Graph Machine Learning
RouteNet-Fermi
Network-performance prediction using graph-based and sequence models.
Problem
Predict network-performance metrics across network topologies and traffic conditions using learned representations of network state.
My contribution
Developed and evaluated machine-learning approaches for network-performance prediction, including graph-based modeling and an LSTM baseline.
Simplified architecture
Modeled network behavior from topology, path, link, and traffic-related inputs, then evaluated prediction accuracy across experiment configurations.
Outcomes
- An LSTM-based model achieved approximately 0.33% MAPE on the FatTree16 experiment configuration.
Technologies
PythonLSTMGraph Neural NetworksNetwork Performance Modeling