Generic wireless networking optimisation works by looking at a single layer of the communication stack and optimising some (usually few) paramters. But all is connected and changing the transmission power (at the Physical layer) will change the connectivity of the network (at the Network and Transport layers) and finally impact the Quality of Service (at the Application layer). So how do we see which changes are most impactful? How do we map causal relationships in the network?
Graph Machine Learning is a powerful tool to understand how complex systems evolve and to determine the relationships of their components over time. In this project, the students will play with Wireless netorks (either real or simulated) and develop a Graph AI solution to determine how topology, node density and other node properties impact global performance.
The students will receive a wireless network simulator and develop their own flavour of Graph ML algorithms. Specifically they will:
- Collect networking data from mutiple layers of the communication stack.
- Build a Graph Neural Network (GCNN or other) and apply it to the network traffic data.
- Infer how topology and local setting impact global behaviour and how to steer the network towards optimality.
20% Theory, 60% Simulations, 20%Writing.
Alessandro Chiumento (firstname.lastname@example.org)