Deep Sparsity: A Distributed Representation of Deep Neural Networks
We present a neural network-based method to predict the distance between two points in a distance graph. The distance graph is a graph with edges that are considered as a point node. In particular, the distance graph contains the edges for which an observation is likely to be true, and it contains the edges for which it is not likely to be true. We study the connection between the likelihood of a new data point to the probability that that observation can be true. A new model is proposed that can predict both the distances between two regions. The new model can predict both the distances between two points, and can be used for predicting the distances between two points if the distance graph is an over-complete tree. We extend existing work in this direction, including a deep CNN architecture and an unidirectional recurrent neural network architecture that can model the prediction of distance between two points in a distance graph. Extensive experiments on various datasets demonstrate that the new model can outperform state of the art networks in predicting distances between two points.