Verifying Binarized Neural Networks using Marabou

Binarized neural networks have recently gained popularity due to their ability to provide good performance with low computational and memory requirements. However, the robustness of binarized neural networks has not been thoroughly studied, and it is important to ensure that they are robust to perturbations of the input. In this work, we use the Marabou framework to verify the robustness of binarized neural networks and evaluate it on a range of binarized architectures and perturbation sizes.