Topologically-Informed Large Language Models for Innovation Intelligence

Short description of the innovation study

TOPO-LLM will develop a large language model that draws on the structure of Futurity Systems’ FAST knowledge graph to improve the discovery and prediction of emerging science and technology. For the entities used to train the model, the study will compute a compact topological embedding that captures structural and temporal information about its relationships within the wider body of scientific knowledge. These embeddings will give the language model information about how scientific knowledge is connected and how those connections develop over time, alongside the information it learns from scientific text. The embedding will then be integrated into the language model, which will be trained and evaluated using European HPC resources. The resulting model will extend FAST’s ability to help researchers, technology companies and other organizations identify emerging areas of science and technology and explore connections across the knowledge graph.

Organisations involved: 

Main SME: Marsbound SL (operating as Futurity Systems)

Technology expert: Dublin Innovation Institute (AC)