Statistical Verification of Cyber-Physical Systems using Surrogate Models and Conformal Inference

We propose a scalable statistical-verification framework for cyber-physical systems that combines data-efficient surrogate models with distribution-free conformal inference. The surrogate learns a probabilistic mapping from system inputs to temporal-logic satisfaction, while conformal prediction supplies valid, finite-sample confidence guarantees. Empirical results on automotive and aerospace benchmarks demonstrate up to a 20× reduction in simulation runs compared with traditional Monte-Carlo verification, without sacrificing statistical soundness.

In International Conference on Cyber-Physical Systems (ICCPS 2022)