UCBFed: Using Reinforcement Learning Method to Tackle the Federated Optimization Problem - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2021

UCBFed: Using Reinforcement Learning Method to Tackle the Federated Optimization Problem

Résumé

Federated learning is a novel research area of AI technology that focus on distributed training and privacy preservation. Current federated optimization algorithms face serious challenge in the aspects of speed and accuracy, especially in non-i.i.d scenario. In this work, we propose UCBFed, a federated optimization algorithm that uses the Upper Confidence Bound (UCB) method to heuristically select participating clients in each round’s optimization process. We evaluate our algorithm in multiple federated distributed datasets. Comparing to most widely-used FedAvg and FedOpt, the UCBFed we proposed is superior in both the final accuracy and communication efficiency.
Fichier principal
Vignette du fichier
509420_1_En_7_Chapter.pdf (265.13 Ko) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03384857 , version 1 (19-10-2021)

Licence

Paternité

Identifiants

Citer

Wanqi Chen, Xin Zhou. UCBFed: Using Reinforcement Learning Method to Tackle the Federated Optimization Problem. 21th IFIP International Conference on Distributed Applications and Interoperable Systems (DAIS), Jun 2021, Valletta, Malta. pp.99-105, ⟨10.1007/978-3-030-78198-9_7⟩. ⟨hal-03384857⟩
74 Consultations
21 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More