Verifying Inverse Model Neural Networks