Imagine this scenario: You're a person from the world of Web2, deeply intrigued by the possibilities of Web3. In the realm of Web2, you have an impeccable track record, boasting a stellar credit history. However, as you step into the uncharted territory of Web3, you find yourself starting from scratch with no digital presence to your name.
Your desire is to establish your identity and prove your worth in this exciting new frontier of the internet, but you also harbor concerns about safeguarding your personal privacy.
In this project, we have employed Fully Homomorphic Encryption (FHE), zkML, and zkOracle technology to create a credit record evaluation system that is resistant to censorship and safeguards privacy.
We have trained a logistic regression model based on real-world data to predict a user's credit rating (credible or not). The predictions generated by this model, along with their zero-knowledge proofs, will be recorded on-chain through zkOracle.
We implemented the project in a cencorship-resistant manner.
When training the model, we TenSEAL, a homomorphic encryption libaray based on Microsoft SEAL
In the training process, training data and model parameters are BOTH encrypted with CKKS, which allows basic operations like vector sum and multiplication.
Inference are done in EZKL, and zkML libaray. Along with each inference, there will also be a zk-snark proof, proving the correctness of inference and preserving the privacy of both user (their critical data), and the model (parameters)
Generating a zk-snark proof is computationally expensive, so it is done off chain. We use zkGraph as Oracle for on-chain and off-chain communication and ensure the consensus.