Federify: A Verifiable Federated Learning Scheme Based on zkSNARKs and Blockchain
Ghazaleh Keshavarzkalhori, Cristina Pérez‐Solà, Guillermo Navarro‐Arribas, Jordi Herrera‐Joancomartí, Habib Allah Yajam
Abstract
Federated learning (FL) has emerged as an alternative to traditional machine learning in scenarios where training data is sensitive. In federated learning, training is held at end devices, and thus data does not need to leave users devices. However, most approaches to federated learning rely on a central server to coordinate the learning process which, in turn, introduces its own security and privacy problems. We propose Federify, a decentralized federated learning framework based on blockchain which employs homomorphic encryption and zero knowledge proofs to provide security, privacy, and transparency. The scheme preserves the confidentiality of both the data used for training and the local models using homomorphic encryption. zkSNARKs are used to provide security by verifying the contributions from the different agents, and transparency of both the learning process and the incentive mechanism is achieved by delegating coordination into a smart contract in a public blockchain. We have also implemented, deployed, and evaluated a proof of concept of our framework, to demonstrate its viability both in terms of computational resources needed and cost to train in a public generic blockchain such as Ethereum.