algorithmic coalitional game theory
play

Algorithmic Coalitional Game Theory Lecture 11: Coalition Structure - PowerPoint PPT Presentation

Algorithmic Coalitional Game Theory Lecture 11: Coalition Structure Generation Oskar Skibski University of Warsaw 12.05.2020 Coalition Structure Generation Coalition Structure Generation Find a partition of players = { " , ,


  1. Algorithmic Coalitional Game Theory Lecture 11: Coalition Structure Generation Oskar Skibski University of Warsaw 12.05.2020

  2. Coalition Structure Generation Coalition Structure Generation Find a partition of players 𝑄 = {𝑇 " , … , 𝑇 # } such that the sum of values of coalitions, i.e. 𝑀 𝑇 " + β‹― + 𝑀(𝑇 # ) , is maximized. In other words: which coalition structure will form? β€’ There are πΆπ‘“π‘šπ‘š(π‘œ) partitions. β€’ We need to check values of all coalitions, so perform at least 𝑃(2 ! ) steps. 2 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  3. Coalition Structure Generation Coalition Structure Generation Find a partition of players 𝑄 = {𝑇 " , … , 𝑇 # } such that the sum of values of coalitions, i.e. 𝑀 𝑇 " + β‹― + 𝑀(𝑇 # ) , is maximized. Notation: β€’ 𝒬(𝑂) – the set of all partitions of set 𝑂 β€’ 𝒬 # (𝑂) – the set of all partitions of set 𝑂 of size 𝑙 π‘œ 𝑙 – size of 𝒬 # 𝑂 , i.e., Stirling number of the second kind β€’ β€’ 𝑀 𝑄 – value of partition 𝑄 , i.e., βˆ‘ $∈& 𝑀(𝑇) 3 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  4. Coalition Structure Generation 𝒬 ' (𝑂) 1|2|3|4 𝒬 ( (𝑂) 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 124|3 134|2 1|234 12|34 13|24 14|23 𝒬 " (𝑂) 1234 4 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  5. Dynamic Programming (DP) [Yeh 1986] Input: Game 𝑂, 𝑀 Output: 𝑄 βˆ— ∈ 𝒬(𝑂) s.t. 𝑀 𝑄 βˆ— β‰₯ 𝑀 𝑄 for every 𝑄 ∈ 𝒬(𝑂) 1: for 𝑙 from 1 to π‘œ do 2: for each 𝑇 βŠ† 𝑂, 𝑇 = 𝑙 do 3: 𝑔 𝑇 ← 𝑀 𝑇 ; 𝑒 𝑇 ← {𝑇} ; 4: for each 𝐡, 𝐢 ∈ 𝒬 " (𝑇) do 5: if 𝑔 𝐡 + 𝑔(𝐢) > 𝑔 𝑇 then 𝑔 𝑇 ← 𝑔 𝐡 + 𝑔 𝐢 ; 6: 7: 𝑒 𝑇 ← 𝑒 𝐡 βˆͺ 𝑒 𝐢 ; 8: return 𝑒 𝑂 ; 5 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  6. Dynamic Programming (DP) [Yeh 1986] 𝑻 π’ˆ 𝑻 𝑻 π’ˆ 𝑻 {1} 𝑀 1 = 2 {1,2,3} 𝑔 1 + 𝑔 2,3 = 9 {2} 𝑀 2 = 4 {1,2,4} 𝑔 1 + 𝑔 2,4 = 11 {3} 𝑀 3 = 3 1,3,4 𝑀 1,3,4 = 10 {4} 𝑀 4 = 5 {2,3,4} 𝑔 2 + 𝑔 3,4 = 12 {1,2} 𝑔 1 + 𝑔 2 = 6 {1,2,3,4} 𝑔 1 + 𝑔 2,3,4 = 14 {1,3} 𝑀 1,3 = 5 {1,4} 𝑀 1,4 = 7 For 𝑀 𝑇 = βˆ‘ #∈% 𝑗 + βˆ‘ #∈% 𝑗 𝑛𝑝𝑒 3 {2,3} 𝑀 2,3 = 7 {2,4} 𝑔 2 + 𝑔 4 = 9 {3,4} 𝑀 3,4 = 8 6 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  7. Dynamic Programming (DP) [Yeh 1986] DP Complexity DP runs in time 𝒫 3 ! . Proof: For every coalition 𝑇 βŠ† 𝑂 , 𝑇 β‰  βˆ… , DP checks 2 $ S" βˆ’ 1 splits. So: ! π‘œ 2 $ S" βˆ’ 1 = < 2 WS" βˆ’ 1 = < 𝑑 $βŠ†T,$Uβˆ… WX" ! = 1 π‘œ 2 WS" βˆ’ 1 = 1 2 3 ! + 1 βˆ’ 2 ! . 2 + < 𝑑 WXY 7 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  8. Dynamic Programming (DP) [Yeh 1986] 𝒬 ' (𝑂) 1|2|3|4 OPTIMAL 𝒬 ( (𝑂) 12|3|4 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 123|4 124|3 124|3 134|2 1|234 12|34 12|34 13|24 14|23 𝒬 " (𝑂) 1234 8 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  9. Improved DP (IDP) [Rahwan & Jennings 2008] Main idea: β€žWe do not have to consider all splits as long as we can reach every coalition structure.” Which splits should we use? For 𝑇 β‰  𝑂 , we consider only splits 𝐡, 𝐢 ∈ 𝒬 ! (𝑇) such that: 𝐡 , 𝐢 ≀ 𝑂 βˆ’ 𝐡 βˆ’ 𝐢 . 9 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  10. Improved DP (IDP) [Rahwan & Jennings 2008] Input: Game 𝑂, 𝑀 Output: 𝑄 βˆ— ∈ 𝒬(𝑂) s.t. 𝑀 𝑄 βˆ— β‰₯ 𝑀 𝑄 for every 𝑄 ∈ 𝒬(𝑂) 1: for 𝑙 from 1 to π‘œ do 2: for each 𝑇 βŠ† 𝑂, 𝑇 = 𝑙 do 3: 𝑔 𝑇 ← 𝑀 𝑇 ; 𝑒 𝑇 ← {𝑇} ; 4: for each 𝐡, 𝐢 ∈ 𝒬 " 𝑇 s.t. |𝐡|, 𝐢 ≀ 𝑂 βˆ’ 𝑇 𝑝𝑠 𝑇 = 𝑂 do 5: if 𝑔 𝐡 + 𝑔(𝐢) > 𝑔 𝑇 then 𝑔 𝑇 ← 𝑔 𝐡 + 𝑔 𝐢 ; 6: 7: 𝑒 𝑇 ← 𝑒 𝐡 βˆͺ 𝑒 𝐢 ; 8: return 𝑒 𝑂 ; 10 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  11. Improved DP (IDP) [Rahwan & Jennings 2008] 𝒬 ' (𝑂) 1|2|3|4 𝒬 ( (𝑂) 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 124|3 134|2 1|234 12|34 13|24 14|23 𝒬 " (𝑂) 1234 11 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  12. Improved DP (IDP) [Rahwan & Jennings 2008] 𝒬 ' (𝑂) 1|2|3|4 𝒬 ( (𝑂) 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 124|3 134|2 1|234 12|34 13|24 14|23 𝒬 " (𝑂) 1234 We removed 12 arrows We removed 12 splits out of 31 arrows. out of 25 splits. 12 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  13. Improved DP (IDP) [Rahwan & Jennings 2008] IDP Correctness IDP is correct, i.e., finds the optimal coalition structure. Proof: It is enough to show that we can reach every coalition structure. Fix arbitrary 𝑄 = {𝑇 " , … , 𝑇 # } and assume 𝑇 " ≀ β‹― ≀ |𝑇 # | . If 𝑙 = 2 , then trivial. Assume 𝑙 > 2 . It is enough to show that 𝑄 has an incoming edge from the lower level. This is true, because for 𝑇 " βˆͺ 𝑇 ) we have: 𝑇 " ≀ 𝑇 ) ≀ 𝑂 βˆ’ 𝑇 " βˆ’ 𝑇 ) . 13 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  14. Improved DP (IDP) [Rahwan & Jennings 2008] IDP Complexity IDP runs in time 𝒫 3 ! . Proof: We can calculate the number of splits that we do not consider and show that it is 𝑝 3 ! . Rahwan and Jennings showed that for π‘œ = 25 , IDP consider only 38.7% splits of DP. 14 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  15. Optimal DP (ODP) [Michalak et al. 2016] Main idea: β€žInstead of looking at coalition sizes, we can take a lexicographical order” Let us write 𝐡 β‰Ό 𝐢 if 𝐡 is lexicographically smaller. E.g., 1,2,3 β‰Ό 1,4 β‰Ό {2,3,4,5} . For 𝑇 β‰  𝑂 , we consider only splits 𝐡, 𝐢 ∈ 𝒬 ! (𝑇) such that: 𝐡, 𝐢 β‰Ό 𝑂 βˆ– (𝐡 βˆͺ 𝐢). 15 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  16. Optimal DP (ODP) [Michalak et al. 2016] Input: Game 𝑂, 𝑀 Output: 𝑄 βˆ— ∈ 𝒬(𝑂) s.t. 𝑀 𝑄 βˆ— β‰₯ 𝑀 𝑄 for every 𝑄 ∈ 𝒬(𝑂) 1: for 𝑙 from 1 to π‘œ do 2: for each 𝑇 βŠ† 𝑂, 𝑇 = 𝑙 do 3: 𝑔 𝑇 ← 𝑀 𝑇 ; 𝑒 𝑇 ← {𝑇} ; 4: for each 𝐡, 𝐢 ∈ 𝒬 " 𝑇 s.t. 𝐡, 𝐢 β‰Ό 𝑂 βˆ– (𝐡 βˆͺ 𝐢) 𝑝𝑠 𝑇 = 𝑂 do 5: if 𝑔 𝐡 + 𝑔(𝐢) > 𝑔 𝑇 then 𝑔 𝑇 ← 𝑔 𝐡 + 𝑔 𝐢 ; 6: 7: 𝑒 𝑇 ← 𝑒 𝐡 βˆͺ 𝑒 𝐢 ; 8: return 𝑒 𝑂 ; 16 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  17. Optimal DP (ODP) [Michalak et al. 2016] 𝒬 ' (𝑂) 1|2|3|4 𝒬 ( (𝑂) 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 124|3 134|2 1|234 12|34 13|24 14|23 𝒬 " (𝑂) 1234 17 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  18. Optimal DP (ODP) [Michalak et al. 2016] 𝒬 ' (𝑂) 1|2|3|4 𝒬 ( (𝑂) 12|3|4 13|2|4 14|2|3 1|23|4 1|24|3 1|2|34 𝒬 ) (𝑂) 123|4 124|3 134|2 1|234 12|34 13|24 14|23 𝒬 " (𝑂) 1234 We removed 17 arrows out We removed 12 splits of 31 arrows. out of 25 splits. 18 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  19. Optimal DP (ODP) [Michalak et al. 2016] ODP Correctness There exists a unique path from 𝑂 to every coalition structure which implies ODP is correct. Proof: Fix arbitrary 𝑄 = {𝑇 " , … , 𝑇 # } and assume 𝑇 " β‰Ό β‹― β‰Ό 𝑇 # . If 𝑙 = 2 , then trivial. Assume 𝑙 > 2 . Now, 𝑄 has exactly one incoming edge from the lower level, specifically from {𝑇 " βˆͺ 𝑇 ) , 𝑇 ( , … , 𝑇 # } , because only for 𝑇 " βˆͺ 𝑇 ) we have: 𝑇 " β‰Ό 𝑇 ) β‰Ό 𝑂 βˆ– (𝑇 " βˆͺ 𝑇 ) ). 19 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

  20. Optimal DP (ODP) [Michalak et al. 2016] ODP Minimality 2 + π‘œ ODP performs π‘œ 3 = " ) 3 !S" βˆ’ 1 splits and less splits cannot be performed. Proof: Number π‘œ 2 corresponds to the number of splits of coalition 𝑂 . Number π‘œ 3 corresponds to splits of 𝐡 βˆͺ 𝐢 into {𝐡, 𝐢} such that 𝐡, 𝐢 β‰Ό 𝑂 βˆ– (𝐡 βˆͺ 𝐢) . It is easy to check that π‘œ ) (2 ! βˆ’ 2) and π‘œ 2 = " 3 = ` 3 ! βˆ’ 3 β‹… 2 ! + 3 = " ) (3 !S" βˆ’ 2 ! + 1) . " 20 Oskar Skibski (UW) Algorithmic Coalitional Game Theory

Recommend


More recommend