Collaborative Multi-device, Multi-network Cellular Throughput Prediction
FedPut improves cellular throughput prediction using federated learning across diverse devices and networks, enhancing accuracy and robustness while preserving data privacy.
Key Features
- Federated Learning: Collaborative training across multiple devices without sharing raw data.
- Cross-Technology Learning: Uses 4G data to bootstrap 5G predictions, improving model adaptability.
- Privacy-Preserving: Ensures user data remains on-device, complying with data security policies.
- High Accuracy: Achieves over 90% R² score in multi-network, multi-device environments.
- Real-World Validation: Evaluated with real and simulated datasets across different geographic regions.
Contributors
Argha Sen
IIT Kharagpur, India
Basabdatta Palit
NIT Rourkela, India
Soumyajit Chatterjee
Nokia Bell Labs, UK
Sandip Chakraborty
IIT Kharagpur, India
Publications
- Argha Sen, Ayan Zunaid, Soumyajit Chatterjee, Basabdatta Palit, and Sandip Chakraborty: "Revisiting Cellular Throughput Prediction over the Edge: Collaborative Multi-device, Multi-network in-situ Learning", EWSN 2023