Private and Sustainable Extreme-Edge AI
This project demonstrates how large language models can be personalized with sensitive data on devices in the home, without transferring that data to the cloud.
The project demonstrated a proof of technology for edge computing applications using Large Language Models (LLMs). The delivered system effectively personalizes a European open-source LLM using highly sensitive personal data, with proven functionality on AI-capable edge devices. The technology incorporates advanced personal data space management, successfully implementing Solid protocols for consent management. The prototype demonstrates efficient workload execution in an extreme edge network deployed in a home environment, while maintaining data privacy and optimizing energy consumption. The solution validates the feasibility of training and deploying LLMs at the edge, specifically handling sensitive personal data without requiring cloud transmission.
Contact: Maxim Chantillon (imec) and Olivier Braet (UAntwerpen)
Researchers: Tanguy Coenen (Strategy lead - imec), Jan Deprez (Innovation specialist - imec) and Matthias Hutsebaut-Buysse (Senior researcher - UAntwerpen)
Research groups: imec – AI & Algorithms and UAntwerpen - IDLab