Zero-knowledge machine learning application in blockchain for decentralized computing

Abstract

This dissertation introduces a novel decentralized application, ZKaggle, designed to facilitate a collaborative yet secure platform for computational task sharing and verification, capitalizing on blockchain technology and Zero-Knowledge Proofs (ZKPs). The development leverages the Filecoin Hyperspace Testnet for deploying smart contracts and Vercel for front-end deployment, employing Next.js to ensure a responsive user interface. The application enables users to act as bounty providers or hunters, engaging in verifiable and monetizable computational tasks. A seamless workflow encompassing task creation, execution, submission, and verification is delineated, underlining the transparent and user-centric design of the platform. Compared to other projects aiming to decentralize computation, our work expands their use case and incorporates decentralized storage to enhance user experience. After multiple experiments, we have successful deployment and functionality with simpler machine learning models, such as handwritten digit recognition. However, the scalability concerning more complex models poses a significant challenge due to blockchain's performance constraints. To address this, a myriad of future recommendations is proposed, focusing on scaling to accommodate intricate models, on-chain verification optimization, user interface enhancement, cross-platform compatibility, security fortification, and community building. Through a blend of modern technologies, frameworks, and cryptographic protocols, the dissertation lays the groundwork for a robust, user-friendly platform, paving the way for further innovation in decentralized computing and machine learning communities.