Rethinking Architecture Design for Tackling Data Heterogeneity in Federated Learning.
Rethinking Architecture Design for Tackling Data Heterogeneity in Federated Learning.
复制标题
DOI:
10.1109/cvpr52688.2022.00982
复制
发表时间:
2022-06
期刊:
影响因子:
--
通讯作者:
Rubin, Daniel
中科院分区:
文献类型:
--
作者:
Qu, Liangqiong;Zhou, Yuyin;Liang, Paul Pu;Xia, Yingda;Wang, Feifei;Adeli, Ehsan;Li Fei-Fei;Rubin, Daniel
Federated learning is an emerging research paradigm enabling collaborative training of machine learning models among different organizations while keeping data private at each institution. Despite recent progress, there remain fundamental challenges such as the lack of convergence and the potential for catastrophic forgetting across real-world heterogeneous devices. In this paper, we demonstrate that self-attention-based architectures (e.g., Transformers) are more robust to distribution shifts and hence improve federated learning over heterogeneous data. Concretely, we conduct the first rigorous empirical investigation of different neural architectures across a range of federated algorithms, real-world benchmarks, and heterogeneous data splits. Our experiments show that simply replacing convolutional networks with Transformers can greatly reduce catastrophic forgetting of previous devices, accelerate convergence, and reach a better global model, especially when dealing with heterogeneous data. We release our code and pretrained models to encourage future exploration in robust architectures as an alternative to current research efforts on the optimization front.
登录
查看更多内容
DOI:
10.3390/s20216230
发表时间:
2020-10-31
期刊:
Sensors (Basel, Switzerland)
影响因子:
--
作者:
Jiang JC;Kantarci B;Oktug S;Soyata T
通讯作者:
Soyata T
影响因子:
4.3
作者:
Hu, Rui;Zhao, Shuang;Su, Juan;Chen, Xiang;Yin, Mingzhu
通讯作者:
Yin, Mingzhu
DOI:
10.1186/s13634-019-0602-z
发表时间:
2019-01-28
影响因子:
1.9
作者:
Chen, Xin;Wang, Ding;Wu, Ying
通讯作者:
Wu, Ying
影响因子:
4.9
作者:
Brisimi TS;Chen R;Mela T;Olshevsky A;Paschalidis IC;Shi W
通讯作者:
Shi W
影响因子:
1.7
作者:
Guo, Ruichao;Jiang, Xiaomeng;Wang, Hongren
通讯作者:
Wang, Hongren