you say it s only constant then something must be
play

You say its only constant? Then something must be hyperfinite! And - PowerPoint PPT Presentation

You say its only constant? Then something must be hyperfinite! And I say your title is too long. Hendrik Fichtenberger July 20, 2019 Property Testing in a Nutshell planar 1 Property Testing in a Nutshell planar non-planar 1 Property


  1. You say itโ€™s only constant? Then something must be hyperfinite! And I say your title is too long. Hendrik Fichtenberger July 20, 2019

  2. Property Testing in a Nutshell planar 1

  3. Property Testing in a Nutshell planar non-planar 1

  4. Property Testing in a Nutshell planar non-planar non-planar 1

  5. Property Testing in a Nutshell planar non-planar non-planar non-planar 1

  6. Property Testing in a Nutshell planar non-planar non-planar non-planar time complexity: ฮฉ(|๐‘Š|) 1

  7. Property Testing in a Nutshell planar non-planar non-planar non-planar time complexity: ฮฉ(|๐‘Š|) 1

  8. Property Testing in a Nutshell planar non-planar non-planar non-planar time complexity: ฮฉ(|๐‘Š|) very quite slightly 1

  9. Property Testing in a Nutshell planar non-planar non-planar non-planar time complexity: ฮฉ(|๐‘Š|) very quite slightly 1

  10. Property Testing in a Nutshell planar non-planar non-planar non-planar ๐œ— 1 0 # edge edits |๐‘Š|+|๐น| 1

  11. Property Testing in a Nutshell planar non-planar non-planar non-planar 3 3 ๐œ— 1 0 # edge edits |๐‘Š|+|๐น| 1 accept w.p. > 2 reject w.p. > 2

  12. Property Testing in a Nutshell 1 ๐œ— -far ๐œ— -far ๐œ— -close |๐‘Š|+|๐น| # edge edits 0 ๐œ— planar 3 3 non-planar non-planar non-planar 1 accept w.p. > 2 reject w.p. > 2

  13. Property Testing in a Nutshell 1 complexity: # queries to adjacency list entries ๐œ— -far ๐œ— -far ๐œ— -close |๐‘Š|+|๐น| # edge edits 0 ๐œ— planar 3 3 non-planar non-planar non-planar 1 accept w.p. > 2 reject w.p. > 2

  14. Property Testing of Bounded Degree Graphs bounded degree graphs: โˆ€๐‘ค โˆˆ ๐‘Š โˆถ ๐‘’(๐‘ค) โ‰ค ๐‘’, ๐‘’ โˆˆ ๐‘ƒ(1) ๐‘Ÿ(๐œ—) planar 2

  15. Property Testing of Bounded Degree Graphs bounded degree graphs: โˆ€๐‘ค โˆˆ ๐‘Š โˆถ ๐‘’(๐‘ค) โ‰ค ๐‘’, ๐‘’ โˆˆ ๐‘ƒ(1) ๐‘Ÿ(๐œ—) planar, degree-regular, cycle-free, subgraph-free, connected, minor-free, hyperfinite, ... 2

  16. Property Testing of Bounded Degree Graphs bounded degree graphs: โˆ€๐‘ค โˆˆ ๐‘Š โˆถ ๐‘’(๐‘ค) โ‰ค ๐‘’, ๐‘’ โˆˆ ๐‘ƒ(1) ๐‘Ÿ(๐œ—) planar, degree-regular, cycle-free, subgraph-free, connected, minor-free, hyperfinite, ... bipartite, expander ฮ˜(โˆš๐‘œ) 2

  17. Property Testing of Bounded Degree Graphs bounded degree graphs: โˆ€๐‘ค โˆˆ ๐‘Š โˆถ ๐‘’(๐‘ค) โ‰ค ๐‘’, ๐‘’ โˆˆ ๐‘ƒ(1) ๐‘Ÿ(๐œ—) planar, degree-regular, cycle-free, subgraph-free, connected, minor-free, hyperfinite, ... bipartite, expander ฮ˜(โˆš๐‘œ) big picture? 2

  18. Property Testing of Bounded Degree Graphs bounded degree graphs: โˆ€๐‘ค โˆˆ ๐‘Š โˆถ ๐‘’(๐‘ค) โ‰ค ๐‘’, ๐‘’ โˆˆ ๐‘ƒ(1) ๐‘Ÿ(๐œ—) planar, degree-regular, cycle-free, subgraph-free, connected, minor-free, hyperfinite, ... bipartite, expander ฮ˜(โˆš๐‘œ) big picture? bounded-degree graphs: litule known about constant-time testability 2

  19. ๐‘™ -Disks and Frequency Vectors disk ๐‘™ (๐‘ค) : subgraph induced by BFS (๐‘ค) of depth ๐‘™ 3

  20. ๐‘™ -Disks and Frequency Vectors disk 1 ( ) disk ๐‘™ (๐‘ค) : subgraph induced by BFS (๐‘ค) of depth ๐‘™ 3

  21. ๐‘™ -Disks and Frequency Vectors disk 2 ( ) disk 1 ( ) disk ๐‘™ (๐‘ค) : subgraph induced by BFS (๐‘ค) of depth ๐‘™ 3

  22. ๐‘™ -Disks and Frequency Vectors disk 2 ( 1 โˆ‘ ) โ‹ฎ 0.6 0.4 freq 2 ( type calculate its share of vertices freq ๐‘™ (๐ป) : for each ๐‘™ -disk isomorphism by BFS (๐‘ค) of depth ๐‘™ disk ๐‘™ (๐‘ค) : subgraph induced ) disk 1 ( ) 3 ) = (

  23. ๐‘™ -Disks and Frequency Vectors 0.4 locality feature frequency vector is a 1 โˆ‘ ) โ‹ฎ 0.6 3 disk 2 ( freq 2 ( type calculate its share of vertices freq ๐‘™ (๐ป) : for each ๐‘™ -disk isomorphism by BFS (๐‘ค) of depth ๐‘™ disk ๐‘™ (๐‘ค) : subgraph induced ) disk 1 ( ) ) = (

  24. Constant-Qvery Testers accept Goldreich, Ron, STOCโ€™09 1 2. accepts ifg โ€–ฬƒ freq ฮ˜(๐‘Ÿ) (๐ป) of freq ฮ˜(๐‘Ÿ) (๐ป) can be transformed into an algorithm that Every property tester with constant query complexity ๐‘Ÿ โˆถ= ๐‘Ÿ(๐œ—) Theorem [GRโ€™09, โ€ฆ] ฮ  ฮ  ฮ  ฮ  ฮ  1 1 reject 4 1. computes an approximation ฬƒ ฮ˜(๐‘Ÿ) for any ๐ป โ€ฒ โˆˆ ฮ  freq ฮ˜(๐‘Ÿ) (๐ป) โˆ’ freq ฮ˜(๐‘Ÿ) (๐ป โ€ฒ )โ€– 1 โ‰ค

  25. Hyperfinite Graphs Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5 In every planar graph, there exists a set of โˆš๐‘œ separators

  26. Hyperfinite Graphs Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5 In every planar graph, there exists a set of โˆš๐‘œ separators

  27. Hyperfinite Graphs 3 |๐‘Š| 3 |๐‘Š| Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5 In every planar graph, there exists a set of โˆš๐‘œ separators > 1 > 1

  28. Hyperfinite Graphs remove ๐œ—๐‘œ edges components of size ๐œ— โˆ’2 Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5 In every planar graph, there exists a set of โˆš๐‘œ separators

  29. Hyperfinite Graphs ๐œ(0.5) ๐œ(0.5) ๐œ(0.5) ๐œ— = 0.5 Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5

  30. Hyperfinite Graphs ๐œ— = 0.5 ๐œ(0.5) ๐œ(0.5) ๐œ(0.5) Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5

  31. Hyperfinite Graphs ๐œ— = 0.5 ๐œ— = 0.6 ๐œ(0.6) ๐œ(0.6) ๐œ(0.6) Definition (๐‘, ๐ญ) -hyperfinite: can remove at most ๐œ—๐‘’๐‘œ edges to obtain connected components of size at most ๐‘ก ๐‡ -hyperfinite: (๐œ—, ๐œ(๐œ—)) -hyperfinite for all ๐œ— โˆˆ (0, 1] 5

  32. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6

  33. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite [GRโ€™09] Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6

  34. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite [GRโ€™09] [NSโ€™11] Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6

  35. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite [GRโ€™09] [NSโ€™11] Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6

  36. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite [GRโ€™09] [NSโ€™11] e.g. ฮ  = connectivity contains expanders Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6

  37. The Story so Far ฮ  has constant query complexity ฮ  is characterized by its k-disk distribution ฮ  is ๐œ -hyperfinite [GRโ€™09] [NSโ€™11] e.g. ฮ  = connectivity contains expanders [FPSโ€™19] Newman, Sohler, STOCโ€™11; Fichtenberger, Peng, Sohler, SODAโ€™19 6 โˆƒ ฮ  โ€ฒ โŠ† ฮ  : ฮ  โ€ฒ is ๐œ โ€ฒ -hyperfinite

  38. Small Frequency-Preserver Graphs freq ๐‘™ (๐ป) freq ๐‘™ (๐ผ) ๐œ€ ๐ป ๐ผ ๐‘œ ๐‘(๐œ€, ๐‘™) Theorem [Alonโ€™11] For every ๐œ€, ๐‘™ > 0 , there exists ๐‘(๐œ€, ๐‘™) such that for every ๐ป there Alon, โ€™11, see: Lovรกsz, Large Networks and Graph Limits , Proposition 19.10 7 exists ๐ผ of size at most ๐‘(๐œ€, ๐‘™) and โ€– freq ๐‘™ (๐ป) โˆ’ freq ๐‘™ (๐ผ)โ€– 1 < ๐œ€ .

  39. Small Frequency-Preserver Graphs freq ๐‘™ (๐ป) freq ๐‘™ (๐ผ) ๐œ€ ๐ป ๐ผ ๐‘œ ๐‘(๐œ€, ๐‘™) For ฯ-hyperfinite graphs M(ฮด,k) is at most Theorem [Alonโ€™11] For every ๐œ€, ๐‘™ > 0 , there exists ๐‘(๐œ€, ๐‘™) such that for every ๐ป there Alon, โ€™11, see: Lovรกsz, Large Networks and Graph Limits , Proposition 19.10 7 exists ๐ผ of size at most ๐‘(๐œ€, ๐‘™) and โ€– freq ๐‘™ (๐ป) โˆ’ freq ๐‘™ (๐ผ)โ€– 1 < ๐œ€ .

  40. Putuing Everything Into The Property Testing Blender in ฮ  ๐‘œ rel. ๐šธ size hypf. ? freq. v. change original start w/ ๐ป โˆˆ ฮ  8

  41. Putuing Everything Into The Property Testing Blender original < ๐œ€/2 ? ? freq. pres. ๐‘ƒ(1) start w/ ๐ป โˆˆ ฮ  change in ฮ  freq. v. ? hypf. size rel. ๐šธ ๐‘œ 8

  42. Putuing Everything Into The Property Testing Blender freq. pres. = 0 ๐œ— -close to ฮ  blow-up hyperfinite (0, ๐‘(๐œ€, ๐‘™)) - ๐‘œ < ๐œ€/2 ? ? ๐‘ƒ(1) in ฮ  start w/ ๐ป โˆˆ ฮ  original change freq. v. ? hypf. size rel. ๐šธ ๐‘œ 8

Recommend


More recommend