Densmitt

**Fast kNN with a self-adaptive compression approach** We present an online learning algorithm for training a convolutional neural network (CNN) model with convolutional layers and an underlying graph-based model which achieves a high accuracy in predicting the data. We train a CNN with the CNN encoder-decoder architecture, which learns to use each layer of the network as a separate layer, and this layer is trained in the CNN model. This approach combines many methods, including the recently developed ResNets and …