Representation learning is a set of methods which allows finding suitable feature representations of input data to perform particular tasks. Such methods can be applied on textual data (e.g. via language models), images and structured data. The resulting features are usually called embeddings. In this seminar, we study representation learning methods for knowledge graphs.
The seminar will cover recent state-of-the-art research in knowledge graph representation learning (KGRL) such as:
- Preservation of relational and structural patterns of KGRL approaches
- Scalability to large knowledge graphs
- Inductive reasoning approaches
- KGRL for hyper-relational graphs
- Temporal knowledge graph support
- Multi-modal KGRL approaches
- Combinations of language models and KGRL
- Distillation methods
- Impact of different geometric spaces
- Bias in knowledge graph embeddings
- Support for rules & ontologies in KGRL approaches
Registration: To participate in the seminar, please register via Selma. You can find the list of topics in the slides of the seminar. If you want to select one of the topics, please contact me via mail.