Communication Efficient on Symmetric and Asymmetric Encrypted Anonymous Federated Learning
Muhammad Senoyodha Brennaf, Po Yang, Vitaveska Lanfranchi · 2023
The rise of privacy concerns and the emergence of privacy and data protection laws push companies to reassess their practices if utilizing traditional machine learning. The collection and processing of users’ private data on the central server may violate the regulation if not appropriately handled. Federated learning comes as a solution where there is no need to upload users’ data to the server. Still, it enables robust learning by collaboratively training on each client’s devices and aggregating the model gradient updates. Federated learning enhanced with proxy as a bridge and encrypted model parameters will boost anonymity, privacy, and data protection guarantee against malicious attacks such as the membership inference attack. However, encrypted data costs more to the clients’ communication and data size by more than twice the initial size. Our paper tries to address these concerns. Throughout our experiments, we found that it is possible to achieve the same level of communication cost as if it was not encrypted. Aside from the experiment analysis on this subject, we also recommend a safe and secure way to implement communication efficiently in an anonymous encrypted federated learning setup as our contribution.