Learning Compact Convolutional Neural Networks with Nested Dropout

Recently, nested dropout has been shown as a method for ordering representation units in autoencoders by their information content, without diminishing reconstruction cost. However, it has only been applied to training fully-connected autoencoders in unsupervised learning. We explore the impact of nested dropout on the convolutional layers in a CNN trained by backpropagation, investigating whether nested dropout can provide a simple and systematic way to determine the optimal representation size with respect to the desired accuracy and desired task and data complexity. Additionally, we hope that ordering parameters may provide additional insights into optimization of deep convolutional neural networks and how the network architecture impacts performance.