Deep neural networks trained in an end-to-end manner are proven to be
efficient in a wide range of machine learning tasks. However, there is one
drawback of end-to-end learning: The learned features and information are
implicitly represented in neural network parameters, which cannot b