This repository contains the code of our paper, Lightweight Blockchain-Empowered Secure and Efficient Federated Edge Learning.
| The golang environment (Go 1.16). The release of the Go 1.16 can be found here, and the installation instruction can be found here.
Python 3.7 is required for the go-python library go-python3. IPFS 0.19.0. Instruction for IPFS installation can be found at https://docs.ipfs.tech/install/command-line/#system-requirements. Packge requirement |
Package | Version |
|---|---|---|
| pytorch | 1.7.1 | |
| numpy | 1.18.1 | |
| scipy | 1.4.1 | |
| torchvision | 0.8.2 |
The reuqired go packages will be automatically downloaded when run the experiment.
FEMNIST: from the LEAF benchmark suite, with the relevant downloading and preprocessing instructions here. The command-line arguments for the LEAF preprocessing utility used were to generate the full-sized non-iid dataset, with minimum 15 samples/user, sample-based 80-20 train-test split were: ./preprocess.sh -s niid --sf 1.0 -k 15 -t sample --tf 0.8. The resulting training .json files files should then be copied to ../data/FEMNIST_data/train and the testing files to ../data/FEMNIST_data/test.
CIFAR10: can be downloaded here. The extracted file should be copied to ../data/CIFAR10_data/.
The hyperpearemeters about the blockchain are set in chain/variables.go。 client/variables_FL.go contains the set hyperparameters for the FL task.
First start the IPFS by running ipfs deamon, then
go run main.go