Convex Language Semantics for Nondeterministic Probabilistic - - PowerPoint PPT Presentation

convex language semantics for nondeterministic
SMART_READER_LITE
LIVE PREVIEW

Convex Language Semantics for Nondeterministic Probabilistic - - PowerPoint PPT Presentation

Convex Language Semantics for Nondeterministic Probabilistic Automata Gerco van Heerdt 1 Justin Hsu 1,2,3 el Ouaknine 4 Alexandra Silva 1 Jo 1 University College London 2 Cornell University 3 University of WisconsinMadison 4 MPI-SWS and


slide-1
SLIDE 1

Convex Language Semantics for Nondeterministic Probabilistic Automata

Gerco van Heerdt1 Justin Hsu1,2,3 Jo¨ el Ouaknine4 Alexandra Silva1

1University College London 2Cornell University 3University of Wisconsin–Madison 4MPI-SWS and Oxford University

October 19, 2018

1 / 17

slide-2
SLIDE 2

Deterministic probabilistic automaton

will get home

taste

1 2 1 2 sleep

taste

2 / 17

slide-3
SLIDE 3

Deterministic probabilistic automaton

1

will get home

taste

1 2 1 2 sleep

taste Expected value after n wine tastings is

  • 1

2

n

2 / 17

slide-4
SLIDE 4

Deterministic probabilistic automaton

Given a finite alphabet A, a probabilistic automaton is a function Q

  • [0, 1] × (DQ)A

with an initial state distribution Language of type A∗ → [0, 1] defined by determinization

3 / 17

slide-5
SLIDE 5

Determinization

NFAs: Q

  • 2 × (PQ)A

⇒ PQ

  • 2 × (PQ)A

Probabilistic automata: Q

  • [0, 1] × (DQ)A

⇒ DQ

  • [0, 1] × (DQ)A

Crucial: P and D are monads for which 2 and [0, 1] are algebras

4 / 17

slide-6
SLIDE 6

D-semantics: transitions

State space DQ Transition DQ → DDQ ⇒ flatten with monad multiplication

q0

taste

1 2 1 2 q1

taste . . .

q0

taste

1 2 q0 + 1 2 q1

taste

1 4 q0 + 3 4 q1

taste

1 2 · (1 2q0 + 1 2q1) + 1 2q1 = 1 4q0 + 3 4q1

5 / 17

slide-7
SLIDE 7

D-semantics: outputs

State space DQ Output DQ → D[0, 1] → [0, 1] D-algebra: expected value 1

q0

taste

1 2 1 2 q1

taste 1

1 2 1 4

. . .

q0

taste

1 2 q0 + 1 2 q1

taste

1 4 q0 + 3 4 q1

taste

1 4 · 1 + 3 4 · 0 = 1 4

6 / 17

slide-8
SLIDE 8

Nondeterministic probabilistic automaton

1

will get home

taste taste

1 2 1 2 sleep

taste Expected value after n tastings is one of 1, 1 2,

1

2

2

,

1

2

3

, . . . ,

1

2

n−1

,

1

2

n

Which ones “make sense”?

7 / 17

slide-9
SLIDE 9

Main results

Nondeterministic probabilistic automata have ◮ two natural (categorical) semantics ◮ more expressive power than deterministic ones ◮ undecidable equivalence problem

8 / 17

slide-10
SLIDE 10

Nondeterministic probabilistic automaton

A nondeterministic probabilistic automaton is a function Q

  • [0, 1] × (PDQ)A

with a set of initial state distributions Language of type A∗ → [0, 1] But: PD is not a monad

9 / 17

slide-11
SLIDE 11

Convex powerset

For a convex set X, PcX = convex subsets of X (finitely generated, nonempty)

10 / 17

slide-12
SLIDE 12

Convex powerset

For a convex set X, PcX = convex subsets of X (finitely generated, nonempty) Convex algebra structure on these sets: pointwise pU + (1 − p)V = {pu + (1 − p)v | u ∈ U, v ∈ V } For example, in Pc[0, 1]: 0.3 · [0.1, 0.5] + 0.7 · [0.2, 1] = [0.17, 0.85]

10 / 17

slide-13
SLIDE 13

Adjusted monad

Gives a monad PcD Multiplication: PcDPcDX → PcPcDX → PcDX convex algebra structure on Pc

11 / 17

slide-14
SLIDE 14

Two natural semantics

Need: PcD-algebra on [0, 1] Comes down to a Pc-algebra α PcD[0, 1]

  • PcE
  • Pc[0, 1]

α

[0, 1]

12 / 17

slide-15
SLIDE 15

Two natural semantics

Need: PcD-algebra on [0, 1] Comes down to a Pc-algebra α PcD[0, 1]

  • PcE
  • Pc[0, 1]

α

[0, 1]

Claim: α ∈ {min, max} Proof outline: Pc[0, 1] generated by {0}, {1}, [0, 1] α({0}) = 0, α({1}) = 1

  • α determined by α([0, 1])

α([0, 1])2 = α([0, 1]) ⇓ α([0, 1]) = 0 or α([0, 1]) = 1

12 / 17

slide-16
SLIDE 16

Expressivity (min semantics)

1 1 1 a, b a

q0 1 2 1 2

a, b

q1

a b

q2 1 2 1 2

a, b

q3

Language: u → 2−n, n = length of longest sequence of a’s in u Not accepted by any WFA

13 / 17

slide-17
SLIDE 17

Undecidability of equivalence (min semantics)

Threshold: given probabilistic automaton X and κ ∈ [0, 1], is there u ∈ A∗ such that LX(u) < κ? Y = κ κ X A A A Z = κ A

14 / 17

slide-18
SLIDE 18

Approximate equivalence

Given c ∈ [0, 1) and l1, l2 : A∗ → [0, 1], dc(l1, l2) =

  • u∈A∗

|l1(u) − l2(u)| ·

c

|A|

|u|

Given λ > 0 we can compute x such that |dc(l1, l2) − x| ≤ λ

15 / 17

slide-19
SLIDE 19

Note on singleton alphabets

Expressivity and undecidability proofs do not extend to |A| = 1 Separate proofs using linear recurrence sequences Expressivity result still holds ◮ Proof uses Skolem–Mahler–Lech and Cayley–Hamilton Deciding equivalence is at least hard ◮ Reduction from Positivity to Threshold

16 / 17

slide-20
SLIDE 20

Conclusions

Contributions ◮ Two natural semantics ◮ More expressivity ◮ Undecidability of equivalence Interplay of convex algebra, number theory, category theory Approximation techniques interesting for verification applications ◮ different metrics?

17 / 17