Abstract: Despite their remarkable performance, deep neural networks remain mostly “black boxes”, suggesting inexplicability and hindering their wide applications in fields requiring making rational ...