Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilities and partial observation From probabilistic omega-automata to stochastic games of imperfect information
Nathalie Bertrand
Inria Rennes + Liverpool visitor
based on joint work with Christel Baier & Marcus Großer¨ Blaise Genest & Hugo Gimbert
Probabilities and partial observation OUCL Verification seminar 1 Monty Hall problem
Ethernet protocol: random choice slot, collisions, maximum nb of trials
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Motivation
Adversarial situations with I probabilities: random choices, uncertainties, losses I partial observation: distributed
Probabilities and partial observation OUCL Verification seminar 2 Ethernet protocol: random choice slot, collisions, maximum nb of trials
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Motivation
Adversarial situations with I probabilities: random choices, uncertainties, losses I partial observation: distributed
Monty Hall problem
Probabilities and partial observation OUCL Verification seminar 2 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Motivation
Adversarial situations with I probabilities: random choices, uncertainties, losses I partial observation: distributed
Monty Hall problem
Ethernet protocol: random choice slot, collisions, maximum nb of trials
Probabilities and partial observation OUCL Verification seminar 2 Problems for POMDPs Given a POMDP ( , ), an LTL formula ϕ and p [0, 1] M ∼ ∈ Question is there a based scheduler with P (ϕ) ./ p? ∼ U U
State of the art (back in 2008) I based, P (F) > 0 is EXPTIME-Comp. [de Alfaro 99] ∃U ∼ U I based, P (^F) ./ p is undecidable [Giro D’Argenio 07] ∃U ∼ (p U(0, 1) and ./ , , = ) ∈ ∈ {≤ ≥ }
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Partially observable MDP
Partially Observable MDP A POMDP ( , ) consists of an MDP equipped with an equiva- lence relationM ∼over states of . M ∼ M
Probabilities and partial observation OUCL Verification seminar 3 State of the art (back in 2008) I based, P (F) > 0 is EXPTIME-Comp. [de Alfaro 99] ∃U ∼ U I based, P (^F) ./ p is undecidable [Giro D’Argenio 07] ∃U ∼ (p U(0, 1) and ./ , , = ) ∈ ∈ {≤ ≥ }
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Partially observable MDP
Partially Observable MDP A POMDP ( , ) consists of an MDP equipped with an equiva- lence relationM ∼over states of . M ∼ M Problems for POMDPs Given a POMDP ( , ), an LTL formula ϕ and p [0, 1] M ∼ ∈ Question is there a based scheduler with P (ϕ) ./ p? ∼ U U
Probabilities and partial observation OUCL Verification seminar 3 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Partially observable MDP
Partially Observable MDP A POMDP ( , ) consists of an MDP equipped with an equiva- lence relationM ∼over states of . M ∼ M Problems for POMDPs Given a POMDP ( , ), an LTL formula ϕ and p [0, 1] M ∼ ∈ Question is there a based scheduler with P (ϕ) ./ p? ∼ U U
State of the art (back in 2008) I based, P (F) > 0 is EXPTIME-Comp. [de Alfaro 99] ∃U ∼ U I based, P (^F) ./ p is undecidable [Giro D’Argenio 07] ∃U ∼ (p U(0, 1) and ./ , , = ) ∈ ∈ {≤ ≥ }
Probabilities and partial observation OUCL Verification seminar 3 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Outline
1 Introduction
2 Probabilistic ω-automata Examples and expressiveness Emptiness problem Positive semantics (PBA>0) Almost-sure semantics (PBA=1)
3 Stochastic games of imperfect information Framework Qualitative determinacy Memory requirements Decidability
4 Conclusion
Probabilities and partial observation OUCL Verification seminar 4 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilistic Buchi-automata¨ [Baier Großer¨ 05]
Probabilistic Buchi¨ Automata NBA where nondeterminism is resolved by probabilities
ω >0( ) = w Σ P( ρ Runs(w) ρ = ^F ) > 0 L A { ∈ ω | { ∈ | | } } =1( ) = w Σ P( ρ Runs(w) ρ = ^F ) = 1 L A { ∈ | { ∈ | | } }
Probabilities and partial observation OUCL Verification seminar 5 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilistic Buchi-automata¨ [Baier Großer¨ 05]
Probabilistic Buchi¨ Automata NBA where nondeterminism is resolved by probabilities
ω >0( ) = w Σ P( ρ Runs(w) ρ = ^F ) > 0 L A { ∈ ω | { ∈ | | } } =1( ) = w Σ P( ρ Runs(w) ρ = ^F ) = 1 L A { ∈ | { ∈ | | } }
PBA are POMDP with trivial equivalence relation: p, q p q. ∀ ∼
word for PBA deterministic blind scheduler for MDP ≡
Probabilities and partial observation OUCL Verification seminar 5 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilistic Buchi-automata¨ [Baier Großer¨ 05]
Probabilistic Buchi¨ Automata NBA where nondeterminism is resolved by probabilities
ω >0( ) = w Σ P( ρ Runs(w) ρ = ^F ) > 0 L A { ∈ ω | { ∈ | | } } =1( ) = w Σ P( ρ Runs(w) ρ = ^F ) = 1 L A { ∈ | { ∈ | | } }
Example 1
ω >0 = (a + b)∗a = NBA L ω L =1 = b∗a L
Probabilities and partial observation OUCL Verification seminar 5 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilistic Buchi-automata¨ [Baier Großer¨ 05]
Probabilistic Buchi¨ Automata NBA where nondeterminism is resolved by probabilities
ω >0( ) = w Σ P( ρ Runs(w) ρ = ^F ) > 0 L A { ∈ ω | { ∈ | | } } =1( ) = w Σ P( ρ Runs(w) ρ = ^F ) = 1 L A { ∈ | { ∈ | | } }
Example 2
ω NBA = ((ac)∗(ab)) L ω >0 = (ab + ac)∗(ab) L ω =1 = (ab) L
Probabilities and partial observation OUCL Verification seminar 5 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilistic Buchi-automata¨ [Baier Großer¨ 05]
Probabilistic Buchi¨ Automata NBA where nondeterminism is resolved by probabilities
ω >0( ) = w Σ P( ρ Runs(w) ρ = ^F ) > 0 L A { ∈ ω | { ∈ | | } } =1( ) = w Σ P( ρ Runs(w) ρ = ^F ) = 1 L A { ∈ | { ∈ | | } }
Example 3
ω NBA = (ab + ac) L 0 = =1 = L> L ∅
Probabilities and partial observation OUCL Verification seminar 5 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Expressiveness: PBA vs NBA
Expressiveness
PBA>0 are more expressive than NBA. PBA=1 and NBA are incomparable.
Probabilities and partial observation OUCL Verification seminar 6 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Expressiveness: PBA vs NBA
Expressiveness
PBA>0 are more expressive than NBA. PBA=1 and NBA are incomparable.
I any NBA can be turned into an equivalent PBA first turn NBA into an equivalent one deterministic in the limit → I example of a PBA>0 whose language is not ω-regular
Probabilities and partial observation OUCL Verification seminar 6 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Expressiveness: PBA vs NBA
Expressiveness
PBA>0 are more expressive than NBA. PBA=1 and NBA are incomparable.
Probabilities and partial observation OUCL Verification seminar 6 Lemma For 0 < λ < µ < 1, ( ) ) ( ). L Pλ L Pµ
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilities matter
Y ( ) = ak1 bak2 b (1 λki ) > 0) L Pλ { · · · | − } i
Probabilities and partial observation OUCL Verification seminar 7 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Probabilities matter
Y ( ) = ak1 bak2 b (1 λki ) > 0) L Pλ { · · · | − } i
Lemma For 0 < λ < µ < 1, ( ) ) ( ). L Pλ L Pµ
Probabilities and partial observation OUCL Verification seminar 7 Proof sketch Reduction of the modified emptiness problem for PFA w P (w) ε or PFA with ∀ R ≤ R w P (w) > 1 ε ∃ R −
1 and↓ 2 PBA s.t. P P >ε ( ) = ( 1) ( 2) = L R ∅ ⇔ L P ∩ L P ∅
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Emptiness for PBA>0 [Baier B. Großer¨ 08]
Theorem The emptiness problem is undecidable for PBA.
Probabilities and partial observation OUCL Verification seminar 8 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Emptiness for PBA>0 [Baier B. Großer¨ 08]
Theorem The emptiness problem is undecidable for PBA.
Proof sketch Reduction of the modified emptiness problem for PFA w P (w) ε or PFA with ∀ R ≤ R w P (w) > 1 ε ∃ R −
1 and↓ 2 PBA s.t. P P >ε ( ) = ( 1) ( 2) = L R ∅ ⇔ L P ∩ L P ∅
Probabilities and partial observation OUCL Verification seminar 8 First undecidability results in qualitative verification of POMDP.
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Consequences for POMDP
Undecidability results for POMDP The following problems are undecidable I Given ( , ) and F set of states of , is there a deterministic M ∼ M based such that P (^F) > 0. ∼ U U I Given ( , ) and F set of states of , is there a deterministic M ∼ M based such that P (^F) = 1. ∼ U U
Probabilities and partial observation OUCL Verification seminar 9 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Consequences for POMDP
Undecidability results for POMDP The following problems are undecidable I Given ( , ) and F set of states of , is there a deterministic M ∼ M based such that P (^F) > 0. ∼ U U I Given ( , ) and F set of states of , is there a deterministic M ∼ M based such that P (^F) = 1. ∼ U U
First undecidability results in qualitative verification of POMDP.
Probabilities and partial observation OUCL Verification seminar 9 Proof sketch 1. emptiness problem and almost-sure reachability are interreducible for PBA
w, Pw (^F) = 1 w, Pw (^M) = 1 ∃ P ≡ ∃ P
2. almost-sure reachability for POMDP is decidable
based, PU (^M) = 1 ∃U ∼ M
Skip proof
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Emptiness for PBA=1
Theorem The emptiness problem is decidable for almost-sure PBA.
Probabilities and partial observation OUCL Verification seminar 10 2. almost-sure reachability for POMDP is decidable
based, PU (^M) = 1 ∃U ∼ M
Skip proof
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Emptiness for PBA=1
Theorem The emptiness problem is decidable for almost-sure PBA.
Proof sketch 1. emptiness problem and almost-sure reachability are interreducible for PBA
w, Pw (^F) = 1 w, Pw (^M) = 1 ∃ P ≡ ∃ P
Probabilities and partial observation OUCL Verification seminar 10 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Emptiness for PBA=1
Theorem The emptiness problem is decidable for almost-sure PBA.
Proof sketch 1. emptiness problem and almost-sure reachability are interreducible for PBA
w, Pw (^F) = 1 w, Pw (^M) = 1 ∃ P ≡ ∃ P
2. almost-sure reachability for POMDP is decidable
based, PU (^M) = 1 ∃U ∼ M
Skip proof
Probabilities and partial observation OUCL Verification seminar 10 I w, Pw (^F) = 1 w, Pw (^M) = 1 ∃ P ∃ P
w, Pw (^F) = 1 Pw (^M) = 1 ∀ P ⇐⇒ P0
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Proof in more details: step 1
I w, Pw (^M) = 1 w, Pw (^F) = 1 ∃ P ∃ P Hint: F = M and add self loops on F with probability one
Probabilities and partial observation OUCL Verification seminar 11 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Proof in more details: step 1
I w, Pw (^M) = 1 w, Pw (^F) = 1 ∃ P ∃ P Hint: F = M and add self loops on F with probability one
I w, Pw (^F) = 1 w, Pw (^M) = 1 ∃ P ∃ P
w, Pw (^F) = 1 Pw (^M) = 1 ∀ P ⇐⇒ P0 Probabilities and partial observation OUCL Verification seminar 11 Idea: reduction to almost-sure reachability for MDP
From POMDP ( , ) build MDP 0 by powerset construction with M ∼ M additional final state F 0. I if δ(r, a) F = : traditional powerset construction ∩ ∅ I if δ(r, a) F , : go to F 0 with probability 1/2, rest of probability mass uniformely∩ ∅ distributed over non final successors
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Proof in more details: step 2
Theorem Given ( , ) and F set of states of , it is decidable whether there M ∼ M exists an observation-based with P (^M) = 1. U U
Probabilities and partial observation OUCL Verification seminar 12 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Proof in more details: step 2
Theorem Given ( , ) and F set of states of , it is decidable whether there M ∼ M exists an observation-based with P (^M) = 1. U U Idea: reduction to almost-sure reachability for MDP
From POMDP ( , ) build MDP 0 by powerset construction with M ∼ M additional final state F 0. I if δ(r, a) F = : traditional powerset construction ∩ ∅ I if δ(r, a) F , : go to F 0 with probability 1/2, rest of probability mass uniformely∩ ∅ distributed over non final successors
Probabilities and partial observation OUCL Verification seminar 12 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Outline
1 Introduction
2 Probabilistic ω-automata Examples and expressiveness Emptiness problem Positive semantics (PBA>0) Almost-sure semantics (PBA=1)
3 Stochastic games of imperfect information Framework Qualitative determinacy Memory requirements Decidability
4 Conclusion
Probabilities and partial observation OUCL Verification seminar 13 Strategy for Player i Based on initial distribution and sequence of signals received so far, Player i chooses a distribution over actions.
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Framework [B. Genest Gimbert 09]
I two-player game I partial observation on both sides signals received by the players → I probability on next state given current one and players’ decisions I qualitative objectives
Probabilities and partial observation OUCL Verification seminar 14 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Framework [B. Genest Gimbert 09]
I two-player game I partial observation on both sides signals received by the players → I probability on next state given current one and players’ decisions I qualitative objectives
Strategy for Player i Based on initial distribution and sequence of signals received so far, Player i chooses a distribution over actions.
Probabilities and partial observation OUCL Verification seminar 14 ϕ: objective of the game for Player 1 (e.g. reachability, Buchi).¨ I From δ, Player 1 wins almost-surely if σ τ Pδ (ϕ) = 1. ∃ ∀ σ,τ I From δ, Player 1 wins positively if σ τ Pδ (ϕ) > 0. ∃ ∀ σ,τ I From δ, Player 2 wins almost-surely if τ σ Pδ (ϕ) = 0. ∃ ∀ σ,τ I From δ, Player 2 wins positively if τ σ Pδ (ϕ) < 1. ∃ ∀ σ,τ
Player i a.-s. or pos. winning from δ only depends on support: X Pδ (ϕ) = δ(s) P1s (ϕ) σ,τ · σ,τ s S ∈
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Winning almost-surely and positively
Initial distribution δ, and strategy profile (σ, τ) induce a probability measure Pδ ( ) on maximal plays. σ,τ ·
Probabilities and partial observation OUCL Verification seminar 15 Player i a.-s. or pos. winning from δ only depends on support: X Pδ (ϕ) = δ(s) P1s (ϕ) σ,τ · σ,τ s S ∈
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Winning almost-surely and positively
Initial distribution δ, and strategy profile (σ, τ) induce a probability measure Pδ ( ) on maximal plays. σ,τ ·
ϕ: objective of the game for Player 1 (e.g. reachability, Buchi).¨ I From δ, Player 1 wins almost-surely if σ τ Pδ (ϕ) = 1. ∃ ∀ σ,τ I From δ, Player 1 wins positively if σ τ Pδ (ϕ) > 0. ∃ ∀ σ,τ I From δ, Player 2 wins almost-surely if τ σ Pδ (ϕ) = 0. ∃ ∀ σ,τ I From δ, Player 2 wins positively if τ σ Pδ (ϕ) < 1. ∃ ∀ σ,τ
Probabilities and partial observation OUCL Verification seminar 15 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Winning almost-surely and positively
Initial distribution δ, and strategy profile (σ, τ) induce a probability measure Pδ ( ) on maximal plays. σ,τ ·
ϕ: objective of the game for Player 1 (e.g. reachability, Buchi).¨ I From δ, Player 1 wins almost-surely if σ τ Pδ (ϕ) = 1. ∃ ∀ σ,τ I From δ, Player 1 wins positively if σ τ Pδ (ϕ) > 0. ∃ ∀ σ,τ I From δ, Player 2 wins almost-surely if τ σ Pδ (ϕ) = 0. ∃ ∀ σ,τ I From δ, Player 2 wins positively if τ σ Pδ (ϕ) < 1. ∃ ∀ σ,τ
Player i a.-s. or pos. winning from δ only depends on support: X Pδ (ϕ) = δ(s) P1s (ϕ) σ,τ · σ,τ s S ∈
Probabilities and partial observation OUCL Verification seminar 15 Initial support: 1, 2 . Objective: reach{ t } Player 2 wins positively.
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Examples
Initial support: 1, 2 . Objective: reach{ t } Player 1 wins almost-surely.
Probabilities and partial observation OUCL Verification seminar 16 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Examples
Initial support: 1, 2 . Objective: reach{ t } Player 1 wins almost-surely.
Initial support: 1, 2 . Objective: reach{ t } Player 2 wins positively.
Probabilities and partial observation OUCL Verification seminar 16 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Consequence of undecidability of PBA>0
Undecidability Buchi¨ games with positive probability (and co-Buchi¨ games with prob- ability one) are undecidable.
I 2-player Buchi¨ games with positive probability generalize PBA>0 I randomness for free in POMDP [Chatterjee Doyen Gimbert Henzinger 10]
Probabilities and partial observation OUCL Verification seminar 17 Implies qualitative determinacy for reachability objectives as well.
NB: co-Buchi¨ games are not qualitatively determined. Details
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy
Qualitative determinacy In Buchi¨ games, every initial distribution is I almost-surely winning for Player 1, or I positively winning for Player 2.
Probabilities and partial observation OUCL Verification seminar 18 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy
Qualitative determinacy In Buchi¨ games, every initial distribution is I almost-surely winning for Player 1, or I positively winning for Player 2.
Implies qualitative determinacy for reachability objectives as well.
NB: co-Buchi¨ games are not qualitatively determined. Details
Probabilities and partial observation OUCL Verification seminar 18 (S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
Probabilities and partial observation OUCL Verification seminar 19 I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P
Probabilities and partial observation OUCL Verification seminar 19 I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution
Probabilities and partial observation OUCL Verification seminar 19 I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
Probabilities and partial observation OUCL Verification seminar 19 I (S M) L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL 1 I M = s S τ P s ( F) = 1 { ∈ | ∃ σ,τ ¬ }
Probabilities and partial observation OUCL Verification seminar 19 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \
Probabilities and partial observation OUCL Verification seminar 19 Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ
Probabilities and partial observation OUCL Verification seminar 19 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Determinacy: Proof sketch
Belief: possible states of the game according to signals received
(S): set of supports where Player 2 does not win positively L ⊆ P I for every L , Player 1 has at least one safe action, i.e. can stay in ∈ L L I σ strategy for Player 1 that, from L selects (uniformly at random) a safe set of actions for L ∈and L plays uniform distribution I σ ensures to stay in I σ is almost-surely winningL
I 1s M = s S τ Pσ,τ( F) = 1 I { (S∈ M|) ∃ ¬ } L ⊆ P \ 1 I N N s S M τ P s (^F) > 1/N ∃ ∈ ∀ ∈ \ ∀ σ,τ Player 1 wins almost-surely from L , with belief-based strategy. ⊆ L
Probabilities and partial observation OUCL Verification seminar 19 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Memory requirements
Player 1 needs exponential memory to win almost-surely. has to remember belief states →
Player 2 needs doubly exponential memory to win positively. has to remember possible beliefs of Player 1 →(beliefs of beliefs)
Probabilities and partial observation OUCL Verification seminar 20 I Player 1 better informed than Player 2: 2EXPTIME-complete I Player 2 better informed than Player 1: EXPTIME-complete I Player 1 perfectly informed: EXPTIME-complete
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Deciding stochastic games with signals
Decidability and complexity Deciding whether Player 1 almost-surely wins a reachability or Buchi¨ game is 2EXPTIME-complete.
Probabilities and partial observation OUCL Verification seminar 21 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Deciding stochastic games with signals
Decidability and complexity Deciding whether Player 1 almost-surely wins a reachability or Buchi¨ game is 2EXPTIME-complete.
I Player 1 better informed than Player 2: 2EXPTIME-complete I Player 2 better informed than Player 1: EXPTIME-complete I Player 1 perfectly informed: EXPTIME-complete
Probabilities and partial observation OUCL Verification seminar 21 Baier, B. and Großer.¨ Probabilistic ω-automata. Journal of the ACM, 2012. B., Genest and Gimbert. Qualitative determinacy and decidability of stochastic games with signals. Proceedings of LICS, 2009.
Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Concluding remarks
More results on probabilistic ω-automata in [Baier B. Großer¨ 12]
Results on stochastic games with signals very dependent on the precise framework. E.g. for deterministic strategies or deterministic memory updates, the memory size for Player 1 may be a tower of exponential [Chatterjee Doyen]
Probabilities and partial observation OUCL Verification seminar 22 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Concluding remarks
More results on probabilistic ω-automata in [Baier B. Großer¨ 12]
Results on stochastic games with signals very dependent on the precise framework. E.g. for deterministic strategies or deterministic memory updates, the memory size for Player 1 may be a tower of exponential [Chatterjee Doyen]
Baier, B. and Großer.¨ Probabilistic ω-automata. Journal of the ACM, 2012. B., Genest and Gimbert. Qualitative determinacy and decidability of stochastic games with signals. Proceedings of LICS, 2009.
Probabilities and partial observation OUCL Verification seminar 22 Introduction Probabilistic ω-automata Stochastic games of imperfect information Conclusion
Detour: co-Buchi¨ games
Co-Buchi¨ games are not qualitatively determined.
Initial state: t Player 1 perfectly informed Player 2 blind
Objective: avoid t from some point on
I Player 1 has no almost-surely winning strategy I Player 2 has no positively winning strategy
Back to main
Probabilities and partial observation OUCL Verification seminar 23