Model Validation: A Probabilistic Formulation

Abstract: This paper presents a probabilistic formulation of the model validation problem. The proposed validation framework is simple, intuitive, and can account both deterministic and stochastic nonlinear systems in presence of parametric and nonparametric uncertainties. Contrary to the hard invalidation methods proposed in the literature, our formulation allows a relaxed notion of validation in probability. The construction of probabilistically robust validation certificates provides provably correct guarantees. Computational complexities and numerical examples are given to illustrate the method.