联邦学习的计算效率优化方法综述OA

Federated learning,as a distributed machine learning paradigm,enables collaborative model training with privacy preservation through parameter aggregation mechanisms,demonstrating potential in addressing data silo issues in fields such as smart healthcare and the Internet of Things.However,computational efficiency bottlenecks in resource-constrained devices se-verely hinder its practical deployment.Compared to traditional centralized learning paradigms,federated learning still faces challenges such as high communication overhead and insufficient adaptability to dynamic environments while protecting data privacy.This paper systematically reviewed efficiency enhancement methods for federated learning from three major directions:model compression,update strategy optimization,and data computation optimization.Existing studies show that collaborative optimization of compression strategies and dynamic resource allocation reduces communication overhead and accelerates conver-gence.Nevertheless,current approaches still encounter challenges including model accuracy loss and inadequate adaptability to dynamic environments.Finally,it proposed future research directions to explore lightweight native model architectures,in-telligent dynamic update mechanisms,and efficient malicious detection methods,aiming to balance efficiency and security for promoting the large-scale application of federated learning in complex scenarios.