QNNVerifier: A Tool for Verifying Neural Networksusing SMT-Based Model Checking

This paper presents QNNVerifier, a tool for verifying implementations of neural networks that takes into accounts the finite word-length (i.e. quantization) of their operations. QNNVerifier achieves this goal by translating the implementation of neural networks to a decidable fragment of first-order logic based on satisfiability modulo theories (SMT) and employing state-of-the-art software model checking (SMC)techniques. The fixed- and floating-point operations are represented through direct implementations of their effects given a hardware-related precision. Furthermore, QNNVerifier allows the user to specify safety properties and verify the resulting model with different verification strategies (incremental and k-induction) and SMT solvers. Furthermore, QNN Verifieradopts invariant inference via interval analysis and discretization of non-linear activation functions to speed up the ANN verification process.