FL-Incentivizer: FL-NFT and FL-Tokens for Federated Learning Model Trading and Training

【Author】 Majeed, Umer; Khan, Latif U.; Hassan, Sheikh Salman; Han, Zhu; Hong, Choong Seon

【Source】IEEE ACCESS

【影响因子】3.476

【Abstract】Federated learning (FL) is an on-device distributed learning scheme that does not require training devices to transfer their data to a centralized facility. The goal of federated learning is to learn a global model over several iterations. It is challenging to claim ownership rights and commercialize the global model efficiently and transparently. Additionally, incentives need to be provided to ensure that devices participate in the FL process. In this paper, we propose a smart contract-based framework called FL-Incentivizer, which relies on custom smart contracts to maintain flow governance of the FL process in a transparent and immutable manner. FL-Incentivizer commercializes and tokenizes the global model using FL-NFT (FL Non-Fungible Token) based on the ERC-721 standard. FL-Incentivizer uses ERC-20 compliant FL-Tokens to incentivize devices participating in FL. We present the system design and operational sequence of the FL-Incentivizer. We provide implementation and deployment details, complete smart contract codes, and qualitative evaluation of the FL-Incentivizer. After implementing FL-Incentivizer for a global iteration of a Federated learning task, we showed the FL-NFT on OpenSea and an FL-Token for a learner on MetaMask. FL-NFTs can be traded on markets such as OpenSea like other NFTs. While FL-Tokens can be transferred in the same manner as other ERC-20-based tokens.

【Keywords】Nonfungible tokens; Federated learning; Smart contracts; Token networks; Decentralized applications; Training; Modeling; Ethereum; smart contracts; token; NFT; ERC-20; ERC-721; incentive; model trading; model learning; DApp

【发表时间】2023

【收录时间】2023-02-27

【文献类型】实验仿真

【主题类别】

区块链技术-协同技术-联邦学习