Description
DARPA’s Information Processing Techniques Office is seeking expert input on approaches, tools, and practices that could dramatically accelerate the verification of network configurations against security requirements expressed in natural language. The agency wants concepts that can reduce the time and cost of compliance checking by two to three orders of magnitude, using a system that ingests a natural‑language requirements document and a concrete network configuration, reasons about completeness and consistency, and then performs near‑real‑time compliance verification. Respondents should submit a concise one‑page narrative plus two PowerPoint slides (one containing a filled table) describing their proposed methodology, any relevant formal systems or DSLs, reasoning strategies, evaluation metrics, and illustrative impact examples. No funding or contract award is implied; this is purely an information‑gathering effort.
• Provide insights on whether a large language model alone can solve the problem or if an intermediate formal representation is required.
• Describe a minimal yet expressive domain‑specific language (DSL) that maps natural‑language abstractions to network elements.
• Suggest reasoning techniques (e.g., induction, decomposition) and evaluation metrics to assess translation fidelity and compliance‑checking accuracy.