LLM-Driven Invariant Discovery for Networked Systems
A new approach to invariant discovery for networked systems partitions the problem into an AI-driven grammar discovery phase followed by a statistical search. This method leverages LLMs for semantic reasoning while addressing their non-determinism and opacity, producing auditable invariants with formal guarantees. The work is detailed in arXiv:2607.22944.
Key facts
- Invariants are relations expected to hold among measured signals of a network.
- Writing invariants by hand requires expertise in formal logic and networking.
- Automatic miners require the grammar of admissible invariants as input.
- Existing miners learn only exact, hard rules and struggle with noise.
- LLMs provide semantic reasoning but are non-deterministic and opaque.
- The proposed method partitions the problem into grammar discovery and search.
- The approach allows hallucination-prone AI to produce auditable invariants.
- The paper is available on arXiv with ID 2607.22944.
Entities
Institutions
- arXiv