On-device Learning for Secure Personalized IoT

Purpose and goal:

This project aims to advance the capabilities of resource-constrained Internet of Things (IoT) devices through the development of on-device learning systems. The objective is to empower these devices to perform intelligent tasks locally, minimizing reliance on external resources and enabling real-time responsiveness. Towards this end, the project has carefully selected and implemented an example application for on-device learning: interference detection and classification for wireless networks.

Expected results and effects:

We have selected an application for on-device training: interference detection and classification, an important problem in wireless networks of resource-constrained devices. A challenge has been the setting of system parameters and the previous solution had only static, hard coded parameter settings. Previous attempts with offline learning had not provided improved results.

We could show that online-learning on the device itself enabled improved interference detection and classification and have gained a basis for further research and developments. Approach and implementation:

The approach of selecting a specific application turned out to be successful. Earlier the parameter settings were hardcoded and hence could not adapt to the specific deployment environment. Offline training had not led to positive results but online-training on the device led to improved results, even though we used simple learning algorithms.

The understanding we got through this project, plus the operating system support we implemented is a very sound base for future work in this exciting area.