FedAdapt: Adaptive Offloading for IoT Devices in Federated Learning
FedAdapt is a holistic framework for an IoT-edge environment that surmounts the challenges of accelerating federated learning on resource constrained devices, reducing the impact of stragglers arising from computational heterogeneity of IoT devices and adapting to varying network bandwidth between devices and an edge server. To accelerate the training process of federated learning, FedAdapt is underpinned by an offloading technique in which the layers of a Deep Neural Network (DNN) model can be offloaded from the device to an edge server to alleviate the computational burden of training on the device. To reduce the impact of stragglers, FedAdapt incorporates a reinforcement learning approach to automatically identify the layers that need to be offloaded from the device to the edge. FedAdapt further optimizes the reinforcement learning approach to develop different offloading strategies for each device while accounting for changing network bandwidth. A clustering technique is implemented to rapidly generate the offloading strategy.
This work was sponsored by funds from Rakuten Mobile, Japan. The last author was also supported by a Royal Society Short Industry Fellowship.
- AM (Accepted Manuscript)
CitationWu , D , Ullah , R , Harvey , P , Kilpatrick , P , Spence , I & Varghese , B 2022 , ' FedAdapt : adaptive offloading for IoT devices in federated learning ' , IEEE Internet of Things Journal . https://doi.org/10.1109/jiot.2022.3176469
Cardiff Met Affiliation
- Cardiff School of Technologies
Cardiff Met AuthorsRehmat Ullah
- © The Publisher
Publisher Rights StatementCopyright © 2022 IEEE. This work has been made available online in accordance with publisher policies or with permission. Permission for further reuse of this content should be sought from the publisher or the rights holder. This is the author created accepted manuscript following peer review and may differ slightly from the final published version. The final published version of this work is available at https://doi.org/10.1109/JIOT.2022.3176469