https://github.com/0xZKML/zk-mnist

Demo of ML for MNIST classification in a zero knowledge proof
https://github.com/0xZKML/zk-mnist

Last synced: 2 months ago
JSON representation

Demo of ML for MNIST classification in a zero knowledge proof

• Host: GitHub
• URL: https://github.com/0xZKML/zk-mnist
• Owner: 0xZKML
• Created: 2022-03-05T00:14:07.000Z (over 2 years ago)
• Default Branch: main
• Last Pushed: 2022-11-16T07:43:55.000Z (over 1 year ago)
• Last Synced: 2024-01-28T16:07:30.210Z (5 months ago)
• Language: TypeScript
• Homepage:
• Size: 50.1 MB
• Stars: 105
• Watchers: 3
• Forks: 11
• Open Issues: 1
• Metadata Files:
  • Readme: README.md

Lists

• awesome-zk - zk-MNIST: web frontend app + Jupyter notebook with ML model generation
• awesome-zkml - zk-mnist - [@hopanml](https://twitter.com/hopanml) [@sunfishstanford](https://twitter.com/fho888) [@henripal](https://twitter.com/henripal) (2022) (Codebases / Articles and podcasts)
• awesome-zkml - zk-mnist - [@hopanml](https://twitter.com/hopanml) [@sunfishstanford](https://twitter.com/fho888) [@henripal](https://twitter.com/henripal) (2022) (Codebases / Articles and podcasts)
• awesome-zero-knowledge-proofs - zk-MNIST: web frontend app + Jupyter notebook with ML model generation on Ethereum

README

        
# 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](https://0xparc.org/blog/program-for-applied-research)'s winter 2021 applied zk learning group, and draws heavily from 0xJOF's [zk learning in public](https://github.com/JofArnold/zkp-learning-in-public) repo and Wei Jie Koh's [zk nft mint repo](https://github.com/weijiekoh/zknftmint/blob/main/contracts/contracts/NftMint.sol)

## Current Functionality:

1) draw a digit or select an image of a hand-drawn digit

2) pass the digit through 2 conv layers and 2 FC layers in browser, generating a dim 84 embedding

3) generate a zkSNARK proof in browser that the embedding represents a given digit

4) verify proof on-chain using ethers + snarkjs

## How to run it locally:

Prerequisites: global install of circom 2.0

1) `git clone` the repo

2) `cd` into the directory

3) `npm i` to install dependencies

4) generate powers of tau `yarn zk:ptau`

5) 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/`

6) compile the contracts `npx hardhat compile`

7) start a local ether node: `npx hardhat node`

8) switch to another terminal

9) deploy the smart contract ` npx hardhat run scripts/deploy.js --network localhost`

10) make a note of where the contract address has been deployed

11) edit `verifierAddress` in `./src/config.js`

12) start web app `npm start`

## Development loop:

### ZK circuits

All zk circuits are in the `zk` directory.

1) generate basic powers of tau phase 1 with `yarn zk:ptau`

2) compile the circuits, generate the solidty validator with `yarn zk:compile` (but see 5a/5b above)