New arXiv survey maps data-driven formal methods for safe AI control

A new survey on arXiv offers a comprehensive overview of data-driven formal methods for complex dynamical systems, written by Behrad Samari, Alessandro Abate, Antoine Girard, Majid Zamani, Amy Nejati, and Abolfolaz Lavaei. The paper organizes the field around three main methodological pillars: (in)finite-abstraction-based techniques, functional certificate approaches such as control barrier certificates, and compositional methods. It covers both deterministic and stochastic systems, classifying data-driven guarantees into three categories: statistical guarantees based on probably approximately correct and scenario-based frameworks, guarantees from Lipschitz continuity, and guarantees exploiting structural properties. The survey also highlights the particular challenges of the stochastic case, where the literature is less developed than for deterministic systems. The proposal for the survey has been accepted at Automatica.