COntext-free model checking for Recursive PrObabilistic pRogrAms

Descripción del proyecto

Garantizar la seguridad de los programas probabilísticos recursivos

Los programas probabilísticos (PP) recursivos son una técnica de aprendizaje automático cada vez más utilizada que, mediante la modelización probabilística, permite que los ordenadores efectúen actividades que antes llevaban a cabo exclusivamente las personas. En el proyecto CORPORA, financiado por las Acciones Marie Skłodowska-Curie, se garantizará la seguridad y corrección de estos programas informáticos, especialmente en el internet de las cosas y los sistemas integrados. Las técnicas facilitarán el estudio teórico y práctico de la comprobación de modelos de PP recursivos, los cuales pueden contener procedimientos recursivos gobernados por una pila y mostrar un comportamiento libre de contexto en las trazas de ejecución. Al ampliar el marco de comprobación de modelos libres de contexto a los PP recursivos, el equipo de CORPORA proporcionará un comprobador de modelos para las propiedades libres de contexto de estos programas, con lo que contribuirá al desarrollo de «software» más seguro basado en inteligencia artificial.

Objetivo

IoT and embedded systems are powered by increasingly sophisticated software components, employing machine learning to create devices that perform activities once exclusively carried out by humans. Since these activities may involve significant risks and responsibilities, ensuring the correctness and safety of involved software components is crucial.
Probabilistic Programs (PPs) are often employed in AI-powered software, particularly to exploit Bayesian inference, and to model randomized algorithms. Thus, studying verification of PPs can enable verification techniques for ensuring safety and correctness of AI-powered programs. PPs are computer programs that, besides ordinary programming constructs, may contain random choices, and variable assignments according to a random distribution.
The CORPORA (COntext-free model checking for Recursive PrObabilistic pRogrAms) project aims at developing new techniques for the verification of Recursive Probabilistic Programs, one of the most expressive classes of PPs. Recursive PPs may contain recursive procedures. Since procedures are governed by a stack, they exhibit a context-free behavior in execution traces: verification of procedural programs has motivated the study of model checking pushdown formalisms. The need for specifying properties concerning the stacks contents has led to the introduction of ad-hoc specification formalisms, including temporal logics featuring context-free-aware modalities, such as CaRet, NWTL and POTL, the latter being introduced by the applicant during his Ph.D.
The CORPORA project will extend the context-free model checking framework to recursive PPs, and its research objectives will encompass the theoretical and practical study of model checking thereof. The projects expected results include a model checker for context-free properties of recursive PPs, which will enable the practical evaluation of the techniques developed in the project.