TY - JOUR
T1 - Incentive Mechanism Design for Federated Learning: Challenges and Opportunities
AU - Zhan, Yufeng
AU - Li, Peng
AU - Guo, Song
AU - Qu, Zhihao
N1 - Funding Information:
AcknowLedgMents This research was supported by the funding from the Hong Kong RGC Research Impact Fund (RIF) with the Project No. R5060-19 and R5034-18; the General Research Fund (GRF) with the Project No. 152221/19E and 15220320/20E; the Collaborative Research Fund (CRF) with the Project No. C5026-18G; the National Natural Science Foundation of China (Grant 61872310); the ROIS NII Open Collaborative Research 2021-(Grant 21S0601); and the China Postdoctoral Science Foundation (Grant 2019M661709).
Publisher Copyright:
© 1986-2012 IEEE.
PY - 2021/7/1
Y1 - 2021/7/1
N2 - Federated learning is a new distributed machine learning paradigm that many clients (e.g., mobile devices or organizations) collaboratively train a model under the orchestration of a parameter server (e.g., service provider), while keeping the training data locally. One of the main challenges in federated learning is the data island, that is, each client maintains its local data and has no incentive for contributing data to model training if no reward is granted. Thus, we must motivate a large number of clients to participate in federated learning to break the limitation of data in the form of isolated islands. We discuss the fundamental research challenges in the incentive mechanism design for federated learning, and present a general framework with potential solutions to the challenges. Experiments are conducted to verify the effectiveness of the proposed framework. With several future research directions identified in incentive mechanism design for federated learning, we expect that more research interest will be stimulated in this novel area.
AB - Federated learning is a new distributed machine learning paradigm that many clients (e.g., mobile devices or organizations) collaboratively train a model under the orchestration of a parameter server (e.g., service provider), while keeping the training data locally. One of the main challenges in federated learning is the data island, that is, each client maintains its local data and has no incentive for contributing data to model training if no reward is granted. Thus, we must motivate a large number of clients to participate in federated learning to break the limitation of data in the form of isolated islands. We discuss the fundamental research challenges in the incentive mechanism design for federated learning, and present a general framework with potential solutions to the challenges. Experiments are conducted to verify the effectiveness of the proposed framework. With several future research directions identified in incentive mechanism design for federated learning, we expect that more research interest will be stimulated in this novel area.
UR - http://www.scopus.com/inward/record.url?scp=85104608044&partnerID=8YFLogxK
U2 - 10.1109/MNET.011.2000627
DO - 10.1109/MNET.011.2000627
M3 - Journal article
AN - SCOPUS:85104608044
SN - 0890-8044
VL - 35
SP - 310
EP - 317
JO - IEEE Network
JF - IEEE Network
IS - 4
M1 - 9409833
ER -