The repository contains the source code for our paper Gradient Compression and Correlation Driven Federated Learning for Wireless Traffic Prediction published by IEEE Transactions on Cognitive Communications and Networking.
We are trying t build a communication-efficient and high-accurate decentralized wireless traffic prediction algorithm for future intelligent communication systems.
- To reduce data communications between local clients the central server, we introduce gradient compression into wireless traffic prediction;
- To keep prediction accuracy, we propose gradient correction and tracking into local client model training;
- To model spatial dependency, we design a family of heuristic gradient re-grouping strategy based on gradient correlation, which can reflect the correlation of BS's wireless traffic.
- Please install necessary python packages to run this code;
- Please confirm that the datasets are downloaded into the data folder;
- You can directly run the main file through
python main.py, or - You can run
bash run_exp.sh