Blockchain-Enabled and Multisignature-Powered Verifiable Model for Securing Federated Learning Systems
Aditya Pribadi Kalapaaking, Ibrahim Khalil, Mohammed Atiquzzaman
Abstract
The Internet of Things (IoT) is revolutionizing numerous industrial applications by employing smart devices in manufacturing and industrial processes. Industries based on IoT generate extensive data, typically analyzed using various machine learning (ML) models. Federated learning (FL) is an emerging, privacy-preserving ML method where clients train models locally and develop a global model based on the aggregation of local models, without sharing the local data set with a third party. However, FL methods struggle to achieve trustworthiness and incorporate accountable ML principles. Blockchain technologies are being developed across different industries to enhance trust and security. This article proposes a blockchain-enabled, verifiable model for securing FL within IoT systems. Our proposed framework combines a trusted execution platform (TEE) to secure each client’s local model training process, and multisignature-powered global model verification to ensure ML model verifiability. We conducted several experiments with different data sets to assess our proposed framework. The experiments demonstrated the high efficiency and scalability of the proposed framework.