the impact of social ignorance on weighted congestion
play

The Impact of Social Ignorance on Weighted Congestion Games Vasilis - PowerPoint PPT Presentation

Introduction Model and Preliminaries Potential Function and Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of -Nash Dynamics The Impact of Social Ignorance on Weighted Congestion Games Vasilis Gkatzelis C.I.M.S. New


  1. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Congestion Games Rosenthal (1973) “A class of games possessing pure strategy Nash equilibria” A set V = { 1 , 2 , ..., n } of selfish players A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A non-decreasing delay function d e : N → R ≥ 0 for e ∈ R Linear congestion games, i.e. d e ( x ) = a e x + b e s = ( s 1 , s 2 , ..., s n ) ∈ Σ =X i ∈ V Σ i where s i ∈ Σ i for i ∈ V A congestion s e = |{ i : e ∈ s i }| for e ∈ R A cost c i ( s ) = � e ∈ s i d e ( s e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , c i ( s − i , s ′ i ) ≥ c i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  2. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Potential Games Monderer and Shapley (1996) “Potential Games” Exact potential function Φ( s ) : Σ → R such that Φ( s ) − Φ( s − i , s ′ i ) = c i ( s ) − c i ( s − i , s ′ i ) ∀ i ∈ V , s ′ i ∈ Σ i Nash dynamics converges to PNE (the directed state graph with payoff improving individual defections is a DAG) � s e For congestion games Φ( s ) = � k =1 d e ( k ) is exact e ∈ R Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  3. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Potential Games Monderer and Shapley (1996) “Potential Games” Exact potential function Φ( s ) : Σ → R such that Φ( s ) − Φ( s − i , s ′ i ) = c i ( s ) − c i ( s − i , s ′ i ) ∀ i ∈ V , s ′ i ∈ Σ i Nash dynamics converges to PNE (the directed state graph with payoff improving individual defections is a DAG) � s e For congestion games Φ( s ) = � k =1 d e ( k ) is exact e ∈ R Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  4. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Potential Games Monderer and Shapley (1996) “Potential Games” Exact potential function Φ( s ) : Σ → R such that Φ( s ) − Φ( s − i , s ′ i ) = c i ( s ) − c i ( s − i , s ′ i ) ∀ i ∈ V , s ′ i ∈ Σ i Nash dynamics converges to PNE (the directed state graph with payoff improving individual defections is a DAG) � s e For congestion games Φ( s ) = � k =1 d e ( k ) is exact e ∈ R Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  5. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  6. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  7. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  8. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  9. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  10. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Inefficiency of Pure Nash Equilibria Σ Social cost function: C ( s ) = � i ∈ V c i ( s ) OPT PNE Let configuration s opt ∈ arg min s ∈ Σ C ( s ) C ( s ) The (pure) Price of Anarchy (PoA): max s ∈ PNE C ( s opt ) C ( s ) The (pure) Price of Stability (PoS): min s ∈ PNE s opt Awerbuch et al. and Christodoulou et al. independently show: Unweighted linear congestion games have PoA=2.5 Awerbuch et al. also show: Weighted linear congestion games have PoA ≈ 2 . 62 Christodoulou et al. and Caragiannis et al. show: Unweighted linear congestion games have PoS ≈ 1 . 58 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  11. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Convergence Time of Dynamics (-) Ackermann et al. show: It is PLS-complete to compute Nash equilibria for congestion games with linear delay functions ( ∃ exponentially long paths) (+) Chien and Sinclair show: ǫ -Nash dynamics for symmetric congestion games converges to ǫ -Nash equilibria in a number of steps polynomial in | V | and ǫ − 1 (given a “ γ -bounded jump” condition for delays) (-) Voecking et al. show: The above approach cannot be extended to asymmetric congestion games (PLS-complete to approximate) (+) Awerbuch et al. show: ǫ -Nash dynamics for congestion games converges to almost optimal solutions w.r.t. social cost in a number of steps polynomial in | V | and ǫ − 1 (“ γ -bounded jump” condition) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  12. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Convergence Time of Dynamics (-) Ackermann et al. show: It is PLS-complete to compute Nash equilibria for congestion games with linear delay functions ( ∃ exponentially long paths) (+) Chien and Sinclair show: ǫ -Nash dynamics for symmetric congestion games converges to ǫ -Nash equilibria in a number of steps polynomial in | V | and ǫ − 1 (given a “ γ -bounded jump” condition for delays) (-) Voecking et al. show: The above approach cannot be extended to asymmetric congestion games (PLS-complete to approximate) (+) Awerbuch et al. show: ǫ -Nash dynamics for congestion games converges to almost optimal solutions w.r.t. social cost in a number of steps polynomial in | V | and ǫ − 1 (“ γ -bounded jump” condition) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  13. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Convergence Time of Dynamics (-) Ackermann et al. show: It is PLS-complete to compute Nash equilibria for congestion games with linear delay functions ( ∃ exponentially long paths) (+) Chien and Sinclair show: ǫ -Nash dynamics for symmetric congestion games converges to ǫ -Nash equilibria in a number of steps polynomial in | V | and ǫ − 1 (given a “ γ -bounded jump” condition for delays) (-) Voecking et al. show: The above approach cannot be extended to asymmetric congestion games (PLS-complete to approximate) (+) Awerbuch et al. show: ǫ -Nash dynamics for congestion games converges to almost optimal solutions w.r.t. social cost in a number of steps polynomial in | V | and ǫ − 1 (“ γ -bounded jump” condition) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  14. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Convergence Time of Dynamics (-) Ackermann et al. show: It is PLS-complete to compute Nash equilibria for congestion games with linear delay functions ( ∃ exponentially long paths) (+) Chien and Sinclair show: ǫ -Nash dynamics for symmetric congestion games converges to ǫ -Nash equilibria in a number of steps polynomial in | V | and ǫ − 1 (given a “ γ -bounded jump” condition for delays) (-) Voecking et al. show: The above approach cannot be extended to asymmetric congestion games (PLS-complete to approximate) (+) Awerbuch et al. show: ǫ -Nash dynamics for congestion games converges to almost optimal solutions w.r.t. social cost in a number of steps polynomial in | V | and ǫ − 1 (“ γ -bounded jump” condition) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  15. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  16. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  17. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  18. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  19. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  20. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  21. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  22. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Motivation: Most of the work is based on the full information setting In many typical applications players have partial information How does the social context affect inefficiency and convergence? Gairing, Monien and Tiemann: Bayesian approach and mostly parallel-link games Koutsoupias, Panagopoulou and Spirakis: Directed social graph (player knows neighbors’ weights and prob. distr. for others’) for two identical parallel links Karakostas, Kim, Viglas and Xia: Fraction of players is totally ignorant of the presence of others Ashlagi, Krysta and Tennenholtz: Social graph not modelling information but individual costs Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  23. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  24. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  25. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  26. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  27. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  28. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  29. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  30. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Ignorance in Congestion Games Bil` o et al. introduce “Graphical Congestion Games”: Players are vertices of a social graph G = ( V , E ) Each player has: full information about the players in his social neighborhood no information whatsoever about the remaining players They show that such games with linear delay functions and unweighted players playing based on their presumed costs: admit a potential function have PoS ≤| V | have PoA= Θ( | V | ( deg max ( G ) + 1)) where deg max ( G ) is the maximum degree of the graph In follow-up paper they show that one can dramatically decrease the PoA by enforcing a carefully selected social graph Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  31. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  32. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  33. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  34. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  35. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  36. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  37. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  38. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  39. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  40. Introduction Model and Preliminaries Motivation and Previous Work Potential Function and Cost Approximation Contribution Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Contribution Is there a more natural parameter of the social graph G that characterizes the inefficiency of PNE and the dynamics? We extend the graphical congestion games’ model for weighted players to investigate the impact of different weights We use the independence number α ( G ) and show that: These games admit a potential function For unweighted players: � 3 α ( G )+7 3 α ( G )+1 α 2 ( G ) if α ( G ) < n 2 , PoA ≤ if α ( G ) ≥ n 2 n ( n − α ( G ) + 1) 2 For weighted players: α ( G )( α ( G )+2+ √ α 2 ( G )+4 α ( G )) PoA ≤ < α ( G )( α ( G ) + 2) 2 α ( G ) ≤ PoS ≤ 2 α ( G ) The techniques of Chien et al. and Awerbuch et al. apply Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  41. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  42. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  43. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  44. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  45. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  46. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  47. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  48. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Linear Congestion Games A set V = { 1 , 2 , ..., n } of selfish players A positive integer weight w i for i ∈ V A set R of resources A strategy space Σ i ⊆ 2 R − {∅} for i ∈ V A delay function d e ( x ) = a e x + b e for e ∈ R A configuration s = ( s 1 , s 2 , ..., s n ) ∈ Σ A congestion s e = � i : e ∈ s i w i for e ∈ R A cost c i ( s ) = w i � e ∈ s i d e ( s e ) for i ∈ V Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  49. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  50. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  51. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  52. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  53. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  54. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Σ OPT PNE Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  55. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Weighted Graphical Congestion Games An undirected social graph G = ( V , E ) A neighborhood Γ i = { j ∈ V : { i , j } ∈ E } An actual congestion s e = � j ∈ V : e ∈ s j w j for e ∈ R A presumed congestion s i e = w i + � j ∈ Γ i : e ∈ s j w j for e ∈ R An actual cost c i ( s ) = � e ∈ s i w i ( a e s e + b e ) for i ∈ V e ∈ s i w i ( a e s i A presumed cost p i ( s ) = � e + b e ) for i ∈ V PNE= { s ∈ Σ : ∀ i ∈ V , ∀ s ′ i ∈ Σ i , p i ( s − i , s ′ i ) ≥ p i ( s ) } Σ PNE OPT Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  56. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example Social graph G = ( V , E ), resource set R and a configuration s . 2 3 4 3 2 1 2 2 1 1 4 2 3 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  57. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example Let e be some resource with d e ( x ) = a e x + b e . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  58. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example The set V e ( s ) of players using it induces subgraph G e . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  59. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example The actual congestion s e of resource e in s is 20 . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  60. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example This player’s presumed congestion for resource e was 14 . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  61. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example This player’s presumed congestion for resource e was 3 . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  62. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example C e ( s )= 20( a e 20 + b e ) and P e ( s )= a e 164 + 20 b e . 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  63. Introduction Model and Preliminaries Weighted Linear Congestion Games Potential Function and Cost Approximation Weighted Graphical Congestion Games Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Social Graph Example We will show that C e ( s ) ≤ α ( G e ) P e ( s ) for all e ∈ R and s ∈ Σ. 2 3 4 3 2 1 2 2 1 1 4 2 3 e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  64. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Potential Function Theorem Every graphical linear congestion game with weighted players admits a potential function and thus a pure Nash equilibrium. Φ( s ) = P ( s ) + U ( s ) 2 where U ( s ) = � n � e ∈ s i w i ( a e w i + b e ) is a potential function. i =1 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  65. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Potential Function Theorem Every graphical linear congestion game with weighted players admits a potential function and thus a pure Nash equilibrium. Φ( s ) = P ( s ) + U ( s ) 2 where U ( s ) = � n � e ∈ s i w i ( a e w i + b e ) is a potential function. i =1 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  66. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  67. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  68. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  69. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  70. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s )= � e ∈ R C e ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s )= � e ∈ R P e ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  71. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s )= � e ∈ R C e ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s )= � e ∈ R P e ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  72. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Total actual cost of s : C ( s ) = � i ∈ V c i ( s )= � e ∈ R C e ( s ) Total presumed cost of s : P ( s ) = � i ∈ V p i ( s )= � e ∈ R P e ( s ) Lemma For any configuration s , C ( s ) ≤ α ( G ) P ( s ) where α ( G ) is the independence number of the social graph G Proof sketch: For any resource e ∈ R let V e ( s ) = { i ∈ V : e ∈ s i } and G e = ( V e ( s ) , E e ( s )) be the graph induced by these players. It suffices to show C e ( s ) ≤ α ( G e ) P e ( s ) We first prove that this is true for the case of unweighted players. We then reduce the case of weighted players to the unweighted. Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  73. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  74. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  75. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  76. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  77. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  78. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  79. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  80. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  81. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  82. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  83. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  84. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  85. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  86. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . C e ( s ) = a e n 2 e + b e n e and P e ( s ) = a e (2 m e + n e ) + b e n e Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  87. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e n e Let n e = | V e ( s ) | , m e = | E e ( s ) | , k = α ( G e ) and r = k − ⌊ k ⌋ . We get that m e ≥ ( k − r )( k + r − 1) α ( G e ) / 2 Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  88. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e m e ≥ ( k − r )( k + r − 1) α ( G e ) / 2 ⇒ P e ( s ) ≥ a e [( k − r )( k + r + 1) + k ] α ( G e ) + b e k α ( G e ) ⇒ α ( G e ) P e ( s ) ≥ a e k 2 α 2 ( G e ) + b e k α ( G e ) = C e ( s ) For weighted players: We create a new graph G ′ e of unweighted players as follows: Replace each player i of weight w i with a clique Q i of size w i For { i , j } ∈ E e ( s ), connect all players of Q i and Q j i) C e ( s ) and P e ( s ) are not affected and ii) α ( G e ) = α ( G ′ e ) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

  89. Introduction Model and Preliminaries Potential Function Potential Function and Cost Approximation Cost Approximation Inefficiency of Pure Nash Equilibria Convergence of ǫ -Nash Dynamics Cost Approximation Proof For any configuration s and resource e , C e ( s ) ≤ α ( G e ) P e ( s ) where α ( G e ) is the independence number of G e m e ≥ ( k − r )( k + r − 1) α ( G e ) / 2 ⇒ P e ( s ) ≥ a e [( k − r )( k + r + 1) + k ] α ( G e ) + b e k α ( G e ) ⇒ α ( G e ) P e ( s ) ≥ a e k 2 α 2 ( G e ) + b e k α ( G e ) = C e ( s ) For weighted players: We create a new graph G ′ e of unweighted players as follows: Replace each player i of weight w i with a clique Q i of size w i For { i , j } ∈ E e ( s ), connect all players of Q i and Q j i) C e ( s ) and P e ( s ) are not affected and ii) α ( G e ) = α ( G ′ e ) Vasilis Gkatzelis Impact of Social Ignorance on Weighted Congestion Games

Recommend


More recommend