ZKML (Zero-knowledge Machine Learning) is a technology that applies zero-knowledge proofs to machine learning, serving as a bridge between AI and blockchain. Foresight Ventures researcher Maggie provides an overview of ZKML, including its background, the problems it solves, use cases, and its scope.
ZKML can solve the issues of privacy protection for AI models/inputs and verification of reasoning process, enabling small models or reasoning ZKP to be put on the chain. The significance of putting model/reasoning proof on the chain is to allow blockchain to perceive the physical world, enable smart contracts to make decisions and protect the privacy of AI models.
Applications of ZKML: 1) On-chain AI: putting AI models/AI reasoning proof on the chain, enabling smart contracts to use AI for decision-making. For example, on-chain trading systems for on-chain investment decisions; 2) Self-improving blockchain: allowing blockchain to use AI capabilities to constantly improve and adjust strategies based on historical data. For example, AI-based on-chain reputation systems; 3) AIGC on-chain: content/artworks generated by AIGC, minted into NFT on the chain, ZK can prove the correctness of the process, and the dataset does not use copyrighted images, etc.; 4) Biometric authentication (KYC) for wallets: proof of facial recognition on the chain, wallet completes KYC; 5) AI security: using AI for fraud detection, witch attack prevention, etc.; 6) On-chain ZKML games: on-chain AI chess players, NFT characters driven by neural networks, etc.
- Evening Read | The Value and Ambition of Worldcoin
- Quick Look at a16z’s Investment Landscape in Q1 2023
- Blockworks Research: In-depth analysis of Curve’s LLAMMA mechanism – Borrowing and Clearing AMM Algorithm
The earliest ZKML library was 2 years ago, and the development history of the entire technology is short. Currently, the latest ZKML library supports some simple neural network ZK conversion and application on the blockchain. It is said that basic linear regression models can be put on the chain, and various smaller neural network models can support proof on the chain. However, there are few demos seen, only one handwritten digit recognition. Some tools are said to support 100M parameters, and some claim to be able to convert GPT2 into ZK circuits and generate ZK proofs. Currently, the development direction of ZKML includes network quantization and attempting to convert large-scale parameter neural networks into ZK circuits and improve proof efficiency (expand ZK capability).
Like what you're reading? Subscribe to our top stories.
We will continue to update Gambling Chain; if you have any questions or suggestions, please contact us!