optimal dislocation with persistent errors in
play

Optimal dislocation with persistent errors in subquadratic time - PowerPoint PPT Presentation

Optimal dislocation with persistent errors in subquadratic time Barbara Geissmann, Stefano Leucci, Chih-Hung Liu, Paolo Penna ETH Zurich The Problem Sorting with erroneous comparisons The Problem Sorting with erroneous comparisons 3 4


  1. Optimal dislocation with persistent errors in subquadratic time Barbara Geissmann, Stefano Leucci, Chih-Hung Liu, Paolo Penna ETH Zurich

  2. The Problem Sorting with erroneous comparisons

  3. The Problem Sorting with erroneous comparisons 3 4 5 7 8 1 2 6

  4. The Problem Sorting with erroneous comparisons 1 2 3 4 5 6 7 8

  5. The Problem 3 4 5 7 8 1 2 6

  6. The Problem 4 > 2 3 4 5 7 8 1 2 6

  7. The Problem 4 < 5 3 4 5 7 8 1 2 6

  8. The Problem 5 < 2 3 4 5 7 8 1 2 6

  9. The Problem 5 < 2 3 4 5 7 8 1 2 6 • error probability p constant

  10. The Problem 5 < 2 3 4 5 7 8 1 2 6 • error probability p constant • independent for each pair

  11. The Problem 5 < 2 3 4 5 7 8 1 2 6 • error probability p constant • independent for each pair • persistent errors

  12. The Problem 5 < 2 3 4 5 7 8 1 2 6 • error probability p constant 5% • independent for each pair • persistent errors

  13. The Problem Repeating does not help 5 < 2 3 4 5 7 8 1 2 6 • error probability p constant • independent for each pair • persistent errors

  14. The Problem Repeating does not help 5 < 2 3 4 5 7 8 1 2 6

  15. Can you sort? Algorithm errors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

  16. Can you sort? Algorithm errors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16

  17. Can you sort? Algorithm errors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 11 12 10 13 14 15 16 Approx Sorted

  18. Can you sort? Algorithm errors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 11 12 10 13 14 15 16 Approx Sorted

  19. Can you sort? Algorithm errors 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 1 2 3 4 5 6 7 8 9 11 12 10 13 14 15 16 Dislocation

  20. What can be done?

  21. Prior Results

  22. Prior Results Dislocation TOTAL MAX O (log n ) O ( n ) Braverman & Mossel (SODA’08)

  23. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) Braverman & Mossel (SODA’08)

  24. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) 110525 (1 / 2 − p ) 4 Braverman & Mossel (SODA’08)

  25. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11)

  26. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  27. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) Ω (log n ) Ω ( n ) Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  28. Prior Results Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) Ω (log n ) Ω ( n ) Subquadratic time? Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  29. Our Contribution YES

  30. Our Contribution YES Dislocation TOTAL MAX Time: O ( n 3 / 2 ) O (log n ) O ( n )

  31. Our Contribution YES Dislocation TOTAL MAX Time: O ( n 3 / 2 ) O (log n ) O ( n ) randomized algorithm “derandomized” algorithm

  32. O ( n 2 )-Time Algorithm Window Sort Geissmann, Leucci, Liu, Penna (ISAAC’17)

  33. O ( n 2 )-Time Algorithm Window Sort errors well-spread ⇒ success Geissmann, Leucci, Liu, Penna (ISAAC’17)

  34. O ( n 2 )-Time Algorithm Window Sort errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  35. O ( n 2 )-Time Algorithm D x Window Sort x d errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  36. O ( n 2 )-Time Algorithm D x Window Sort x d with high prob d = O (log n ) errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  37. O ( n 2 )-Time Algorithm D x Window Sort n x d errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  38. O ( n 2 )-Time Algorithm New Algo D x Window Sort x d errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  39. O ( n 2 )-Time Algorithm New Algo D x tricky part Window Sort x d errors well-spread ⇒ success initial dislocation D ⇒ time O ( Dn ) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  40. Simple Faster Algo New Algo D x Window Sort x d

  41. Simple Faster Algo x D x Window Sort x d

  42. Simple Faster Algo x √ n D x Window Sort x d

  43. Simple Faster Algo x √ n D x Window Sort x d

  44. Simple Faster Algo x √ n D x Window Sort x d

  45. Simple Faster Algo x √ n O ( n ) O ( n ) O ( n ) D x Window Sort x d

  46. Simple Faster Algo x √ n O ( n 3 / 2 ) O ( n ) O ( n ) O ( n ) D x Window Sort x d

  47. Simple Faster Algo x √ n O ( n 3 / 2 ) NOT ENOUGH! D x Window Sort x d

  48. Simple Faster Algo x 1 √ n O ( n 3 / 2 ) NOT ENOUGH! D x Window Sort x d

  49. Simple Faster Algo x 1 √ n O ( n 3 / 2 ) 1 NOT ENOUGH! D x Window Sort x d

  50. Simple Faster Algo x 1 √ n O ( n 3 / 2 ) 1 NOT ENOUGH! D x 1 Window Sort x d

  51. Simple Faster Algo x O ( n 3 / 2 ) D x Window Sort x d

  52. Simple Faster Algo x O ( n 3 / 2 ) D x Window Sort x d

  53. Simple Faster Algo x O ( n 3 / 2 ) D x Window Sort x d

  54. Simple Faster Algo x O ( n 3 / 2 ) STILL NOT ENOUGH D x Window Sort x d

  55. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) D x Window Sort x d

  56. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) 1 2 D x Window Sort x d

  57. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) 1 2 D x Window Sort x d

  58. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) √ n 1 2 D x Window Sort x d

  59. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) √ n 1 2 1 D x Window Sort x d

  60. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) √ n 1 2 1 2 D x Window Sort x d

  61. Simple Faster Algo x 1 2 · · · n O ( n 3 / 2 ) √ n 1 2 √ n 1 2 D x Window Sort x d

  62. Simple Faster Algo random permutation x O ( n 3 / 2 ) D x Window Sort x d

  63. Simple Faster Algo random permutation x O ( n 3 / 2 ) = n 3 / 4 D x Window Sort x d

  64. Simple Faster Algo random permutation x O ( n 3 / 2 ) = n 3 / 4 D x O ( n 7 / 4 ) Window Sort x d

  65. That was the simple version...

  66. Window Sort Window Sort Window Sort Window Sort Window Sort O ( n 2 ) O ( n 2 − δ )

  67. Window Sort Window Sort Window Sort Window Sort Window Sort O ( n 2 ) O ( n 2 − δ ) Window Sort Window Sort Window Sort Window Sort O ( n 2 − δ )

  68. Window Sort Window Sort Window Sort Window Sort Window Sort O ( n 2 ) O ( n 2 − δ ) Window Sort Window Sort Window Sort Window Sort O ( n 2 − δ ) Window Sort

  69. Window Sort Window Sort Window Sort Window Sort Window Sort O ( n 2 ) O ( n 2 − δ ) Window Sort Window Sort Window Sort Window Sort O ( n 2 − δ ) Window Sort · · · O ( n 3 / 2 )

  70. Part II: Derandomization

  71. Comparisons ⇒ Randomness < > > < p p 1 − p 1 − p ⇒ < > · · · ? ? ? XOR 1 2 ± 1 1 2 ± 1 n 4 n 4 c log n Comparisons One random bit

  72. Naive Approach · · · k ? ? ? k ( n − k ) log n

  73. Naive Approach · · · k ? ? ? SORT k ( n − k ) log n

  74. Naive Approach · · · k ? ? ? SORT k ( n − k ) log n REINSERT

  75. Open Questions Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) O ( n 3 / 2 ) O (log n ) O ( n ) Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  76. Open Questions Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) ∼ O ( n 3 / 2 ) O (log n ) O ( n ) Faster? O ( n log n ) comparisons Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  77. Open Questions Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) p < 1 / 16 O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) O ( n 3 / 2 ) O (log n ) O ( n ) Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

  78. Open Questions Dislocation TOTAL MAX Time: O ( n 3+ C ) O (log n ) O ( n ) p < 1 / 16 O ( n 2 ) O (log n ) O ( n 2 ) O (log n ) O ( n ) O ( n 3 / 2 ) O (log n ) O ( n ) Any p < 1 / 2? Braverman & Mossel (SODA’08) Klein, Penninger, Sohler, Woodruff (ESA’11) Geissmann, Leucci, Liu, Penna (ISAAC’17)

Recommend


More recommend