Neural Variational Inference For Estimating Uncertainty in Knowledge Graph Embeddings

• DOI: 10.48550/arxiv.1906.04985
• Publication Date: 2019-01-01
• Authors: Alexander I. Cowe...; Pasquale Minervini; Tim Rocktäschel; Matko Bošnjak; Sebastian Riedel; Jun Wang
• Keywords: Knowledge graph

Recent advances in Neural Variational Inference allowed for a renaissance latent variable models variety of domains involving high-dimensional data. While traditional variational methods derive an analytical approximation the intractable distribution over variables, here we construct inference network conditioned on symbolic representation entities and relation types Knowledge Graph, to provide distributions. The new framework results highly-scalable method. Under Bernoulli sampling framework, alternative justification commonly used techniques large-scale stochastic inference, which drastically reduce training time at cost additional lower bound. We introduce two from this highly scalable probabilistic namely Latent Information Fact models, reasoning knowledge graph-based representations. Our improve upon baseline performance under certain conditions. use learnt embedding variance estimate predictive uncertainty during link prediction, discuss quality these estimates. source code datasets are publicly available online https://github.com/alexanderimanicowenrivers/Neural-Variational-Knowledge-Graphs.