SMARTY: Federated Knowledge Representation for secure knowledge integration
INFAI’s Federated Knowledge Representation Framework leverages graph-based systems to unify and manage machine learning models, enabling intelligent lifecycle management and collaborative learning across distributed environments.
The solution combines Knowledge Graphs, federated learning techniques and privacy-preserving approaches to support knowledge integration and management. A key capability is the Data-to-Embeddings Pipeline, which enables the regularisation of embeddings generated by AI-based models and their use for the secure comparison and transfer of data among partners and stakeholders, regardless of their type or format.
Among the developed functionalities, the framework also includes a Text-to-SPARQL Transformation module, which uses locally fine-tuned Large Language Models adapted to specific Knowledge Graphs and ontologies to translate natural language questions into SPARQL queries, simplifying data access while maintaining privacy and control.
The two-step fine-tuning strategy combines reasoning-based approaches and reward-based optimisation to improve the generation of SPARQL queries by LLMs.
Follow SMARTY for updates:
LinkedIn: https://www.linkedin.com/company/smartyeu/?viewAsMember=true
Website: https://www.smarty-project.eu