A Fully Decentralized Homomorphic Federated Learning Framework

The convergence of Big Data, Metaverse, and Large Language Models (LLM) has presented a new horizon for Artificial Intelligence (AI). However, due to legal and security constraints in data sourcing from multiple channels, it has become challenging to realize this potential. This paper proposes a possible solution to this problem using a Homomorphic Encrypted Federated Learning technique that is verifiable and decentralized. Unlike existing Encrypted Federated Learning systems, our method does not require local model updates to be decrypted during aggregation by the server. This results in reduced network overhead, computation time, and memory usage without compromising the convergence rate of the system. Furthermore, our approach provides a verifiable proof that the model update remains untampered.