚΟΣ ΣΥΝΔΕΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY 21 ος ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ Δεύτερος Γύρος - 30 Μαρτίου 2011

Congruence Classes of Invertible Matrices of Order 3 over F 2

Math 6 SL Probability Distributions Practice Test Mark Scheme

Homomorphism in Intuitionistic Fuzzy Automata

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 19/5/2007

Finite Field Problems: Solutions

Econ 2110: Fall 2008 Suggested Solutions to Problem Set 8 questions or comments to Dan Fetter 1

Abstract Storage Devices

ΕΙΣΑΓΩΓΗ ΣΤΗ ΣΤΑΤΙΣΤΙΚΗ ΑΝΑΛΥΣΗ

New bounds for spherical two-distance sets and equiangular lines

ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΥΠΡΟΥ ΤΜΗΜΑ ΠΛΗΡΟΦΟΡΙΚΗΣ. ΕΠΛ342: Βάσεις Δεδομένων. Χειμερινό Εξάμηνο Φροντιστήριο 10 ΛΥΣΕΙΣ. Επερωτήσεις SQL

Reminders: linear functions

CHAPTER 25 SOLVING EQUATIONS BY ITERATIVE METHODS

Cyclic or elementary abelian Covers of K 4

Πτυχιακή Εργασία Η ΠΟΙΟΤΗΤΑ ΖΩΗΣ ΤΩΝ ΑΣΘΕΝΩΝ ΜΕ ΣΤΗΘΑΓΧΗ

Matrices and vectors. Matrix and vector. a 11 a 12 a 1n a 21 a 22 a 2n A = b 1 b 2. b m. R m n, b = = ( a ij. a m1 a m2 a mn. def

Partial Differential Equations in Biology The boundary element method. March 26, 2013

Distances in Sierpiński Triangle Graphs

3.4 SUM AND DIFFERENCE FORMULAS. NOTE: cos(α+β) cos α + cos β cos(α-β) cos α -cos β

Practice Exam 2. Conceptual Questions. 1. State a Basic identity and then verify it. (a) Identity: Solution: One identity is csc(θ) = 1

Lecture 2: Dirac notation and a review of linear algebra Read Sakurai chapter 1, Baym chatper 3

The Simply Typed Lambda Calculus

ORDINAL ARITHMETIC JULIAN J. SCHLÖDER

Uniform Convergence of Fourier Series Michael Taylor

ANSWERSHEET (TOPIC = DIFFERENTIAL CALCULUS) COLLECTION #2. h 0 h h 0 h h 0 ( ) g k = g 0 + g 1 + g g 2009 =?

Models for Probabilistic Programs with an Adversary

Dynamic types, Lambda calculus machines Section and Practice Problems Apr 21 22, 2016

Statistics 104: Quantitative Methods for Economics Formula and Theorem Review

Second Order RLC Filters

Areas and Lengths in Polar Coordinates

Areas and Lengths in Polar Coordinates

Πρόβλημα 1: Αναζήτηση Ελάχιστης/Μέγιστης Τιμής

Assalamu `alaikum wr. wb.

SCITECH Volume 13, Issue 2 RESEARCH ORGANISATION Published online: March 29, 2018

CE 530 Molecular Simulation

ΚΥΠΡΙΑΚΗ ΕΤΑΙΡΕΙΑ ΠΛΗΡΟΦΟΡΙΚΗΣ CYPRUS COMPUTER SOCIETY ΠΑΓΚΥΠΡΙΟΣ ΜΑΘΗΤΙΚΟΣ ΔΙΑΓΩΝΙΣΜΟΣ ΠΛΗΡΟΦΟΡΙΚΗΣ 6/5/2006

Lecture 2. Soundness and completeness of propositional logic

Homework 3 Solutions

Example Sheet 3 Solutions

Chapter 6: Systems of Linear Differential. be continuous functions on the interval

The challenges of non-stable predicates

Μηχανική Μάθηση Hypothesis Testing

Approximation of distance between locations on earth given by latitude and longitude

Σχέσεις, Ιδιότητες, Κλειστότητες

CHAPTER 48 APPLICATIONS OF MATRICES AND DETERMINANTS

Homomorphism of Intuitionistic Fuzzy Groups

derivation of the Laplacian from rectangular to spherical coordinates

Lecture 34 Bootstrap confidence intervals

Bounding Nonsplitting Enumeration Degrees

Solutions to the Schrodinger equation atomic orbitals. Ψ 1 s Ψ 2 s Ψ 2 px Ψ 2 py Ψ 2 pz

Commutative Monoids in Intuitionistic Fuzzy Sets

On a four-dimensional hyperbolic manifold with finite volume

ΑΡΙΣΤΟΤΕΛΕΙΟ ΠΑΝΕΠΙΣΤΗΜΙΟ ΘΕΣΣΑΛΟΝΙΚΗΣ ΤΜΗΜΑ ΟΔΟΝΤΙΑΤΡΙΚΗΣ ΕΡΓΑΣΤΗΡΙΟ ΟΔΟΝΤΙΚΗΣ ΚΑΙ ΑΝΩΤΕΡΑΣ ΠΡΟΣΘΕΤΙΚΗΣ

4.6 Autoregressive Moving Average Model ARMA(1,1)

Right Rear Door. Let's now finish the door hinge saga with the right rear door

Supplementary Materials for Evolutionary Multiobjective Optimization Based Multimodal Optimization: Fitness Landscape Approximation and Peak Detection

A Note on Intuitionistic Fuzzy. Equivalence Relation

Second Order Partial Differential Equations

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

Problem Set 3: Solutions

C.S. 430 Assignment 6, Sample Solutions

ΤΕΧΝΟΛΟΓΙΚΟ ΠΑΝΕΠΙΣΤΗΜΙΟ ΚΥΠΡΟΥ ΣΧΟΛΗ ΓΕΩΠΟΝΙΚΩΝ ΕΠΙΣΤΗΜΩΝ ΒΙΟΤΕΧΝΟΛΟΓΙΑΣ ΚΑΙ ΕΠΙΣΤΗΜΗΣ ΤΡΟΦΙΜΩΝ. Πτυχιακή εργασία

VBA ΣΤΟ WORD. 1. Συχνά, όταν ήθελα να δώσω ένα φυλλάδιο εργασίας με ασκήσεις στους μαθητές έκανα το εξής: Version ΗΜΙΤΕΛΗΣ!!!!

The Probabilistic Method - Probabilistic Techniques. Lecture 7: The Janson Inequality

6.1. Dirac Equation. Hamiltonian. Dirac Eq.

Iterated trilinear fourier integrals with arbitrary symbols

Εργαστήριο Ανάπτυξης Εφαρμογών Βάσεων Δεδομένων. Εξάμηνο 7 ο

þÿ Ç»¹º ³µÃ ± : Ãż²» Ä Â

SOLUTIONS TO MATH38181 EXTREME VALUES AND FINANCIAL RISK EXAM

Matrices and Determinants

Section 8.3 Trigonometric Equations

Inverse trigonometric functions & General Solution of Trigonometric Equations

Τ.Ε.Ι. ΔΥΤΙΚΗΣ ΜΑΚΕΔΟΝΙΑΣ ΠΑΡΑΡΤΗΜΑ ΚΑΣΤΟΡΙΑΣ ΤΜΗΜΑ ΔΗΜΟΣΙΩΝ ΣΧΕΣΕΩΝ & ΕΠΙΚΟΙΝΩΝΙΑΣ

Solutions to Exercise Sheet 5

EPL 603 TOPICS IN SOFTWARE ENGINEERING. Lab 5: Component Adaptation Environment (COPE)

Heisenberg Uniqueness pairs

ΑΛΓΟΡΙΘΜΟΙ Άνοιξη I. ΜΗΛΗΣ

Statistics & Research methods. Athanasios Papaioannou University of Thessaly Dept. of PE & Sport Science

Higher Derivative Gravity Theories

ΑΓΓΛΙΚΑ Ι. Ενότητα 7α: Impact of the Internet on Economic Education. Ζωή Κανταρίδου Τμήμα Εφαρμοσμένης Πληροφορικής

ΙΠΛΩΜΑΤΙΚΗ ΕΡΓΑΣΙΑ. ΘΕΜΑ: «ιερεύνηση της σχέσης µεταξύ φωνηµικής επίγνωσης και ορθογραφικής δεξιότητας σε παιδιά προσχολικής ηλικίας»

Transcript:

Optimal Impartial Selection Max Klimm Technische Universität Berlin Head of Junior Research Group Optimization under Uncertainty Einstein-Zentrum für Mathematik

Introduction select member of a set of agents based on nominations by agents from the set applications Selection of representatives Award of prizes Assignment of responsibilities Peer review: of papers, projects Optimal Impartial Selection 2

Introduction A, B, C form committee vote for speaker, arcs represent votes p B = 0 B C p C = 1 A p A = 0 naive mechanism chose vertex with max. in-degree uniformly at random Optimal Impartial Selection 3

Introduction A, B, C form committee vote for speaker, arcs represent votes p B = 01/3 B C p C = 1/3 Vote of A influences its own probability of selection! A p A = 01/3 naive mechanism chose vertex with max. in-degree uniformly at random Optimal Impartial Selection 4

Impartial selection Goal: Find a mechanism that is impartial: probability of selecting an agent independent of its votes competitive: select (in expectation) agents with many nominations Optimal Impartial Selection 5

Formalization set G of directed graphs G = (N,E) without loops (approval voting) subset G =1 G of graphs with uniform out-degree 1 (plurality) mechanism f maps G = (N,E) G to a probability distribution over N f is impartial if f(n,e) i = f(n,f) i, whenever E \ ({i} N) = F \ ({i} N). f is α-optimal, for α [0,1], if for all G G if E i~f(g) [deg(i)] Δ(G) α, where Δ(G) = max i N deg(i). Optimal Impartial Selection 6

Related work Deterministic mechanisms are very limited, even for G =1 [Moulin, Holzman, Econometrica, `13] 1/4-optimal randomized mechanism (2-partition) Upper bound of 1/2 for any mechanism on G [Alon et al., TARK, `11] Achievable approximation factors G G =1 deterministic 0 1/n randomized [1/4, 1/2] [1/4, 1] Optimal Impartial Selection 7

Outline and main results Achievable approximation factors G G =1 deterministic 0 1/n randomized [1/4, 1/2] [1/4, 1] Optimal Impartial Selection 8

Outline and main results different analysis of the 2-partition mechanism 1/2-optimal mechanism for G Achievable approximation factors G G =1 deterministic 0 1/n randomized 1/2 [1/4, 1] Optimal Impartial Selection 9

Outline and main results different analysis of the 2-partition mechanism 1/2-optimal mechanism for G same mechanism is 67/108-optimal ( 0.62-optimal) for G =1 Achievable approximation factors G G =1 deterministic 0 1/n randomized 1/2 [67/108, 1] Optimal Impartial Selection 10

Outline and main results different analysis of the 2-partition mechanism 1/2-optimal mechanism for G same mechanism is 67/108-optimal ( 0.62-optimal) for G =1 upper bound for G =1 of 3/4 Achievable approximation factors G G =1 deterministic 0 1/n randomized 1/2 [67/108, 3/4] Optimal Impartial Selection 11

Outline and main results different analysis of the 2-partition mechanism 1/2-optimal mechanism for G same mechanism is 67/108-optimal ( 0.62-optimal) for G =1 upper bound for G =1 of 3/4 Achievable approximation factors G G =1 deterministic 0 1/n randomized 1/2 [67/108, 3/4] Optimal Impartial Selection 12

The 2-partition mechanism Partition each i N into (S 1, S 2 ) uniformly at random Return i arg max j S2 deg(j,s 1 ) uniformly at random (unless S 2 =, then return an arbitrary i S 1 ) deg(i,s 1 ) = (j,i) E : j S 1 Optimal Impartial Selection 13

The 2-partition mechanism Partition each i N into (S 1, S 2 ) uniformly at random Return i arg max j S2 deg(j,s 1 ) uniformly at random (unless S 2 =, then return an arbitrary i S 1 ) deg(i,s 1 ) = (j,i) E : j S 1 D E C A B Optimal Impartial Selection 14

The 2-partition mechanism Partition each i N into (S 1, S 2 ) uniformly at random Return i arg max j S2 deg(j,s 1 ) uniformly at random (unless S 2 =, then return an arbitrary i S 1 ) deg(i,s 1 ) = (j,i) E : j S 1 S 1 C S 2 E D A B Optimal Impartial Selection 15

The 2-partition mechanism Partition each i N into (S 1, S 2 ) uniformly at random Return i arg max j S2 deg(j,s 1 ) uniformly at random (unless S 2 =, then return an arbitrary i S 1 ) deg(i,s 1 ) = (j,i) E : j S 1 S 1 C S 2 E D A B Optimal Impartial Selection 16

The 2-partition mechanism Partition each i N into (S 1, S 2 ) uniformly at random Return i arg max j S2 deg(j,s 1 ) uniformly at random (unless S 2 =, then return an arbitrary i S 1 ) deg(i,s 1 ) = (j,i) E : j S 1 S 1 C 1/3 S 2 E D A 1/3 B 1/3 Optimal Impartial Selection 17

The 2-partition mechanism Theorem 2-Partition is impartial and 1/4-optimal. [Alon et al., TARK, `11] Proof: Impartiality: only nominations in S 1 considered, only agents in S 2 selected 1/4-Optimality: i* arg max i N deg(i) E i~f [deg(i)] P[i* S 2 ] E[deg(i*,S 1 ) i* S 2 ] E i~f [deg(i)] 1/2 E x~bin(1/2,δ) [x] =Δ/4 Bound is tight: Optimal Impartial Selection 18

The 2-partition mechanism - Revisited P[i* S 1 ] = 1/2 i* P[i* S 2 ] = 1/2 S 1 S 2 d* = deg(i*,s 1 ) d = max i S2 deg(i,s 1 ) for every 2-partition S = (S 1,S 2 ), E i~f [deg(i) S] d + (Δ-d) χ(d* > d)/2 E i~f [deg(i) S] min t N {t + (Δ-t) χ(d* > t)/2} minimum is attained for either t = 0, then E i~f [deg(i) S] = Δ/2 or t = d*, then E i~f [deg(i) S] = d* thus E i~f [deg(i)] E x~bin(1/2,δ) min{x, Δ/2} Δ/4 Optimal Impartial Selection 19

The 2-partition mechanism - Revisited Theorem [Fischer and K., SICOMP, `15] 2-Partition is α(δ)-optimal where α(δ) = Ex~Bin(1/2,Δ) min{x, Δ/2}. α(δ) is increasing lim Δ α(δ) = 1/2 1/2 α(δ) 3/8 1/4 1/8 0 0 2 4 6 8 10 Δ Optimal Impartial Selection 20

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i Optimal Impartial Selection 21

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i Optimal Impartial Selection 22

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i Optimal Impartial Selection 23

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i Optimal Impartial Selection 24

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i Optimal Impartial Selection 25

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 0 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 26

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 0 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 27

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 0 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 28

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 0 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 29

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 1 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 30

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 1 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 31

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 2 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 32

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 2 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 33

The k-partition mechanism Randomly partition N into (S 1,S 2,,S k ), denote S <j = i<j S i {i } :=, d := 0 for j:= 2,,k if max i Sj deg(i, S <j {i }) d Choose i arg max i Sj deg(i,s <j ) uniformly at random i := i, d = deg(i,s <j ) Return i i d = 2 S 1 S 2 S 3 S 4 S 5 Optimal Impartial Selection 34

The k-partition mechanism - Analysis P[i* S j ] = 1/k for all j = 1 k i* S 1 S 2 S 3 S 4 S 5 i* looses For k-partition S = (S 1, S k ) i* wins v j = deg(i*,s j ) and d j * = deg(i*,s <j ) = l [j-1] v l d = max j [k] max i Sj deg(i,s <j ) Then E i~f [deg(i) S] d + 1 k j [k](δ-d) χ(d j * > d) E i~f [deg(i) S] min t N {t + 1 k j [k](δ-t) χ(d j * > t)} minimum attained for t = d j * with j [k], and v = (v 1,,v k )~Mul(Δ,k) E i~f [deg(i)] E v~mul(δ,k) min j [k] { l=1,,j-1 v l + (k-j)/k l=j, k v l } Optimal Impartial Selection 35

The k-partition mechanism - Analysis Lemma k-partition is α(δ)-optimal, where α(δ) = E v~mul(δ,k) min j [k] { l=1,,j-1 v l + (k-j)/k l=j, k v l }. 0 1 α(δ) = E v~mul(δ,k) min v B @ k 1 k 1 k k k 2 1 k k 1 k 1 k... k k 2 k 2 k... k k 3 k... k 1 1 k 3......... 1 1... 1 0 C A Theorem [Fischer and K., SICOMP `15] k-partition is impartial and (k-1)/2k-optimal. α(δ) is non-decreasing in Δ α(1) E v~mul(1,k) min {v (k-1, k-2,,0)}/k (1/k 2 ) i=1,,k k-i = (k-1)/2k Optimal Impartial Selection 36

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i Optimal Impartial Selection 37

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 38

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 39

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 40

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 41

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 42

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 43

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 44

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 45

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 0 i Optimal Impartial Selection 46

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 1 i Optimal Impartial Selection 47

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 1 i Optimal Impartial Selection 48

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 2 i Optimal Impartial Selection 49

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 2 i Optimal Impartial Selection 50

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 2 i Optimal Impartial Selection 51

The permutation mechanism Choose permutation (π 1,π 2,,π n ) of N, denote π <j = i<j {π i } {i } :=, d := 0 for j:= 1,,k if deg(π j, π <j {i }) d i := π j, d = deg(π j,π <j ) return i d = 2 i Optimal Impartial Selection 52

The permutation algorithm Theorem [Fischer and K., SICOMP `15] Permutation is impartial and 1/2-optimal. Proof by standard limit arguments d = 2 i Optimal Impartial Selection 53

Highly nominated agents Theorem [Fischer and K., SICOMP `15] For graphs with sufficiently large Δ, permutation is (3/4-ε)-optimal. Proof For large Δ, v Mul(Δ,k): v i (1-ε)(Δ/k) whp 0 B B @ k 1 k 1 k k k 2 1 k k 1 k 1 k... k k 2 k 2 k... k k 3 k... 1 1 k 3......... 1 1... 1 0 With W = B k C, α(δ) = E v~mul(δ,k) min Wv (1-ε)(Δ/k) min W (1,..,1) T 3/4 - ε 1 C C A Δ Δ/2 Δ/2 Optimal Impartial Selection 54

Plurality Any G G =1 contains a directed cycle Permutation mechanism always selects a vertex with deg(i) 1 If Δ = 1, then α = 1 If Δ = 2, then α 2/3 Approximation guarantees parametrized in Δ give better bounds for Δ 3 67/108 for permutation (quite technical) Optimal Impartial Selection 55

Upper bound for plurality Optimal Impartial Selection 56

Upper bound for plurality p 1 p 1 p 1 p 1 p 1 p 1 np 1 = 1 Optimal Impartial Selection 57

Upper bound for plurality p 1 p 1 p 2 p 4 p 1 p 1 p 1 p 5 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 Optimal Impartial Selection 58

Upper bound for plurality p 1 p 1 p 2 p 4 p 2 p 1 p 1 p 1 p 5 p 2 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 Optimal Impartial Selection 59

Upper bound for plurality p 1 p 1 p 2 p 4 p 2 p 1 p 1 p 1 p 5 p 2 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 p 3 p 3 2p 2 +2p 3 1 Optimal Impartial Selection 60

Upper bound for plurality p 1 p 1 p 2 p 4 p 2 p 1 p 1 p 1 p 5 p 2 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 p 3 p 4 p 4 p 4 p 3 2p 2 +2p 3 1 p 4 4p 4 1 Optimal Impartial Selection 61

Upper bound for plurality p 1 p 1 p 2 p 4 p 2 p 1 p 1 p 1 p 5 p 2 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 p 3 p 6 p 6 p 4 p 4 p 5 p 5 p 4 p 3 2p 2 +2p 3 1 2p 5 +2p 6 1 p 4 4p 4 1 Optimal Impartial Selection 62

Upper bound for plurality p 1 p 1 p 2 p 4 p 2 p 1 p 1 p 1 p 5 p 2 p 1 p 1 np 1 = 1 p 3 p 5 p 1 +p 2 +p 3 +p 4 +(n 4)p 5 = 1 p 3 p 6 p 6 p 6 p 4 p 4 p 5 p 5 p 4 p 3 2p 2 +2p 3 1 2p 6 +(1-p 6 ) α 2 = p 6 /2+1/2 2p 5 +2p 6 1 p 4 4p 4 1 combining these inequalities: α 3/4-1/(8n) Optimal Impartial Selection 63

Selecting more than one agent Using permutation mechanism as a building block Selecting up to 2 agents [Bjelde, Fischer, K., WINE `15] 2/3-optimal while no impartial mechanism is better than 3/4-optimal Selecting exactly 2 agents 7/12-optimal while no impartial mechanism is better than 2/3-optimal Optimal Impartial Selection 64

Conclusions Impartial mechanisms select agents independently of their vote(s). Permutation mechanism is 1/2-optimal, i.e., it selects on average an agent with 1/2 of the maximal number of votes (subject to impartiality this is optimal) For highly nominated agents, it is 3/4-optimal. For plurality, it is 67/108-optimal and no impartial mechanism is better than 3/4-optimal. Multiple agents can be selected by running the permutation mechanism several times. Thank you. Optimal Impartial Selection 65