KAKURENBO: Adaptively Hiding Samples in Deep Neural Network Training

This paper proposes a method for hiding the least-important samples during the training of deep neural networks to increase efficiency, i.e., to reduce the cost of