back to list

Project: Peer-to-peer Federated Learning


--update--: This project is now taken by Davis Eisaks

The goal of this project is to study how to train a machine learning model in a gossip-based approach, where if two devices (e.g smartwatches) pass each other in the physical space, they could exchange part of a model vector with each other and then incorporate the little information that they received into their own model. If you a consider an environment like a city, where different people pass each other all the time, like on the metro etc, would over time a a collaborative model emerge on each person's smartwatch? 

During the preparation phase, the feasibility of implementing this idea using raspberry pi's or simulating it and observing the convergence behaviour would be studied. This will lead to defining the scope of the MSc thesis project on peer-to-peer federated learning, i.e. without communication to a server.

Starting points could be:

Mykola Pechenizkiy
Secondary supervisor
Tim d'Hondt
Get in contact