BriefGPT.xyz
Feb, 2024
回顾：联邦学习中的理解和减轻遗忘
Flashback: Understanding and Mitigating Forgetting in Federated Learning
HTML
PDF
Mohammed Aljahdali, Ahmed M. Abdelmoniem, Marco Canini, Samuel Horváth
TL;DR
研究探索了联邦学习中遗忘现象对算法收敛的影响，引入了度量遗忘程度的指标，并提出了一种动态蒸馏方法的联邦学习算法Flashback，以提高模型聚合的效果和减少遗忘现象，实现更快的收敛速度和准确性。
Abstract
In
federated learning
(FL),
forgetting
, or the loss of knowledge across rounds, hampers algorithm convergence, particularly in the presence of severe
→