DENOS Lab, University of Calgary ← Back to research

Federated Learning for Edge Service Orchestration

Federated learning methods that budget clients by availability and carbon footprint and guide cloud-edge orchestration toward resilient, real-world deployment.

Overview

As federated learning (FL) spreads across heterogeneous infrastructures, slow convergence drives excessive energy use on cloud and battery-powered edge devices. The lab designs FL methods that budget clients by availability and carbon footprint and guide cloud-edge orchestration toward resilient, real-world deployment.

Three threads run through the work. Client selection decides who trains and when, with lifespan extension for battery-powered IoT networks in FedLE, availability budgets that drop the assumption that every client is reachable every round, and Owen sampling that makes contribution estimation fast enough to use while training is still going. Model adaptation changes what each client receives, adapting model size to a client’s carbon profile and location in FedGreen and allocating size dynamically for energy efficiency. Orchestration carries the result up to the platform layer, where federated prediction of workload energy lets container schedulers such as Kubernetes account for carbon without centralizing operational data from any enterprise.

Underneath all three sits the question of how the network shapes federated training, surveyed for edge computing in ACM Computing Surveys and pushed to its limit in aerial and space networks, where connectivity is intermittent and clients differ from one another far more than in a data centre.

Researchers

Publications

  • IEEE SWC2025

    Towards Carbon-Aware Container Orchestration: Predicting Workload Energy Consumption with Federated Learning

    Zainab Saad, Jialin Yang, Henry Leung, and Steve Drew

    paper

  • arXiv2025

    Owen Sampling Accelerates Contribution Estimation in Federated Learning

    Hossein KhademSohi, Hadi Hemmati, Jiayu Zhou, and Steve Drew

    paper

  • IEEE INFOCOM-W2025

    Semi-decentralized Federated Time Series Prediction with Client Availability Budgets

    Yunkai Bao, R. Safarzadeh, Xin Wang, and Steve Drew

  • IEEE BigData2024

    Energy-efficient Federated Learning with Dynamic Model Size Allocation

    M. S. C. Kumar, Yunkai Bao, Xin Wang, and Steve Drew

  • IEEE ICC-W2024

    FedGreen: Carbon-aware Federated Learning with Model Size Adaptation

    Ali Abbasi, Fan Dong, Xin Wang, Henry Leung, Jiayu Zhou, and Steve Drew

    paper

  • ACM CSUR2024

    Topology-aware Federated Learning in Edge Computing: A Comprehensive Survey

    Jiajun Wu, Fan Dong, Henry Leung, Zhuangdi Zhu, Jiayu Zhou, and Steve Drew

    paper

  • IEEE GLOBECOM2023

    Federated Learning with Client Availability Budgets

    Yunkai Bao, Steve Drew, Xin Wang, Jiayu Zhou, and Xiaoguang Niu

  • IEEE ICC2023

    FedLE: Federated Learning Client Selection with Lifespan Extension for Edge IoT Networks

    Jiajun Wu, Steve Drew, and Jiayu Zhou

    paper

← All research projects