Efficient Formal Safety Analysis of Neural Networks