Explanation of predictions made by black-box deep learning models has been rather challenging, especially when neither the model details nor its training data are known. Various techniques based on shadow-model have been ...