핵심 개념
Fluent introduces a round-efficient secure aggregation scheme for private federated learning, reducing communication rounds and latency.
초록
The article introduces Fluent, a secure aggregation scheme for private federated learning, comparing it with existing solutions like Bell et al. (CCS 2020) and Ma et al. (SP 2023). Fluent aims to minimize communication rounds and latency, improving computational efficiency and reducing communication overhead. The article discusses the challenges of privacy inference and inversion attacks in federated learning and the need for secure aggregation schemes. It details the key contributions of Fluent, such as one-time handshake and secret sharing, one-round consistency check and unmasking, and dynamically joining clients. Experimental results show significant improvements in computational cost and communication overhead compared to existing solutions.
통계
FL은 수천 개의 지리적으로 분산된 클라이언트를 동시에 포함하므로 통신 지연이 발생할 수 있습니다.
Fluent는 기존 솔루션에 비해 최소한의 통신 라운드를 요구합니다.
Fluent는 서버 및 클라이언트의 계산 비용을 줄이고 전역 통신 오버헤드를 완화합니다.
인용구
"Fluent achieves the fewest communication rounds in the malicious server setting."
"Experimental results show that Fluent improves computational cost by at least 75% and communication overhead by at least 25% for normal clients."