Privacy and Security Requirements Challenges in Blockchain-Based Decentralized Federated Learning

The Internet of Things (IoT), a growing technology, is revolutionizing different fields, e.g., healthcare, smart cities, surveillance, etc. However, IoT devices become vulnerable due to the heterogeneous nature of devices, which affects the sensitive data in them. This research prioritizes users' privacy and data security through decentralized, federated learning by leveraging blockchain technology. Unlike traditional centralized learning, our approach keeps data on local devices, enhancing the users' privacy. To provide coordination and trust among the participating IoTs, we propose a Hyperledger Fabric blockchain for its decentralized, transparent, and permissioned nature, to ensure secure and reliable communication between devices in the federated learning model. We plan to write different chain codes for tasks, e.g., model updates, versioning, and integration. Additionally, we plan to use Directed Acyclic Graphs for efficient and traceable model versioning, addressing the challenges of managing updates in decentralized environments. This framework aims to transform learning models and offer a scalable, secure, and privacy-conscious solution. Model integration is achieved through weighted average techniques, which boost the accuracy of the global model by incorporating the appropriate model into it.