Z Zhu, J Hong, J Zhou - International conference on machine …, 2021 - proceedings.mlr.press
… model using aggregated knowledge from heterogeneous users, … Moreover, the ensemble knowledge is not fully utilized to … In this work, we propose a data-free knowledgedistillation …
… the knowledge from previous rounds, and the local training induces forgetting the knowledge … To this end, we propose a novel and effective algorithm, Federated Not-True Distillation (…
D Yao, W Pan, Y Dai, Y Wan, X Ding, H Jin… - arXiv preprint arXiv …, 2021 - arxiv.org
… For all the CV tasks, we set the channels of each group as 16. The optimizer of client training used is SGD, we tune the learning rate in {0.1, 0.05, 0.01} for FedAvg and set the learning …
H Wang, Y Li, W Xu, R Li, Y Zhan… - Proceedings of the …, 2023 - openaccess.thecvf.com
… of existing federateddistillation methods, we in this paper propose a novel federated distillation … training with a proximal term in the model objective [16]. FEDDFUSION is a data-based …
… (2) for one class, the importance of knowledge are different among local models of clients. To facilitate more effective knowledgedistillation, we propose customized label sampling and …
X Li, B Chen, W Lu - Applied Intelligence, 2023 - Springer
… knowledgedistillation process (FedDKD). In FedDKD, we introduce a decentralized knowledge distillation (DKD) module to distill the knowledge of local … speed of the algorithms [16]. …
C Wu, F Wu, L Lyu, Y Huang, X Xie - Nature communications, 2022 - nature.com
… We compare FedPAQ with 16-bit or 8-bit precision levels. … for personalized learning and knowledgedistillation. In addition, … -efficient than other compared federated learning-based …
J Zhang, S Guo, X Ma, H Wang… - Advances in Neural …, 2021 - proceedings.neurips.cc
… Motivated by the paradigm of KnowledgeDistillation (KD) [12–16] that knowledge can be transferred from a neural network to another via exchanging soft predictions instead of using …
… distillation algorithm that aggregates structural knowledge with … inherent heterogeneity of decentralized federated learning. … address this issue by distilling on public data [16, 22, 4], they …