eduzhai > Applied Sciences > Engineering >

Scheduling Policy and Power Allocation for Federated Learning in NOMA Based MEC

  • Save

... pages left unread,continue reading

Document pages: 6 pages

Abstract: Federated learning (FL) is a highly pursued machine learning technique thatcan train a model centrally while keeping data distributed. Distributedcomputation makes FL attractive for bandwidth limited applications especiallyin wireless communications. There can be a large number of distributed edgedevices connected to a central parameter server (PS) and iterativelydownload upload data from to the PS. Due to the limited bandwidth, only asubset of connected devices can be scheduled in each round. There are usuallymillions of parameters in the state-of-art machine learning models such as deeplearning, resulting in a high computation complexity as well as a highcommunication burden on collecting distributing data for training. To improvecommunication efficiency and make the training model converge faster, wepropose a new scheduling policy and power allocation scheme usingnon-orthogonal multiple access (NOMA) settings to maximize the weighted sumdata rate under practical constraints during the entire learning process. NOMAallows multiple users to transmit on the same channel simultaneously. The userscheduling problem is transformed into a maximum-weight independent set problemthat can be solved using graph theory. Simulation results show that theproposed scheduling and power allocation scheme can help achieve a higher FLtesting accuracy in NOMA based wireless networks than other existing schemes.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×