Convolutional neural networks usually learn useful features from training data. Depending on the task, the features of the first convolution layers might represent something elementary for the training data. For example, in image data, the learned features could represent edges and blobs. In the subsequent layers of the network the…