Sign in for Members and Observers

IFIP WG1.3 Foundations of System Specification

Talk "Why do you need to learn if you can program it?"

by Michele Loreti

Thu, 08 September 2022 at 11:30 am in Lipari, Italy

Joint work with: M. Belenchia, F. Corradini, M. Quadrini

Abstract: In the last years different attempts have been done to apply Machine Learning and Neural Networks to Formal Methods with the goal of developing new verification techniques. Unfortunately, many of these approaches are often limited to specific contexts and limited to specific classes of systems or properties. In this talk we will present a novel calculus that can be used to program Graph Neural Networks. The proposed language is based on simple graph operators that can be also used to integrate GNN models obtained from starndard learning techniques. A simple compiler has been developed that permits executing programs on high parallel infrastructures, such as GPUs or TPUs. The proposed approach has been used to develop a CTL model-checker. The performance of the obtained tool have been compared with the state-of-the-art tool $\mu$-CRL2 using the benchmark of the Model Checking Contest 2022.