Research
Cluster based federated learning for IoT environments, under a research collaboration scholarship.
- Period
- 2025-2026
- Universitat Jaume I
- GEOTEC
- Line of work
- Federated learning, IoT
The starting point
Federated learning trains a model across many devices without the data ever leaving them. Each node trains locally and only sends parameters. It is the sensible way to learn from data that cannot move, whether for privacy or for volume.
The problem appears when the nodes are heterogeneous, which is exactly the IoT case. Different sensors, different environments, different data distributions. A single global model averages all of that away and ends up mediocre for every node taken on its own.
What I researched
A system that groups nodes into clusters by behaviour and trains one model per cluster instead of one for everybody. Each device receives a model that resembles the data it actually sees.
Evaluation ran under typical IoT conditions: intermittent connectivity, nodes appearing and disappearing mid round, and limited compute at the edge.
- 2 Conference papers In publication process
- 1 Scientific article In publication process