Loading...

Proposal af a New Approach to Improve Communication and Computation Efficiency of Federated Edge Learning with Heterogeneous Resources

Moazam Sedeh, Marjan | 2023

62 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56608 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Hesabi, Shahin
  7. Abstract:
  8. In practice, to implement Federated learning on edge networks, devices must repeatedly transmit their trained models to the edge server through wireless links to update the global model; Due to the heterogeneity in the Federated Edge learning system, such as lack of power, some devices may not be able to connect to the base station and reduce the performance of the trained model or, if connected, impose a large delay on the learning process. To overcome the mentioned challenge, we propose a Collaborative Federated learning framework. In this framework, some devices can participate in Federated learning without connecting to the base station; In this way, devices that are not able to participate in Federated learning can collaborate with other devices and participate in learning through them. The connection of edge devices can be implemented with various topologies; As a result, in order to minimize the weighted sum of energy consumption and delay, we simultaneously optimize the speed of calculations and the topology of all devices involved in learning, and we solve the mixed nonlinear problem by providing a penalty-based successive convex approximation solution
  9. Keywords:
  10. Federated Learning ; Collaborative Edge Computing (CEC) ; Heterogeneous Resources ; Successive Approximation ; Collaborative Federated Learning ; Edge Nodes Topology ; Computing Speed Optimization

 Digital Object List

 Bookmark

...see more