Exploiting ladder networks for gene expression classification

The application of deep learning to biology is of increasing relevance, but it is difficult; one of the main difficulties is the lack of massive amounts of training data. However, some recent applications of deep learning to the classification of labeled cancer datasets have been successful. Along this direction, in this paper, we apply Ladder networks, a recent and interesting network model, to the binary cancer classification problem; our results improve over the state of the art in deep learning and over the conventional state of the art in machine learning; achieving such results required a careful adaptation of the available datasets and tuning of the network.