摘要
自从谷歌提出联邦学习架构以来,越来越多的人开始关注通过联邦学习的方法来训练神经网络,其中以Fed AVG代表的联邦学习聚合算法,在CNN、RNN中取得了不错的效果。但随着神经网络技术的不断发展和创新,图神经网络GNN近年来受到越来越多的关注。不同于CNN、RNN处理的结构化数据,GNN面对的更多是非结构化的数据,因此这也对联邦学习的聚合算法提出了新的挑战。
Since Google proposed the federated learning architecture, more and more people have begun to pay attention to training neural networks through federated learning. Among them, the federated learning aggregation algorithm represented by FedAVG has achieved good results in CNN and RNN. But with the continuous development and innovation of neural network technology, graph neural network GNN has received more and more attention in recent years. Different from the structured data processed by CNN and RNN, GNN faces more unstructured data, so this also poses new challenges to the aggregation algorithm of federated learning.
出处
《工业控制计算机》
2022年第10期85-87,90,共4页
Industrial Control Computer
关键词
图神经网络
联邦学习
聚合算法
graph neural network
federate learning
aggregation method
