Formal Modeling and Verification of Convolutional Neural Networks based on MSVL

Abstract

With the rapid development and wide application of neural networks, it is more and more important to use formal methods to verify and ensure their security. In this paper, we propose a comprehensive formal framework for the modeling and verification of convolutional neural networks (CNN). The framework is developed based on Modeling, Simulation and Verification Language (MSVL), a formal language with temporal-logic basis. First, the structure and basic behavior of a CNN are characterized hierarchically as MSVL specifications. On this basis, the prediction model, training model and verification module are developed. Experimental results show that the framework constructs formal models of CNNs effectively and supports the verification of various network properties.