Towards Efficient Verification of Quantized Neural Networks