ranking and repulsing supermartingales for reachability
play

Ranking and Repulsing Supermartingales for Reachability in - PowerPoint PPT Presentation

Ranking and Repulsing Supermartingales for Reachability in Probabilistic Programs Toru Takisaka, Yuichiro Oyabu, Natsuki Urabe, Ichiro Hasuo A robot resolves a set of tasks Mode 1: safe mode N tasks Mode 1: safe mode 3 min. N-1 tasks N tasks


  1. Ranking and Repulsing Supermartingales for Reachability in Probabilistic Programs Toru Takisaka, Yuichiro Oyabu, Natsuki Urabe, Ichiro Hasuo

  2. A robot resolves a set of tasks

  3. Mode 1: safe mode N tasks

  4. Mode 1: safe mode 3 min. N-1 tasks N tasks

  5. Mode 2: urgent mode N tasks

  6. Mode 2: urgent mode 1 min. N-1 tasks N tasks 90 %

  7. Mode 2: urgent mode 1 min. N-1 tasks N tasks 90 % N+3 tasks 10 %

  8. Complete 15 tasks within 30 minutes

  9. Complete 15 tasks within 30 minutes What is the probability that the robot completes the tasks?

  10. Problem formulation Input: probabilistic program

  11. Problem formulation Input: probabilistic program Nondet. / Prob. branching

  12. Problem formulation Input: probabilistic program Nondet. / Prob. branching Nondet. / Prob. assignment

  13. Problem formulation Input: probabilistic program Problem What is the probability that the program terminates? (under angelic/demonic scheduler) Nondet. / Prob. branching We admit continuous variable Nondet. / Prob. ⇒ Generally one can’t compute assignment this value efficiently

  14. Problem formulation Input: probabilistic program Problem What is the probability that the program terminates? (under angelic/demonic scheduler) Nondet. / Prob. branching We admit continuous variable Nondet. / Prob. ⇒ Generally one can’t compute assignment this value efficiently ⇒ Certification by supermartingale

  15. Certification by supermartingale Probabilistic modification of real-world benchmarks (in Alias+, SAS’10) Almost-sure termination is certified in 20/28 examples (Agrawal+, POPL’18)

  16. Certification by supermartingale System: a pendulum under Gaussian noise The log-base-10 of the failure probability (failure = within 1h) >99% safety is guaranteed (Pr(enter a bad state) <1%) (Steinhardt-Tedrake, IJRR’12)

  17. Control flow graph � � � � Start � • A state is a pair (program location, memory state) • As powerful as MDP � finite

  18. Control flow graph � � � � Start � • A state is a pair (program location, memory state) • As powerful as MDP � finite

  19. Control flow graph � � � � Start � • A state is a pair (program location, memory state) • As powerful as MDP � finite

  20. Control flow graph � � � � Start � • A state is a pair (program location, memory state) • As powerful as MDP � finite

  21. Control flow graph � � � � Start � • A state is a pair (program location, memory state) • As powerful as MDP � finite

  22. Control flow graph � � � � Start Problem 𝟔 � (Locations) (Variables) ⇒ Pr(the system eventually • A state is a pair (program location, memory state) visits the region )? • As powerful as MDP � finite

  23. Supermartingale = a function over states that is “non-increasing” through transitions � (angelic) � � � …(demonic) � �

  24. Ranking function

  25. Ranking function

  26. Ranking function Int-valued

  27. Ranking function Int-valued The system eventually visits (under any nondeterministic choice)

  28. Ranking function Int-valued The system eventually visits (under any nondeterministic choice)

  29. Ranking supermartingale

  30. Ranking supermartingale

  31. Ranking supermartingale - valued decreases at least 1

  32. Ranking supermartingale - valued decreases at least 1 The system eventually visits almost surely

  33. Barrier certificate Safe region Unsafe region

  34. Barrier certificate Safe region Unsafe region

  35. Barrier certificate Safe region Unsafe region

  36. Barrier certificate Safe region Unsafe region

  37. Barrier certificate Safe region Unsafe region The system does not enter the unsafe region

  38. Probabilistic barrier certificate (a.k.a. nonnegative repulsing supermartingale) Safe region Unsafe region 𝑦 ����

  39. Probabilistic barrier certificate (a.k.a. nonnegative repulsing supermartingale) - Safe region valued Unsafe region ���� 𝑦 ����

  40. Probabilistic barrier certificate (a.k.a. nonnegative repulsing supermartingale) - Safe region valued Unsafe region ���� 𝑦 ����

  41. Probabilistic barrier certificate (a.k.a. nonnegative repulsing supermartingale) - Safe region valued Unsafe region ���� 𝑦 ����

  42. Probabilistic barrier certificate (a.k.a. nonnegative repulsing supermartingale) - Safe region valued Unsafe region ���� 𝑦 ���� Pr(the system enters the unsafe region)

  43. Our contributions Comprehensive account of martingale-based approximation methods via fixed point argument Soundness/completeness for uncountable-states MDP s, under angelic/demonic nondeterminism Implementation and experiments

  44. Our contributions Comprehensive account of martingale-based approximation methods via fixed point argument Soundness/completeness for uncountable-states MDP s, under angelic/demonic nondeterminism Implementation and experiments

  45. Two objective functions • Given: a control flow graph, and a subset of its states • and are

  46. Two objective functions • Given: a control flow graph, and a subset of its states • and are …under angelic/demonic scheduler

  47. Soundness/completeness Ranking supermartingale Soundness: ����� ( ) ���� Completeness: Nonnegative repulsing supermartingale Soundness: Completeness:

  48. Soundness/completeness Ranking supermartingale Soundness: Known ����� ( ) ���� Partly Completeness: known Nonnegative repulsing supermartingale Partly Soundness: known Not Completeness: known

  49. Soundness/completeness For certain endofunctions and and

  50. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � …

  51. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � … Soundness is a RankSM

  52. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � … Soundness is a RankSM

  53. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � … Knaster-Tarski theorem Soundness is a RankSM

  54. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � … Knaster-Tarski theorem Soundness is a RankSM Completeness

  55. Soundness/completeness Our theorem The lattice … the set of all (measurable) functions � … Knaster-Tarski theorem Soundness is a RepSM Completeness

  56. Our contributions Comprehensive account of martingale-based approximation methods via fixed point argument Soundness/completeness for uncountable-states MDP s, under angelic/demonic nondeterminism Implementation and experiments

  57. Soundness/completeness for martingale methods Approximation method It certifies Soundness Completeness Additive ranking Yes (MDP, Yes (MDP, Supermartingale continuous variable) discrete variable) ����� (Chakarov-Sankaranarayanan, CAV’13 etc.) Nonnegative repulsing Yes (Markov Chain) - supermartingale (Steinhardt+, IJRR’12 etc.) Yes (Markov Chain) - -scaled submartingale (Urabe+, LICS‘17) -decreasing repulsing Yes (MDP, - supermartingale continuous variable, (Chatterjee+, POPL’17) linearity assumpt.)

  58. Soundness/completeness for martingale methods Approximation method It certifies Soundness Completeness Yes (MDP, Additive ranking Yes (MDP, Yes (MDP, continuous Supermartingale continuous variable) discrete variable) ����� variable) (Chakarov-Sankaranarayanan, CAV’13 etc.) Nonnegative repulsing Yes (Markov Chain) - Yes (MDP, continuous variable) supermartingale (Steinhardt+, IJRR’12 etc.) Yes (MDP, Yes (Markov Chain) - -scaled submartingale continuous (Urabe+, LICS‘17) variable) -decreasing repulsing Yes (MDP, - No supermartingale continuous variable, (Chatterjee+, POPL’17) linearity assumpt.)

  59. Our contributions Comprehensive account of martingale-based approximation methods via fixed point argument Soundness/completeness for uncountable-states MDP s, under angelic/demonic nondeterminism Implementation and experiments

  60. Implementation and experiments ① ① ② • Implemented template-based synthesis algorithms • Nontrivial bounds are found ( ① ) • Observed comparative advantage of nonnegative RepSM over -decreasing RepSM ( ② )

  61. Summary • Martingale can evaluate reachability of probabilistic programs in various ways • We gave a comprehensive account of martingale-based approximation methods via fixed point argument • We proved soundness/completeness of several methods for uncountable-states MDPs , which extends known results • We demonstrated implementation and experiments

Recommend


More recommend