文献类型：期刊文献

英文题名：Client selection based weighted federated few-shot learning

作者：Xu, Xinlei[1,2];Niu, Saisai[3,4];Zhe, Wanga[1,2];Li, Dongdong[2];Yang, Hai[2];Du, Wenli[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Dept Comp Sci & Engn, Shanghai 200237, Peoples R China;[3]Shanghai Aerosp Control Technol Inst, Shanghai 201109, Peoples R China;[4]China Aerosp Sci & Technol Corp, Res & Dev Ctr Infrared Detect Technol, Shanghai 201109, Peoples R China

年份：2022

卷号：128

外文期刊名：APPLIED SOFT COMPUTING

收录：;EI(收录号：20223512628852);WOS:【SCI-EXPANDED(收录号:WOS:000884754600008)】；

基金：This work is supported by Shanghai Science and Technology Program, China ''Federated based cross-domain and cross-task incremental learning'' under Grant No. 21511100800, Shanghai Science and Technology Program, China ''Distributed and generative few-shot algorithm and theory research'' under Grant No. 20511100600, Natural Science Foundation of China under Grant No. 62076094, Chinese Defense Program of Science and Technology under Grant No. 2021-JCJQ-JJ-0041, China Aerospace Science and Technology Corporation Industry-University-Research Cooperation Foundation of the Eighth Research Institute under Grant No. SAST2021-007, National Science Foundation of China for Distinguished Young Scholars under Grant No. 61725301.

语种：英文

外文关键词：Federated learning; Few-shot learning; Meta learning; Image classification

摘要：With the advancement of technology, clients have a large amount of personal data. Human beings are paying more and more attention to the privacy and security of this part of data. Clients do not want to share private data, which directly leads to the existence of data islands. Especially in few-shot scenarios, due to the insufficient amount of personal data, constructing an effective few-shot model is difficult. To solve the above problems, we propose Federated Few-shot Learning (FedFSL) in this paper. We utilize Federated Learning (FedL) to ensure the privacy and security issues of joint training. Moreover, the global model obtained by FedL has the characteristics of universally applicable to all clients. That universality satisfies the scenario of universally applicable to all meta tasks in the fewshot meta-learning stage. What is more, to obtain a more effective global federated universal few-shot model, we respectively proposed Weighted FedL (WFedL) and Client Selection based FedL (CSFedL) strategies. WFedL takes into account the difference between clients performance when building the global model and assigns different weights to different clients. CSFedL considers the malicious participation of clients, and we propose an adaptive client selection strategy to mitigate the impact caused by malicious participation. Extensive federated experiments on CIFAR-10 and CIFAR-100 show the advantage of proposed WFedL, CSFedL and combined Client Selection based WFedL (CSWFedL). We further verify the performance improvement of FedFSL on miniImagenet and propose our overall framework Client Selection based WFedFSL (CSWFedFSL). The best performance of CSWFedFSL is higher than both the few-shot baseline and FedFSL, and CSWFedFSL protects clients data privacy in the few-shot scenario. (c) 2022 Elsevier B.V. All rights reserved.