Presentation
Balancing and Minimizing Energy Consumption of Federated Learning in Heterogeneous Mobile Edge IoT
DescriptionFederated Learning is a privacy-centric distributed learning paradigm that aims to build a highly accurate global model.in Mobile Edge IoT, FL training can drain device energy. Current optimization methods focus on reducing overall energy use, potentially causing high consumption in some devices, shortening their lifespan.For enhancing the accuracy of global model and balancing the energy consumption between devices,we introduce a novel FL training approach.We propose a client selection strategy integrates cluster partitioning and utility-driven approaches,then introduce a Sequential Least Squares Quadratic Programming scheme for effective communication resource allocation. Our approach outperforms existing methods,increasing model accuracy and reducing energy consumption gap.
Event Type
Work-in-Progress Poster
TimeTuesday, June 256:00pm - 7:00pm PDT
LocationLevel 2 Lobby
AI
Autonomous Systems
Cloud
Design
EDA
Embedded Systems
IP
Security