Secure Data Sharing in the Internet of Vehicles Using Blockchain-based Federated Learning

This project implements the proof-of-concept of blockchain-based federated learning system with zero-knowledge proofs for model aggregation verification.

We used:

• Hyperledger Fabric for private blockchain network
• EZKL for zero-knowledge proofs
• Flower framework for federated learning

Full Thesis: Link (NB! Implementation details on page 70)

DEMO Video: Link

Project Structure

Blockchain Component (/blockchain)

Class diagram can be found in figs/blockchain_uml.txt

The blockchain component is built using Hyperledger Fabric and consists of:

Smart Contracts

• Local Model Chaincode (asset-transfer-basic/chaincode-go/chaincode/local_model_chaincode.go)
  • Handles metadata for local ML models (no weights)
  • Tracks model hashes, number of examples, round IDs etc.
  • Uses LocalModelSmartContract to manage LocalModel assets
• Global Model Chaincode (asset-transfer-private-data/chaincode-go/chaincode/global_model_chaincode.go)
  • Manages aggregated global models
  • Stores model weights in Private Data Collections
  • Uses GlobalModelSmartContract to manage GlobalModelAsset and GlobalModelPrivateDetails

REST API Gateway

Golang gateway server (/rest-api-go)

• Provides HTTP endpoints for organizations to interact with the blockchain
• Implements OrgSetup for managing organization credentials and connections
• Exposes methods to invoke and query chaincode

Federated Learning Component (/fl)

Class diagram can be found in figs/fl_uml.txt

The FL system is built using Flower framework and consists of:

Core Components

• Client (src/client_app.py)
  • FlowerClient class for training local models
  • Integrates with blockchain API to verify and submit local models
  • Uses PyTorch for model training
• Server (src/server_app.py)
  • Implements federated averaging strategy
  • Verifies local models via blockchain before aggregation
  • Submits aggregated models to blockchain
• Strategy (src/strategy.py)
  • BlockchainStrategy extends Flower's FedAvg
  • Handles model verification and blockchain integration
  • Generates zero-knowledge proofs for aggregation
• Task (src/task.py)
  • Defines the ML model architecture
• ZKP (src/zkp.py)
  • Zero-knowledge proof setup, generation and verification logic
• Blockchain API (src/blockchain_api.py)
  • Interface to interact with the blockchain network

Setup and Usage

Blockchain Network Setup

0. Installation:

Follow the instructions in https://hyperledger-fabric.readthedocs.io/en/release-2.5/getting_started.html

1. Start the Fabric test network:

cd blockchain/test-network
./network.sh up createChannel -ca

2. Deploy local model chaincode:

./network.sh deployCC -ccn local_model_chaincode -ccp ../asset-transfer-basic/chaincode-go -ccl go -ccep "OR('Org2MSP.peer')" -ccv 1.0

3. Deploy global model chaincode with private data collections:

./network.sh deployCC -ccn global_model_chaincode -ccp ../asset-transfer-private-data/chaincode-go/ -ccl go -ccep "OR('Org1MSP.member')" -cccg ../asset-transfer-private-data/chaincode-go/collections_config.json -ccv 2.0

4. Start Org1 (aggregation server) gateway REST API:

cd blockchain/asset-transfer-basic/rest-api-go
go run main.go

5. Start Org2 (client) gateway REST API:

go run main.go -orgName Org2 -mspID Org2MSP -cryptoPath ../../test-network/organizations/peerOrganizations/org2.example.com -peerEndpoint localhost:9051 -gatewayPeer peer0.org2.example.com -port 3001

Federated Learning Setup

0. Follow the detailed instructions in fl/README.md to:

• Set up the Python environment
• Generate certificates for secure communication
• Prepare the dataset

1. Start the Flower server with superlink authentication:

flower-superlink \
    --ssl-ca-certfile certificates/ca.crt \
    --ssl-certfile certificates/server.pem \
    --ssl-keyfile certificates/server.key \
    --auth-list-public-keys keys/client_public_keys.csv \
    --auth-superlink-private-key keys/server_credentials \
    --auth-superlink-public-key keys/server_credentials.pub

2. Start the Flower supernode:

flower-supernode \
    --root-certificates certificates/ca.crt \
    --auth-supernode-private-key keys/client_credentials_1 \
    --auth-supernode-public-key keys/client_credentials_1.pub \
    --node-config 'dataset-path="datasets/cifar10_part_1"' \
    --clientappio-api-address="0.0.0.0:9094"

3. Start the second supernode:

flower-supernode \
    --root-certificates certificates/ca.crt \
    --auth-supernode-private-key keys/client_credentials_2 \
    --auth-supernode-public-key keys/client_credentials_2.pub \
    --node-config 'dataset-path="datasets/cifar10_part_2"' \
    --clientappio-api-address="0.0.0.0:9095"

4. Initiate the FL run:

flwr run . my-federation

System Sequence Diagrams

Federated Learning Data Sharing Flow

Blockchain Operations

Reading Models from the Ledger

Creating Models on the Ledger

License and Notice

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

As required by the Apache License, a NOTICE file is included with the distribution.