On-device Learning for resource-constrained 6G Ambient IoT devices
Reference number | |
Coordinator | RISE Research Institutes of Sweden AB |
Funding from Vinnova | SEK 99 744 |
Project duration | January 2025 - June 2025 |
Status | Completed |
Venture | 6G - Competence supply |
Call | 6G - Supervision of degree work |
Important results from the project
The goal of the thesis was to make Mondrian forests ready for embedded devices. Mondrian forests enable on-device learning also on low-power devices without the need for training in the cloud. Before the thesis, only a Python version existed. The thesis has ported the existing implementation to C and evaluated the performance. The thesis is available at https://uu.diva-portal.org/smash/record.jsf?pid=diva2%3A1979257&dswid=-6317
Expected long term effects
The expected result was to demonstrate the feasibility of online learning on Ambient IoT devices by implementing an online learning system that enables model training on Ambient IoT devices. On-device training allows (1) training models via local data without sharing data, thus enabling privacy-preserving computation by design, (2) model personalization and environment adaptation, and (3) deploying accurate models in any location without stable internet connectivity. This has been achieved.
Approach and implementation
After a literature review that studied the theoretical foundations of Mondrian Forests and related work, the student ported the existing Mondrian Forests implementation from Python to C. The implementation was then optimized for resource-constraints of IoT devices and different approaches for reducing the required memory size were implemented and evaluated. The students has weekly meetings with their supervisors.