cargo build --release --target wasm32-wasip1
RUST_LOG=debug cargo run --release
The path to the WASM binary may need to be corrected in zk/src/main.rs.
Welcome to the zKML project, a initiative designed to revolutionize the way data privacy is handled in collaborative machine learning efforts. This project aims to address one of the most pressing challenges in the digital age: enabling data sharing for health or financial predictions while preserving the privacy and security of sensitive information.
Imagine a world where governments, healthcare institutions, financial organizations, and other entities could pool their data to make more accurate predictions that could save lives or improve financial stability. However, due to privacy concerns and trust issues, these organizations are often reluctant to share their data openly. zKML (Zero-Knowledge Machine Learning) offers a groundbreaking solution to this problem.
- Data Sensitivity: Health and financial data are extremely sensitive. Governments and institutions are wary of sharing this data due to privacy concerns and the potential for misuse.
- Trust Issues: Even in a collaborative environment, there is a lack of trust among parties about how data will be used and whether the work done by each participant can be verified without exposing the data itself.
- Proof of Work: In collaborative models, it's crucial to ensure that each party has contributed honestly without compromising the privacy of the data or the model itself.
zKML leverages Zero-Knowledge Proofs (ZKPs) to create a framework where multiple parties can collaborate on training machine learning models without revealing their private data. This is done by proving that the data and the model were used as claimed, without ever exposing the underlying sensitive information.
- Privacy-Preserving Data Sharing: zKML allows multiple parties to contribute data to a machine learning model without revealing any of the data to each other or any external party.
- Trustless Collaboration: With zKML, all parties can trust that the model is being trained correctly and that their data is being used as intended, without the need to trust other participants blindly.
- Proof of Contribution: zKML provides verifiable proof that a party’s data has been used to train the model, ensuring that all contributions are acknowledged and validated.
- Versatility in Applications: While the primary use case revolves around health and financial predictions, zKML can be applied to various domains where privacy, trust, and proof of work are crucial.
- Data Providers: Governments, institutions, and other stakeholders prepare their data for model training.
- Model Training: The machine learning model is trained across these data sets using the zKML framework, ensuring that data never leaves the provider’s control.
- Zero-Knowledge Proofs: Throughout the process, zero-knowledge proofs are generated to verify that each step of the model training is performed correctly, without revealing the actual data.
- Verification and Deployment: Once the model is trained, all parties can verify the model’s integrity and the contributions made, enabling its deployment for practical use in health or financial predictions.
The implications of zKML are profound. By allowing secure, private collaboration on sensitive data, zKML can unlock new levels of insight and predictive power in critical areas like healthcare and finance. This can lead to better decision-making, improved public health outcomes, and more robust financial systems, all while maintaining the privacy and trust of the data providers.
This project was created as part of the zK Montreal Hackathon to explore the possibilities of Zero-Knowledge Machine Learning (zKML) in solving real-world privacy challenges. We’re just getting started, and we’re excited to see where this journey takes us!