ScaleSFL: A Sharding Solution for Blockchain-Based Federated Learning

TL;DR

ScaleSFL introduces a scalable sharding approach for blockchain-based federated learning, enhancing security and performance by separating off-chain components and verifying model updates efficiently.

Contribution

The paper presents ScaleSFL, a novel sharding solution that improves scalability and security in blockchain federated learning systems, demonstrated through a Hyperledger Fabric prototype.

Findings

Sharding improves validation performance linearly.

The solution remains efficient and secure.

Prototype demonstrates feasibility with Hyperledger Fabric.

Abstract

Blockchain-based federated learning has gained significant interest over the last few years with the increasing concern for data privacy, advances in machine learning, and blockchain innovation. However, gaps in security and scalability hinder the development of real-world applications. In this study, we propose ScaleSFL, which is a scalable blockchain-based sharding solution for federated learning. ScaleSFL supports interoperability by separating the off-chain federated learning component in order to verify model updates instead of controlling the entire federated learning flow. We implemented ScaleSFL as a proof-of-concept prototype system using Hyperledger Fabric to demonstrate the feasibility of the solution. We present a performance evaluation of results collected through Hyperledger Caliper benchmarking tools conducted on model creation. Our evaluation results show that sharding can improve validation performance linearly while remaining efficient and secure.