Learning Compact Representations of Neural Networks using DiscriminAtive Masking (DAM)

The key intuition behind DAM is to discriminatively prefer some of the neurons to be refined during the training process, while gradually masking out other neurons.