Federated Learning Architecture to Enable Continuous Learning at the Tactical Edge for Situational Awareness
Open AccessThis research demonstrates how federated learning (FL) can be conducted across edge computing devices that are connected by limited bandwidth networks. Of interest are networks that cannot transport bulk raw data from imagery sensors across network connections and therefore any processing of imagery data must be done on-device. To make the proposed solution applicable for the Internet of Things (IoT) at the network edge, and for defense systems at the tactical edge, the size of the FL model-update message (MUM) is limited to 465 bits or less. The size of MUM is used to tailor a machine learning (ML) model architecture and a corresponding model training protocol. The model training protocol is applied differently across the system lifecycle stages to account for availability of data and expected network connectivity. The novel aspects of this research include a tailored federated learning architecture which extends systems learning into environments where inference-only ML would have typically been used. The results demonstrate the feasibility of using FL to improve upon inference-only image classification performance, while requiring significantly less computational and network resources than a traditional centralized approach. The results show the proposed federated learning architecture enables post-deployment learning to improve upon the image classification performance of an inference-only, centralized approach.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.