MARS
MARS is a repository service of Mason Publishing and the Data and Digital Scholarship Services (DDSS) at the George Mason University Libraries. MARS provides enduring, stable, well-indexed access to a wide range of scholarship from the Mason community, such as Electronic Theses and Dissertations (ETDs), articles, presentations, reports, and creative work. Learn more about publishing, sharing, and preserving research data with the George Mason University Institutional Dataverse, and our other repository services.
To start publishing your content in MARS, please contact us by using our online form. Questions? Please email publish@gmu.edu.
Communities in MARS
Select a community to browse its collections.
Recent Submissions
Using The Oxford Nanopore MinION MK1B sequencer to identify human and animal DNA
(2026-05) Bishop, Robin; Wilson, Mark
Animal DNA is valuable to forensic science research and development because it lends strong evidence that can be used to potentially include or exclude a person from involvement in a crime. Oxford Nanopore sequencing is considered a highly innovative tool used for forensic applications, but present-day knowledge and documentation of animal sequencing using this device is limited. Examples of previous studies where animal species were positively identified exist; however, due to limitations of current technology and research, it has not always been accurate. This research investigates if the Oxford Nanopore MinION MK1B sequencer can correctly identify human versus animal DNA from a mixture. Using the innovative Oxford Nanopore Technology, four samples were analyzed using human DNA (NA12878) and dog DNA (GDM-150). One sample was 100% human DNA, one sample was 100% dog DNA, one was a 75% dog 25% human DNA mixture, and one was a 75% human 25% dog DNA mixture. Mixtures are evidence samples containing DNA from more than one individual. The procedure included DNA library preparation, DNA quantification, sequencing in MinKNOW, flow cell wash, and data analysis in EPI2ME software. FASTQ files from sequencing were converted to BAM files using the EPI2ME workflow wf-alignment. The BAM files were then uploaded to IGV for analysis against a reference genome of human and dog DNA. Based on analysis in IGV and the wf-metagenomics workflow, the dog DNA was the major contributor in the samples that were predominately dog, and the human DNA was the major contributor for the samples that were predominately human. This study demonstrates that ONT can be used to identify different species successfully.
Sparse K-Means Compression for Federated Machine Learning and Linear Regression Using Sketched and Quantized Predictors
(2025-05) Hill, Daniel; Kepplinger, David
The Information Age has led to the generation of vast and unquantifiable amounts of data, but technology has struggled to keep pace with the growing demand for efficient storage and transmission. Compression algorithms provide a means to reduce storage and transmission costs while preserving essential information for learning and analysis. This dissertation makes two contributions in this area: a novel compression scheme for federated learning and a statistical analysis framework regarding the use of compressed data in linear regression.
Regarding the first contribution, we propose the Sparse $k$-Means (SparK) algorithm specifically designed for Federated Learning applications. SparK compresses model parameter updates between clients and a server by combining sparsification with $k$-means clustering. Using the desired inverse compression rate as its sole hyperparameter, SparK optimizes the degree of sparsification and the number of clusters in $k$-means for each model layer to achieve the desired compression with minimal distortion. Experimental results demonstrate that SparK performs comparably or better than similar sparsification and clustering methods on a standard test bed across various compression levels.
Regarding the second contribution, we examine dithered 1-bit compression of predictors and response variables in the context of linear regression. We propose an M-estimator of the associated regression coefficients and establish its asymptotic Normality and asymptotic mean squared error (MSE). This is complemented by a non-asymptotic analysis of the MSE for three compressors: 1-bit stochastic quantization, Gaussian sketching, and their combination. High-probability upper bounds are derived for each compressor under both fixed and random design assumptions. The relative efficiency in comparison to the ordinary least squares estimator with access to uncompressed data is studied as well.
Business Plan Archive
(Roy Rosenzweig Center for History and New Media, 2002)
Business Plan Archive was an online repository for business plans and related planning documents. Webmergers.com and the University of Maryland's Robert H. Smith School of Business have built this site, in cooperation with the Center for History and New Media at George Mason University and with financial support from the Alfred P. Sloan Foundation, to collect business plans for posterity. All of the documents we collect will ultimately be deposited in the Archives and Manuscript Library at the University of Maryland, College Park, where future entrepreneurs and business researchers will have access to learn from this remarkable period of technological and organizational creativity.
Dot Com Archive
(Roy Rosenzweig Center for History and New Media, 2003)
The Dot Com Archive was a digital collecting project to gather the histories of people who worked at internet technology companies during the 1990s.
Do History
(Roy Rosenzweig Center for History and New Media, 2003)
This website was intended to show students how to piece together information about the past from the fragments that have survived to the present day, using the case study of Martha Ballard. It was originally a project of the Harvard University Film Study Center and transferred to RRCHNM in 2003.