Doctoral Dissertations

Keywords and Phrases

Cryptography; Federated Learning; Machine Learning; Satellite Communication; Security

Abstract

"The advancement of satellite technology has enabled the launch of small satellites equipped with high-resolution cameras into low Earth orbit (LEO), enabling the collection of extensive Earth data for training AI models. However, the conventional approach of downloading satellite-related data to a ground station (GS) for training a centralized machine learning (ML) model faces significant challenges. Firstly, the transmission of raw data raises security and privacy concerns, especially in military applications. Secondly, the download bandwidth is limited, which puts a stringent limit on image transmissions to the GS. Lastly, LEO satellites have sporadic visibility with the GS, and orbit the Earth in irregular patterns, thus providing very limited time windows for data exchange between satellites and GS.

The objective of this series of research is to tackle the above challenges by introducing federated learning (FL) and tailoring it to LEO networks. FL, a recent advancement, represents a cutting-edge distributed learning paradigm that prioritizes privacy. In FL, each client (a satellite in our case) trains its own ML model onboard without sharing raw data with any other entity. However, integrating FL into LEO networks is confronted with substantial challenges: (i) transmitting models on insecure communication channels faces security and privacy issues like model inversion, membership inference, poisoning, and persisting replay attacks, (ii) LEO satellites have limited computation and storage capabilities that hinder them from training ML models onboard, and (iii) the FL training process requires a large number of rounds, and hence because of the sporadic and irregular visibility pattern mentioned above, it can take several days or even weeks for the FL process to converge.

This dissertation proposes various approaches that (i) enable LEO satellites to run FL despite their limited computing capability, (ii) safeguard the transmitted models and their associated data against potential adversarial attackers and eavesdroppers, and (iii) expedite FL convergence within LEO networks to complete in less than two hours, under challenging satellite-GS communication conditions, while achieving competitive model accuracy" -- Abstract, p. iv

Advisor(s)

Luo, Tony T.

Committee Member(s)

Madria, Sanjay Kumar
Nadendla, V. Sriram Siddhardh
Tripathy, Ardhendu S.
Alsharoa, Ahmad

Department(s)

Computer Science

Degree Name

Ph. D. in Computer Science

Publisher

Missouri University of Science and Technology

Publication Date

Summer 2024

Pagination

xviii, 213 pages

Note about bibliography

Includes_bibliographical_references_(pages 50, 90, 122, 170 & 196-198)

Rights

©2024 Mohamed Elmahallawy , All Rights Reserved

Document Type

Dissertation - Open Access

File Type

text

Language

English

Thesis Number

T 12380

Electronic OCLC #

1460025279

Download

Share

 
COinS