Geometric representation learning is a fundamental aspect of representation learning on non-Euclidean structured data and achieved promising progress in several areas like graph representations, knowledge graph embeddings and query answering. The main idea of geometric representation learning is to design geometric models that provide geometric inductive bias for data with specific geometry. To preserve the complex underlying logical patterns in data, Dr. Xiong considers geometric representations of symbolic data like knowledge graphs for knowledge reasoning.
From left to rigth: Prof. Dr. Staab, Dr. Xiong, Prof. Dr. Schulte, and Prof. Dr. Niepert.
Here is a list of selected publications by Dr. Xiong during his PhD:
- Hyperbolic embedding inference for structured multi-label prediction [NeurIPS 2022]
- Pseudo-Riemannian Graph Convolutional Networks [NeurIPS 2022]
- Ultrahyperbolic Knowledge Graph Embeddings [KDD 2022]
- Faithful Embeddings for Knowledge Bases [ISWC Best paper award 2022]
- Shrinking Embeddings for Hyper-Relational Knowledge Graphs [ACL 2023]
- NestE: Modeling Nested Relational Structures for Knowledge Graph Reasoning [AAAI 2024]
Bo Xiong (in the middle) and some members of the Analytic Computing department.