Federated Learning Model Aggregation in Heterogeneous Aerial and Space Networks

Dong, Fan

Federated Learning Model Aggregation in Heterogeneous Aerial and Space Networks

dc.contributor.advisor	Drew, Steve
dc.contributor.advisor	Leung, Henry
dc.contributor.author	Dong, Fan
dc.contributor.committeemember	Drew, Steve
dc.contributor.committeemember	Leung, Henry
dc.contributor.committeemember	Ye, Qiang
dc.contributor.committeemember	Wang, Mea
dc.date.accessioned	2024-10-10T18:40:17Z
dc.date.available	2024-10-10T18:40:17Z
dc.date.issued	2024-10-09
dc.description.abstract	Federated learning offers a promising solution for overcoming the challenges of networking and data privacy in aerial and space networks by harnessing large-scale private edge data and computing resources from drones, balloons, and satellites. Although existing research has extensively explored optimizing the learning process, improving computing efficiency, and reducing communication overhead, statistical heterogeneity remains a substantial challenge for federated learning optimization. While state-of-the-art algorithms have made progress, they often overlook diversity heterogeneity and fail to significantly improve performance in high-degree label heterogeneity conditions. In this thesis, statistical heterogeneity is further dissected into two categories: diversity heterogeneity and label heterogeneity, allowing for a more nuanced analysis. It also emphasizes the importance of addressing both diversity heterogeneity and high-degree label heterogeneity in aerial and space network applications. A theoretical analysis is provided to guide optimization in these two challenging scenarios. To tackle diversity heterogeneity, the WeiAvgCS algorithm is introduced to accelerate federated learning convergence. This algorithm employs weighted aggregation and client selection based on an estimated diversity measure, termed projection, enabling WeiAvgCS to outperform other benchmarks without compromising privacy. For high-degree label heterogeneity, the FedBalance algorithm is proposed, utilizing the label distribution information of each client. A novel metric, termed relative scarcity, is introduced to determine the aggregation weights assigned to clients. During the training process, fully homomorphic encryption is employed to protect clients’ label distributions. Additionally, two communication protocols are designed to facilitate training across different scenarios. Extensive experiments were conducted, demonstrating the effectiveness of WeiAvgCS and FedBalance in addressing the research gaps in diversity heterogeneity and high-degree label heterogeneity.
dc.identifier.citation	Dong, F. (2024). Federated learning model aggregation in heterogeneous aerial and space networks (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.
dc.identifier.uri	https://hdl.handle.net/1880/119967
dc.language.iso	en
dc.publisher.faculty	Schulich School of Engineering
dc.publisher.institution	University of Calgary
dc.rights	University of Calgary graduate students retain copyright ownership and moral rights for their thesis. You may use this material in any way that is permitted by the Copyright Act or through licensing that has been assigned to the document. For uses that are not allowable under copyright legislation or licensing, you are required to seek permission.
dc.subject	Federated Learning
dc.subject	Heterogeneity
dc.subject	Weighted Aggregation
dc.subject	Communication
dc.subject	Aerial and Space Networks
dc.subject.classification	Artificial Intelligence
dc.subject.classification	Computer Science
dc.title	Federated Learning Model Aggregation in Heterogeneous Aerial and Space Networks
dc.type	master thesis
thesis.degree.discipline	Engineering – Electrical & Computer
thesis.degree.grantor	University of Calgary
thesis.degree.name	Master of Science (MSc)
ucalgary.thesis.accesssetbystudent	I do not require a thesis withhold – my thesis will have open access and can be viewed and downloaded publicly as soon as possible.