Deep Neural Networks in Representation Learning and Tensor Factorization

Representation learning is hugely successful in natural language processing and knowledge graph modelling. The basic concept is that entities or words are mapped to latent vectors of a given dimension and a shallow or a deep neural network then maps those representations to one or several probabilistic statements. We discuss relationships between representation learning and tensor factorization. We show how deep neural networks can be used to learn representations for new entities and new events in the industrial TIA selection tool and in visual perception. We will also discuss how tensor-trains can be used in combination with recurrent neural networks for video classification.