Date: | 2022, May 24 |
Time: | 10:30 a. m. |
Place: | G29-018 |
Author: | Memariani, Adel |
Title: | Geometric Embeddings and their Applications |
Most of the real-world knowledge graphs (KG) are notably incomplete. Link prediction (A.k.a. KG-completion) methods aim to predict missing information in the KGs and add the inferred facts to the KGs. However, KG-completion methods lead to a significantly larger graph, which slows down the reasoning process because traversing intermediary nodes in a reasoning path is computationally expensive. Therefore, finding answers to complex questions is challenging in KGs. Query embedding techniques can implicitly account for missing edges in the KGs. In this presentation, we will talk about recent advances in query embedding methods where logical queries and entities are modeled as closed regions in the embedding spaces. Furthermore, we will discuss the benefits of reasoning in hyperbolic space instead of euclidean space.