Periodic Reporting for period 1 - ENEXA (Efficient Explainable Learning on Knowledge Graphs)
Berichtszeitraum: 2022-10-01 bis 2024-03-31
Knowledge graphs are a family of data structures that are used in an increasing number of user-facing applications. They are now widely regarded as a key enabler for explainable machine learning for the masses. However, the knowledge graphs used in real applications are often large, inconsistent and incomplete. Hence, they come with three main challenges. First, they challenging to store and query, especially when analytical queries are required to gather high-value data. Second, standard inference engines cannot be used on them, as they simply terminate their computation when faced with inconsistencies. Finally, their incompleteness often leads to downstream applications remaining unaware of highly relevant assertions. Moreover, their non-vectorial representation makes it challenging to deploy classical machine learning techniques for analysis or exploitation.
The goal of ENEXA is to address the challenge of making explainable machine learning on Web-scale knowledge graphs possible. To achieve this goal, the project specificies, implements, and evaluates novel extraction, storage, inference, and explainable AI techniques for knowledge graphs at large scale. Once completed, our approaches will make machine learning on Web-scale knowledge graph possible and empower small and large Europoean enterprises to use knowledge graphs reliably as cornerstone for their data strategy.
The goal of ENEXA is to address the challenge of making explainable machine learning on Web-scale knowledge graphs possible. To achieve this goal, the project specificies, implements, and evaluates novel extraction, storage, inference, and explainable AI techniques for knowledge graphs at large scale. Once completed, our approaches will make machine learning on Web-scale knowledge graph possible and empower small and large Europoean enterprises to use knowledge graphs reliably as cornerstone for their data strategy.
ENEXA has already been able to deliver innovation in the following areas:
– In the area of storage, we are continuing to contribute to the development of Tentris, one of the world‘s most efficient graph databases. We were already able to show that our solution is up to 1000x faster than the state of the art on datasets such as WikiData (5.5 billion triples)
– We have unified a significant portion of knowledge graph embedding techniques into the Keci approach, which exploits the expressive power of Clifford algebras to generalize over the state of the art. With Keci, we are now able to learn embedding spaces as well as embeddings.
– We have developed a modular reasoner for large-scale knowledge graph that is able to detect and alleviate inconsistencies in large knowledge graphs with up to 10,000 assertions in under 3 min.
– We have unified class expression learning and reinforcement learning to achieve significant improvement of the runtime of the former by an order of magnitude while retaining the accuracy and explainability of the computed models.
– We have developed a retrieval-augmented generated explainer for class expression learning that can combine class expressions and provenance information to achieve co-constructive explainations of machine learning results.
– In the area of storage, we are continuing to contribute to the development of Tentris, one of the world‘s most efficient graph databases. We were already able to show that our solution is up to 1000x faster than the state of the art on datasets such as WikiData (5.5 billion triples)
– We have unified a significant portion of knowledge graph embedding techniques into the Keci approach, which exploits the expressive power of Clifford algebras to generalize over the state of the art. With Keci, we are now able to learn embedding spaces as well as embeddings.
– We have developed a modular reasoner for large-scale knowledge graph that is able to detect and alleviate inconsistencies in large knowledge graphs with up to 10,000 assertions in under 3 min.
– We have unified class expression learning and reinforcement learning to achieve significant improvement of the runtime of the former by an order of magnitude while retaining the accuracy and explainability of the computed models.
– We have developed a retrieval-augmented generated explainer for class expression learning that can combine class expressions and provenance information to achieve co-constructive explainations of machine learning results.
See section 2. The corresponding publications are at https://enexa.eu/publications/(öffnet in neuem Fenster). Key needs to ensure further uptake will be elaborated upon during the second phase of the project.