About

zk-MNIST: web frontend app + Jupyter notebook with ML model generation Authors: @horacepan @sunfishstanford @henripal

You can play with the webapp demo at: https://zkmnist.netlify.app/ (Note: mobile browsers not supported; wallet connected to Goerli testnet required to demonstrate ZK verifier.) Tutorial blog post: https://hackmd.io/Y7Y79_MtSoKdHNAEfZRXUg

This project is part of 0xPARC's winter 2021 applied zk learning group, and draws heavily from 0xJOF's zk learning in public repo and Wei Jie Koh's zk nft mint repo

Current Functionality:

draw a digit or select an image of a hand-drawn digit
pass the digit through 2 conv layers and 2 FC layers in browser, generating a dim 84 embedding
generate a zkSNARK proof in browser that the embedding represents a given digit
verify proof on-chain using ethers + snarkjs

Prerequisites: global install of circom 2.0

git clone the repo
cd into the directory
npm i to install dependencies
generate powers of tau yarn zk:ptau
compile circuits: (yarn zk:compile doesn't work) 5a) cd zk 5b) zsh compile.sh 5c) cd .. (back to main directory) 5d) copy over some of the wasm files mkdir public/static mkdir public/static/js cp node_modules/onnxruntime-web/dist/*.wasm public/static/js/
compile the contracts npx hardhat compile
start a local ether node: npx hardhat node
switch to another terminal
deploy the smart contract npx hardhat run scripts/deploy.js --network localhost
make a note of where the contract address has been deployed
edit verifierAddress in ./src/config.js
start web app npm start

All zk circuits are in the zk directory.

generate basic powers of tau phase 1 with yarn zk:ptau
compile the circuits, generate the solidty validator with yarn zk:compile (but see 5a/5b above)