SeBRUS: Mitigating Data Poisoning Attacks on Crowdsourced Datasets with Blockchain

With the rise in prominence of crowdsourced datasets in machine learning, data poisoning attacks pose a considerable threat. Many current defenses fall short because they are overly specialized for certain attacks, lack contribution incentives, and are difficult to integrate into current platforms. This paper explores the underaddressed system security problem posed by data poisoning through SeBRUS, a comprehensive data contribution application that leverages Ethereum smart contracts to secure crowdsourced datasets. SeBRUS introduces a voting network and poisoned data detection model, allowing for easy implementation with current platforms to defend against label-flipping, clean-label, and backdoor attacks.

This is SeBRUS (SEH-BRUS), or Secure Blockchain Regularization Upload System. This repository consists of the code for the SeBRUS application associated with our group's research. SeBRUS is built to allow for simple and secure machine learning dataset contribution by using the Ethereum blockchain to ensure the security of crowdsourced contributions to machine learning datasets.

This research project was conducted as part of the Governor's School of Engineering and Technology 2023 program. To read more about our research, please read our paper which was accepted to the 2023 MIT IEEE Undergraduate Research Technology Conference. For a demonstration of how to use our project, please watch this video.

Dependencies

In order to use or develop SeBRUS, you will need the following dependencies:

• Node.js: A JavaScript runtime environment.
• Yarn: A package manager for Node.js used for our React frontend.
• Python 3.0: Used for our Flask backend server.
• Metamask: A browser extension that allows you to interact with the Ethereum blockchain.
• Ganache: A local blockchain for Ethereum development.
• Truffle: A development environment for Ethereum.

Starting SeBRUS

Make sure to always start from scratch if the system has been started before. Delete test networks and accounts from MetaMask, the folder blockchain/build, the folder server/abi, and server/sqlite.db. After installing the above dependencies, clone this repository and follow the instructions below.

git clone https://github.com/bliutech/SeBRUS.git
cd SeBRUS/

1. Start Ganache & MetaMask

Start running a Quickstart workspace in Ganache. Under Contracts, link a Truffle project (truffle-config.js) in the Workspace section, and then set the gas limit to a larger number (ex. 1000000000000000) under the Chain section. This is to account for the large gas costs that come with contract calls for larger images.

In MetaMask, create a new test network under the Networks tab. Set any name for the network, and complete the following fields:

• New RPC URL: http://127.0.0.1:7545
• Chain ID: 1337
• Currency symbol: ETH

Next, import a new account in MetaMask. Under the Accounts section in Ganache, select an account. Do not select the first account (Index 0), this is reserved for deploying contracts with Truffle. Copy and paste the private key (not the account address) into MetaMask to import this wallet.

2. Start the Server

The next step is to start the Flask server. This will automatically create the SQLite database and populate it with the necessary tables. Inside the repository, run the following commands. The server should be deployed to http://localhost:5000/.

cd server
virtualenv venv --python=3.10
source venv/bin/activate
pip install -r requirements.txt
python app.py

3. Deploy the Smart Contracts

The next step is to deploy the smart contracts to the blockchain. This will create the necessary contracts and populate them with the necessary data. A supplemental script has been created to help register the corresponding ABI information to the SeBRUS backend. Inside the repository, run the following commands.

cd blockchain
truffle compile
truffle migrate
python register.py

4. Run the Client

The final step is to run the React client to interface with SeBRUS. This will automatically open a browser window to http://localhost:3000/. Inside the repository, run the following commands.

cd client
yarn install
yarn start

Due to gas limits, uploaded images must be limited to a smaller size. Images from the MNIST (28x28) and CIFAR (32x32) datasets are suitable.

Linting

This project supports CI/CD for linting. For the server application, utilize black for linting. To check if any formatting changes need to be made, run the following.

python -m black --check *.py

To resolve any needed changes run

python -m black *.py

For the client application, utilize prettier for linting. To check if any formatting changes need to be made, run the following.

npm run check

To resolve any needed changes run

npm run fix

Troubleshooting

When trying to run virtualenv venv --python=3.10, a zsh: command not found: virtualenv error may come up. If Anaconda is installed on your computer, you will have to run:

cd server
conda deactivate 
pip install virtualenv
virtualenv venv
source venv/bin/activate

Authors

SeBRUS was developed by Anusha Iyer, Chloe Lee, Cyrus Rosenberg, Rebekah Wang, & Trisha Reddy for GSET 2023. The project was advised by Benson Liu as their project mentor. For any questions or requests for additional information, please contact the authors.