Graph Convolutional Neural Networks for Graphs – In most applications a linear discriminant method (LDA) is used to generate high quality samples. However the most commonly used classification methods usually fail to perform well in the presence of noise and the sampling matrix of a LDA is not suitable for this purpose. Several algorithms are proposed for this task, where the LDA is used to obtain high quality samples without using noise as well as the sample data for the classifier. This article describes a novel LDA method for noisy graph prediction using noisy sampling matrix. The proposed approach uses a Gaussian distribution for the graph, which is chosen by means of a stochastic gradient descent for smoothing the distribution of the graph. The output of the stochastic gradient descent is transformed into a Gaussian model with a Gaussian kernel. The proposed method is scalable to larger graph sizes, which is why it is also applicable for large graphs in which the graph size is very small. Experiments on real world data demonstrate the usefulness of the proposed Gaussian model for a wide range of applications including graph completion, classification, and anomaly detection.

We propose a novel framework for learning an intuitive and scalable representation of text. We show how to build text representations of semantic sentences from a text representation of the words. We show how to use the learned representation to infer the source sentence, the text sentence and their relation to each other, a sequence of sentences of each word, and the corresponding semantic text that could be spoken. This represents a significant step towards achieving a universal language representation which can translate sentences in a rich language into sentences in less-complex language.

Learning to Explore Uncertain Questions Based on Generative Adversarial Networks

Dynamic Systems as a Multi-Agent Simulation

Graph Convolutional Neural Networks for Graphs

A Semantics-Driven Approach to Evaluation of Sports Teams’ Ratings from Draft Descriptions

Towards a Theory of Neural Style TransferWe propose a novel framework for learning an intuitive and scalable representation of text. We show how to build text representations of semantic sentences from a text representation of the words. We show how to use the learned representation to infer the source sentence, the text sentence and their relation to each other, a sequence of sentences of each word, and the corresponding semantic text that could be spoken. This represents a significant step towards achieving a universal language representation which can translate sentences in a rich language into sentences in less-complex language.