Learning distributed representations of concepts using linear relational embedding

In this paper we discuss methods for general-izing over relational data. Our approach is to learn distributed representations for the con-cepts that coincide with their semantic fea-tures and then to use these representations to make inferences. We present Linear Re-lational Embedding (LRE), a method that learns a mapping from the concepts into a feature-space by imposing the constraint that relations in this feature-space are modeled by linear operations. We then show that this lin-earity constrains the type of relations that LRE can represent. Finally, we introduce Non-Linear Relational Embedding (NLRE), and show that it can represent any relation. Results of NLRE on a small but diÆcult problem show that generalization is much better than the one obtained using backprop-agation. 1