eduzhai > Applied Sciences > Engineering >

Accelerating Federated Learning over Reliability-Agnostic Clients in Mobile Edge Computing Systems

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 14 pages

Abstract: Mobile Edge Computing (MEC), which incorporates the Cloud, edge nodes and enddevices, has shown great potential in bringing data processing closer to thedata sources. Meanwhile, Federated learning (FL) has emerged as a promisingprivacy-preserving approach to facilitating AI applications. However, itremains a big challenge to optimize the efficiency and effectiveness of FL whenit is integrated with the MEC architecture. Moreover, the unreliable nature(e.g., stragglers and intermittent drop-out) of end devices significantly slowsdown the FL process and affects the global model s quality Xin suchcircumstances. In this paper, a multi-layer federated learning protocol calledHybridFL is designed for the MEC architecture. HybridFL adopts two levels (theedge level and the cloud level) of model aggregation enacting differentaggregation strategies. Moreover, in order to mitigate stragglers and enddevice drop-out, we introduce regional slack factors into the stage of clientselection performed at the edge nodes using a probabilistic approach withoutidentifying or probing the state of end devices (whose reliability isagnostic). We demonstrate the effectiveness of our method in modulating theproportion of clients selected and present the convergence analysis for ourprotocol. We have conducted extensive experiments with machine learning tasksin different scales of MEC system. The results show that HybridFL improves theFL training process significantly in terms of shortening the federated roundlength, speeding up the global model s convergence (by up to 12X) and reducingend device energy consumption (by up to 58 ).

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×