Computer Vision News - September 2021

2 Summary Deep Learning Research 6 colorectal cancer, which creates a need to develop an optimized, robust model across varied cluster size and density. Deep learning approaches such as CNN provide a more granular correlation of features but, as discussed in earlier works, this doesn’t apply to a small training dataset. In this work, a novel architecture is proposed, which attempts to be computational efficient and utilize the phylogenetic tree to predict both continuous and binary outcome, as can be seen in Fig 1. Methodology The network is effectively a CNN which regularizes the phylogeny of the microbiome prediction. The taxa are clustered based on a phylogeny-induced correlation structure. To achieve high accuracy, convolutional layers are designed to include the phylogenetic correlation across different phylogenetic depths as much as possible. With a clever use of convolutional layers, it achieves both the dimensionality reduction but also the encoding of the phylogenetic information, leading to the recognition of spatially local input patterns . Figure 1. The architecture of the proposed MDeep. You can see the input, convolutional layers, fully connected and the output of the network, together with the dimensions of each. Notice, that in the last fully connected layer, a single node for continuous outcome and two nodes for binary outcome are shown.