Deep Neural Networks and the Tree of Life

In Evolutionary Biology, species close in the tree of evolution are identified by similar visual features. In computer vision, deep neural networks perform image classification by learning to identify similar visual features. This leads to an interesting question: is it possible to leverage the advantage of deep networks to construct a tree of life? In this paper, we make the first attempt at building the phylogenetic tree diagram by leveraging the high-level features learned by deep neural networks. Our method is based on the intuition that if two species share similar features, then their cross activations in the softmax layer should be high. Based on the deep representation of convolutional neural networks trained for image classification, we build a tree of life for species in the image categories of ImageNet. Further, for species not in the ImageNet categories that are visually similar to some category, the cosine similarity of their activation vectors in the same layer should be high. By applying the inner product similarity of the activation vectors at the last fully connected layer for different species, we can roughly build their tree of life. Our work provides a new perspective to the deep representation and sheds light on possible novel applications of deep representation to other areas like Bioinformatics.