FedChain

This repository is for our submission.

Start

1. We use Hyperledger Fabric v2.2 as the default blockchain.

Prerequisites:

1. HyperLedger Fabric v2.2.x LTS
2. Download this repository, and merge BEAS/fabric-samples with the HyperLedger fabric-samples directory.

Network Setup:

$  cd fabric-samples/blockchain-network
$  ./network.sh fedchain

2. Each client has a lightweight server based on node.js listening and processing requests off the chain.

1. Create a working dictionary in fabric-samples

$  cd fabric-samples/offchain-server

2. Install server dependancies:

$  npm install

3. Run web server for each client:

$  node client.js

3. Install Caliper dependancies:

$  npm install --only=prod @hyperledger/caliper-cli
$  npx caliper bind --caliper-bind-sut fabric:2.1

3. We start multiple client processes for local training.

1. Start 10 clients

$ ./start_clients.sh 10

2. End

$ ./end.sh

4. To facilitate the performance comparison of various federated learning methods, we implemented an off-chain version to simplify operation.

1. Modify the hyperparameters in ./config/conf.json
2. Run the script such as:

$  python main.py -c ./config/conf.json

5. We plot the convergence comparison with plot.py based on the results in ./offchain/offchain-train/results