Anticipatory Monitoring Framework for LTLfMT Properties
A new foundational framework for anticipatory monitoring of linear-time properties enriched with SMT theories over finite traces (LTLfMT) has been proposed. The framework addresses the challenge of monitoring dynamic AI systems where internal specifications are inaccessible and model checking is inapplicable. It combines automata-theoretic methods for temporal logic with automated reasoning for SMT theories, enabling evaluation of properties based on both observed trace prefixes and all possible finite continuations. The correctness of the approach is formally proven under reasonable assumptions on the background theory. This work targets complex, heterogeneous AI systems and offers an alternative to traditional verification techniques.
Key facts
- The framework monitors LTLfMT properties over finite traces.
- It handles anticipatory monitoring where state depends on past and future continuations.
- Automata-theoretic methods are combined with SMT reasoning.
- Correctness is formally proven under assumptions on the background theory.
- Targets dynamic AI systems without accessible internal specifications.
- Offers an alternative to model checking for verification.
- The approach is foundational and novel.
- Published as arXiv:2605.14666v1.
Entities
Institutions
- arXiv