zkml demo

Private quantized linear regression on Ethereum.

This is a demo of the zkml protocol, which implements a zk-SNARK circuit where the proof verifies that a private model has a certain accuracy under a public dataset, as well as the public encrypted model is exactly the private model encrypted using the shared key.

Presentation: https://www.loom.com/share/fdbe451385dd4a22b5593bf80747a4be

Running

• download powersOfTau28_hez_final_18.ptau from circom and move to circuits
• mkdir artifacts && mkdir artifacts/lr to store artifacts
• cd circuits && yarn prod to build the circuits
• cd eth && yarn compile && yarn deploy-{NETWORK} to deploy the contracts
• set up Infura IPFS with keys/ipfs.json containing {"id": ..., "secret": ...}
• export PRIVATE_KEY=... && export URL=... to export private key and RPC URL
• for the jupyter demo, run jupyter kernelspec list to find kernel.json and add an "env": {"PRIVATE_KEY": ..., "URL": ...} entry
• ./zkml to interact with cli (xDai)

Check it out on-chain

There is an outstanding 20 xDai bounty on the contract, available for claim as of December 2021.

https://blockscout.com/xdai/mainnet/address/0x5B54f06991871cd7EAE76a3D270D9EFFBdC01207/contracts

Protocol Overview

Special Thanks

• ETH Summer
• Lattice