A Privacy-Preserving Federated Learning with Mutual Verification on Vector Spaces

A Privacy-Preserving Federated Learning with Mutual Verification on Vector Spaces

연구 분야: Verification

학회: International Symposium on Security and Privacy in Social Networks and Big Data

Federated learning has received widespread attention in recent years, since it trains a model by only sharing gradients without accessing training sets. In this paper, we consider two security issues in the training process of federated learning, i.e., privacy preservation and message verification, which mainly consider the security of the local gradients uploaded by clients and the aggregation result. We give the detail design about the privacy preserving federated learning with mutual authentication, which provides the privacy-preserving and mutually verifiable federated learning framework on the vector space. To extend the numerical operations to the vector space, we modify the secret sharing of numbers to that of vectors, and advance the commitment to numbers to a commitment to polynomials.