Practical verification of multi-agent systems against Slk specifications
read more
Citations
Agent systems verification : systematic literature review and mapping
Probabilistic Strategy Logic.
Verification of multi-agent systems with public actions against strategy logic
Natural Strategic Ability under Imperfect Information
Enforcing Equilibria in Multi-Agent Systems
References
Graph-Based Algorithms for Boolean Function Manipulation
The temporal logic of programs
Reasoning About Knowledge
An Introduction to MultiAgent Systems
Introduction to Multiagent Systems
Related Papers (5)
Frequently Asked Questions (14)
Q2. What are the future works in "Practical verification of multi-agent systems against slk specifications" ?
Future Work. The authors found that the main impediment to better performance of the tool is the size of the BDDs required to encode sets of extended states. Future efforts will be devoted to mitigate this problem as well as to support logics stronger than Slk, including relaxing the assumption on epistemic sentences.
Q3. How many exabytes of m1 would be needed to synthesise a?
if the authors encode each output of fz using only ⌈ log2∣∣Acshr(ϕ,z)∣∣⌉ = 4 bits and store the whole mapping in a large array, m1 will use at most 500 bytes while m2 might need up to 434 exabytes.
Q4. How many entries of the form would be needed to map m1?
While 22the mapping m1 for behavioural strategies would require at most 1000 entries of the form (s, fx(s), fy(s)) 7→ fz(s), the mapping m2 for non-behavioural strategies might have up to 1021 entries of the form (fx, fy, s) 7→ fz(s).
Q5. What is the main impediment to better performance of the tool?
The authors found that the main impediment to better performance of the tool is the size of the BDDs required to encode sets of extended states.
Q6. Why is the length of the Sl specification non-elementary?
The non-elementariness w.r.t. the length of the Sl specification is due to the alternation of the memoryful strategy quantifiers that requires alternating projection operations on the automaton, each of which induces an exponential blow-up.
Q7. How do the authors get the Ltl property verified?
Once the strategies are assigned to all corresponding variables in the Slk formula, the authors can then project them onto the interpreted structure obtaining a labelled graph, where the Ltl property is verified recursively on the structure of the formula, by mimicking the semantics definition.
Q8. What is the simplest way to fix the set of variables in the Slk formula?
Vr to always be the set of variables quantified in the Slk formula the authors are considering (e.g., if the formula to be checked is ϕ = 〈〈x〉〉[[y]](a, x)(b, y)X p, then the authors set Vr = {x, y}).
Q9. What is the proof of the model-checking result for Slk?
Before concluding this section with a proof of the model-checking result for Slk, the authors stress that although memoryless strategies are less powerful than the memoryful ones, they are more compact and, therefore, easier to handle.
Q10. How many global states can be reached in this system?
The authors also label those states corresponding to the valuation of the variable set to 1 (resp. 0).Now note that the number of reachable global states in this interpreted system is 1+m+3k.
Q11. What is the main performance bottleneck of MCMASSlk?
The experimental results presented in this section confirm that the main performance bottleneck of MCMASSlk is the BDD encoding of the extended states, which allocates separate BDD variables for each shared local state of a strategy (see Subsection 3.3).
Q12. What is the worst case time complexity of the model checking algorithm?
The worst case time complexity of the symbolic implementation of the model checking algorithm CheckI(ϕ, ∅) is:O(|ϕ| × |Ag|)× 2O(|St|×|vars(ϕ)|×log2|Ac|)
Q13. What is the worst case of the two optimisations?
Note that despite both optimisations, the worst case still remains K = |vars(ϕ)|×|St|×dlog2 (maxa∈Ag |Aca|)e, i.e., the authors need polynomially many Boolean variables with respect to both the size of the model |I| and the number of strategy variables in the formula |vars(ϕ)|.
Q14. what is the possible action of the strategy mapped to xi?
the authors find a possible action c ∈ ⋂ a∈shr(ψ,xi) Pa(sa(s′)) such that C[v, u] ∧ c[us′,xi ] is not equivalent to false, where c[us′,xi ] is the Boolean formula representing the fact that the next action of the strategy mapped to variable xi at the global state s′ is c.