Coloring Distributed Algorithms Below the Greedy Regime Yannic Maus Mohsen Ghaffari , Juho Hirvonen, Fabian Kuhn, Jara Uitto This project has received funding from the European Unionβs Horizon 2020 Research and Innovation Programme under grant agreement no. 755839.
LOCAL Model [Linial ; FOCS β87] π― = πΎ, π , Communication Network = Problem Instance: π = πΎ I am green 11 11 11 15 15 15 5 5 5 6 6 6 1 1 1 2 2 2 9 9 9 21 21 21 7 7 7 26 26 26 33 33 33 8 8 8 27 27 27 Discrete synchronous rounds: β’ local computations β’ exchange messages with all neighbors (computations unbounded, message sizes are unbounded) time complexity = number of rounds 3
CONGEST MODEL π― = πΎ, π , Communication Network = Problem Instance: π = πΎ 11 11 15 15 5 5 6 6 1 1 2 2 9 9 21 21 7 7 26 26 33 33 8 8 27 27 Discrete synchronous rounds: β’ local computations β’ exchange messages with all neighbors π·(π¦π©π‘ π) bits (computations unbounded, message sizes are unbounded) ) time complexity = number of rounds 4
Classic Big Four (Greedy Regime) π¬ + π -Vertex Coloring Maximal Ind. Set (MIS) (ππ¬ β π) -Edge Coloring Maximal Matching ( Ξ : maximum degree of π» ) 5
In the LOCAL Model β¦ Greedy Below Greedy 2 π log π 2 π log π Maximal IS Maximum IS, π β π -approx. poly log π vertex cover, 2 π log π vertex cover, π + π -approx. 2 -approx. 2 π log π 2 π log π min. dominating set, min dominating set, ( π + π) π¦π©π‘ π¬ -approx. π + π -approx. 2 π log π 2 π log π hypergraph vertex cover, hypergraph vertex cover, π + π -approx. rank -approx. π¬ + π -vertex coloring π¬ -vertex coloring 2 π log π 2 π log π ππ¬ β π -edge coloring poly log π poly log π π + π π¬ -edge coloring poly log π poly log π maximal matching Maximum Matching, π + π -approx. CONGEST 6
In the LOCAL Model β¦ Greedy Below Greedy 2 π log π 2 π log π Maximal IS Maximum IS, π β π -approx. βProblems that do have β Problems that do not have easy sequential greedy easy sequential greedy algorithms .β algorithms .β π¬ + π -vertex coloring π¬ -vertex coloring 2 π log π 2 π log π ππ¬ β π -edge coloring poly log π poly log π π + π π¬ -edge coloring poly log π poly log π maximal matching Maximum Matching, π + π -approx. CONGEST 7
Outline This Talk: How do we use LOCAL? Below Greedy Ξ©(π 2 ) 2 π log π Maximum IS Technique 1: Ball growing Maximum IS, π β π -approx. 7/8 -approx. [Ghaffari, Kuhn, Maus; STOC β 17 ] Ξ©(π 2 ) vertex cover, 2 π log π vertex cover, π + π -approx. exact Ξ©(π 2 ) 2 π log π min. dominating set, min dominating set, π + π -approx. exact Ξ©(π 2 ) 2 π log π hypergraph vertex cover, hypergraph vertex cover, π + π -approx. exact Ξ©(π 2 ) π(π―) -vertex coloring π¬ -vertex coloring Technique 2: Local filling 2 π log π [Ghaffari, Hirvonen, Kuhn, Maus; PODC β 18 ] ? poly log π π + π π¬ -edge coloring edge coloring Technique 3: Aug. paths ? poly log π [Ghaffari, Kuhn, Maus , Uitto; STOC β 18 ] Maximum matching Maximum Matching, π β π -approx. exact CONGEST 8
Technique 1: Ball Growing This Talk: How do we use LOCAL? Below Greedy Ξ©(π 2 ) 2 π log π Maximum IS Technique 1: Ball growing Maximum IS, π β π -approx. 7/8 -approx. Ξ©(π 2 ) vertex cover, 2 π log π vertex cover, π + π -approx. exact Ξ©(π 2 ) 2 π log π min. dominating set, min dominating set, π + π -approx. exact Ξ©(π 2 ) 2 π log π hypergraph vertex cover, hypergraph vertex cover, π + π -approx. exact Ξ©(π 2 ) π(π―) -vertex coloring π¬ -vertex coloring Technique 2: Local filling 2 π log π ? poly log π π + π π¬ -edge coloring edge coloring Technique 3: Aug. paths ? poly log π Maximum matching Maximum Matching, π β π -approx. exact 9
Sequential Ball Growing (MaxIS) π Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 10
Sequential Ball Growing (MaxIS) Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 11
Sequential Ball Growing (MaxIS) Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 12
Sequential Ball Growing (MaxIS) Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 13
Sequential Ball Growing (MaxIS) Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 14
Sequential Ball Growing (MaxIS) Sequentially computes a 1 + π β1 -approximation for MaxIS. (using unbounded computation) Safe Ball πͺ π (π) : |πππ¦π½π(πΆ π +1 )| < (1 + π) β πππ¦π½π(πΆ π ) Find safe ball: Set π = 0 and increase π until ball πΆ π is safe. Terminates with small radius π = π(π β1 log π) . 15
Parallel Ball Growing Theorem Using ( πͺπ©π¦π³ π¦π©π‘ π , πͺπ©π¦π³ π¦π©π‘ π) -network decompositions βsequentially ball growingβ can be βdone in parallelβ in LOCAL. [ STOC β 17, Ghaffari, Kuhn, Maus ] Corollary π¦π©π‘ π deterministic There are πͺπ©π¦π³ π¦π©π‘ π randomized and π π· π + π -approximation algorithms for covering and packing integer linear programs . This includes maximum independent set, minimum dominating set, vertex cover, β¦ . [ STOC β 17, Ghaffari, Kuhn, Maus ] 16
Technique 2: Local Filling This Talk: How do we use LOCAL? Below Greedy Ξ©(π 2 ) 2 π log π Maximum IS Technique 1: Ball growing Maximum IS, π β π -approx. 7/8 -approx. Ξ©(π 2 ) vertex cover, 2 π log π vertex cover, π + π -approx. exact Ξ©(π 2 ) 2 π log π min. dominating set, min dominating set, π + π -approx. exact Ξ©(π 2 ) 2 π log π hypergraph vertex cover, hypergraph vertex cover, π + π -approx. exact Ξ©(π 2 ) π(π―) -vertex coloring π¬ -vertex coloring Technique 2: Local filling 2 π log π ? poly log π π + π π¬ -edge coloring edge coloring Technique 3: Aug. paths ? poly log π Maximum matching Maximum Matching, π β π -approx. exact 17
Ξ -Coloring Previous Work: [Panconesi, Srinivasan; STOC β93 ] Definition: An induced subgraph πΌ β π» is called an easy component if any π¦ π» -coloring of π» β πΌ can be extended to a π¦ π» -coloring of π» without changing the coloring on π» β πΌ . Well studied under the name degree chosable components. [ ErdΕs et al. β79, Vizing β76 ] β β Theorem: Let π» be a graph ( β clique) with max. degree Ξ β₯ 3 . Every node of π» has a small diameter easy component in distance at most O(log n) . [ PODC β 18; Ghaffari, Hirvonen, Kuhn, Maus ] 18
Find an MIS π΅ of small diameter easy components Define π log π Layers: π΄ π = π€ π€ in distance π to some component in π΅} For π = π(log π) to 1 color nodes in π π through solving a (deg+1)-list coloring Color easy components in M 19
Find an MIS π΅ of small diameter easy components Define π log π Layers: π΄ π = π€ π€ in distance π to some component in π΅} For π = π(log π) to 1 color nodes in π π through solving a (deg+1)-list coloring Color easy components in M 20
Find an MIS π΅ of small diameter easy components Greedy regime Define π log π Layers: π΄ π = π€ π€ in distance π to some component in π΅} For π = π(log π) to 1 color nodes in π π through solving a (deg+1)-list coloring Color easy components in M 21
Find an MIS π΅ of small diameter easy components Greedy regime Define π log π Layers: π΄ π = π€ π€ in distance π to some component in π΅} For π = π(log π) to 1 color nodes in π π through solving a (deg+1)-list coloring Color easy components in M 22
Find an MIS π΅ of small diameter easy components Greedy regime Define π log π Layers: π΄ π = π€ π€ in distance π to some component in π΅} For π = π(log π) to 1 color nodes in π π through solving a (deg+1)-list coloring Color easy components in M 23
Recommend
More recommend