Digital Commons Network^™

Distributed System For Meshnet, Pratyush Reddy Gaggenapalli

Computer Science Theses

This thesis explores the integration of Meshnet models with distributed learning techniques to enhance MRI brain scan analysis, with a focus on optimizing brain tissue segmentation while maintaining secure distributed systems. Through refining Meshnet's architecture and training strategies, the goal is to enhance accuracy in identifying brain segmentations. Distributed learning strategies, particularly centralized aggregation, are investigated to enable collaborative model training while ensuring data privacy. Additionally, Coinstac is integrated for secure gradient aggregation from diverse nodes, facilitating collaborative analysis without compromising confidentiality. Implementation of a serverless architecture using public clouds extends global accessibility while upholding robust security measures. The primary …

Detection Of Brain Communities In The General Children Population, Britny Farahdel

Computer Science Theses

The fingerprint is known to be unique in every individual, and there is evidence that such individuality exists with the brain. Neuroimaging studies that research brain fingerprint patterns typically consider relationships between individuals and their brain patterns. However, there remains a question as to how such fingerprint patterns can be grouped among the general population. In this study, we implemented clustering-based methods to evaluate whether such subgrouping exists among individuals and evaluated the relationships between these clusters and individuals’ developmental, cognitive, demographical, psychological status in the Adolescent Brain and Cognitive Development study cohort. Multiplex community detection and K-means clustering revealed …

Towards Vision And Language Models Aided Object Navigation, Weizhen Liu

Computer Science Theses

In this work, we present a novel hierarchical navigation policy for object navigation that leverages both object detection models and large language models (LLMs) to enhance the interpretation and interaction with complex indoor environments. Our approach integrates object detection to accurately assess the surrounding space and employs a layout reconstruction strategy to model the environment’s structure. By defining our navigation strategy hierarchically, we separate the decision-making into long-term and short-term goals, effectively utilizing the existing concept of ”frontier-based goal selection.” We refine this method by representing frontiers through a series of observations transformed into language via object detection models. Each …

The Analysis Of Longitudinal Change Patterns In Developing Brain Using Functional And Structural Magnetic Resonance Imaging Data, Rekha Saha

Computer Science Theses

This paper introduces a novel approach to explore longitudinal changes in brain func- tional network connectivity (FNC) and gray matter (GM) in adolescents, utilizing data from the Adolescent Brain and Cognitive Development (ABCD) study. The study focuses on mul- tivariate patterns of FNC changes over a two-year period, identifying structured Functional Change Patterns (FCPs). One noteworthy finding is the strengthened functional connec- tivity between visual (VS) and sensorimotor (SM) domains as participants age. Moreover, the research highlights gender-specific variations in these patterns. This approach offers a robust means of assessing whole-brain functional changes longitudinally.

Additionally, the paper presents two complementary …

Classifying Different Cancer Types Based On Transcriptomics Data Using Machine Learning Algorithms, Eunice Olorunshola

Computer Science Theses

Cancer, a complex group of diseases characterized by abnormal cell growth, presents a significant global health challenge. Accurate classification of cancer types is vital for effective treatment and improved patient outcomes. This master’s thesis addresses the crucial issue associated with accurate cancer classification. It analyzes transcriptomic data of RNA sequencing, from six cancer subtypes (breast, colorectal, glioblastoma, hepatobiliary, lung, pancreatic) and a healthy control group. This research utilizes several machine learning algorithms to construct accurate cancer classification models using gene expression profiles and gene count data. The study incorporates advanced techniques such as feature selection, data preprocessing, and model optimization. …

Dynalay: An Introspective Approach To Dynamic Layer Selection For Deep Networks, Mrinal Mathur

Computer Science Theses

Deep learning models have become increasingly computationally intensive, requiring extensive computational resources and time for both training and inference. A significant contributing factor to this challenge is the uniform computational effort expended on each input example, regardless of its complexity. We introduce \textbf{DynaLay}, an alternative architecture that features a decision-making agent to adaptively select the most suitable layers for processing each input, thereby endowing the model with a remarkable level of introspection. DynaLay reevaluates more complex inputs during inference, adjusting the computational effort to optimize both performance and efficiency. The core of the system is a main model equipped with …

Geant4 Simulation For Radon Propagation In Soil, Mayur P. Aitavadekar

Computer Science Theses

Radon is a radioactive element in the periodic table that is generated from the decay of radium below the earth’s surface. After a generation, it travels through the soil and reaches the earth’s surface. Radon poses a radiological health risk. Therefore, the research community pays close attention to radon.

This thesis reviews the complete process of radon propagation from soil to air and studies the effects of different environmental and geological parameters which influence radon flux. A Geant4 simulation program based on C++ has been developed to study radon propagation in soil. The process of simulation setup and the results …

Semantic Structure Based Query Graph Prediction For Question Answering Over Knowledge Graph, Mingchen Li

Computer Science Theses

Building query graphs from questions is an important step in complex question answering over knowledge graph (Complex KGQA). In general, a question can be correctly answered if its query graph is built correctly and the right answer is then retrieved by issuing the query graph against the KG. Therefore, this paper focuses on query graph generation from natural language questions. Existing approaches for query graph generation ignore the semantic structure of a question, resulting in a large number of noisy query graph candidates that undermine prediction accuracies. In this paper, we define six semantic structures from common questions in KGQA …

Decentralized Harmonization Algorithm And Application To Functional Network Connectivity, Biozid Bostami

Computer Science Theses

Neuroimage data collected from multiple research institutions may incur additional source dependency, affecting the overall statistical power and leading to erroneous conclusions. This problem can be mitigated with data harmonization approaches. While open neuroimaging datasets are becoming more common, a substantial amount of data can still not be shared for various reasons. In addition, current approaches require moving all the data to a central location, which requires additional resources and creates redundant copies of the same datasets. To address these issues, we propose a decentralized harmonization approach called "Decentralized ComBat" that performs remote operations on the datasets separately without sharing …

A Systematic Development Of A Baseline Standard For Evaluating Virtual Camera Based Hand Rehabilitation Performance, Bhagirath Tallapragada

Computer Science Theses

Stroke is among the leading causes of disability, with 795,000 individuals experiencing a new or recurrent stroke each year. Upper extremity sensorimotor deficits, including diminished grip strength, are the most common long-term deficits among stroke survivors. Diminished hand function is a significant challenge for stroke survivors and health professionals. Most technology-driven approaches to address rehabilitation rely on intrusive and sensory device-based systems to analyze and assist hand rehabilitation. We strive to develop a non-intrusive system for hand rehabilitation using camera-based virtual rehabilitation. Taking steps towards this goal, in this thesis, we develop the baseline standard for hand function by analyzing …

Functional Enrichment Analysis Of Transcriptomics Data Of Breast Cancer Rna-Seq, Babatunde Bello

Computer Science Theses

The biochemical mechanism driving cancer metastasis and primary cancer invasion of new sites are still unclear as the process can be complex. The mutations in somatic cells often include mutation drivers and some passenger mutations. In this study, we have analyzed RNA-Seq datasets from primary breast cancer and metastatic lung cancer for differentially expressed gene lists to gain insight into transcriptomic profiles of the two conditions. The gene lists are analyzed for pathway and functional enrichment annotations. It is interesting to note that the top enriched pathways are major ones involving some connected cancer-related signaling processes. The enriched gene sets …

Epidemic Vulnerability Index: Vaccine Dissemination Criteria For Successful Resolution To Epidemics, Hunmin Lee

Computer Science Theses

Vaccination is the preventative measure that effectively decelerates the virus proliferation in a community. A successful response strategy toward pandemics can be obtained through selecting the optimal vaccine distribution route and minimizing the casualties by lowering the death rate and infection rate. In this thesis paper, we propose the Epidemic Vulnerability Index (EVI) that quantifies the potential risk of the subject via analyzing the COVID-19 patient dataset that correlates with mortality and social network analysis that affects the infection rate. We propagate the virus and vaccination in an Agent-based model based on real-world statistics of physical connections and features to …

A Data Visualization Framework For Esda: Understanding Pre-Diabetes And Diabetes Prevalence In Florida, Brindal Dhol

Computer Science Theses

Exploratory spatial data analysis (ESDA) is a technique for analyzing data from different geographic regions. To examine patterns, ESDA uses univariate and multivariate graphical approaches. Through a case study of diabetes and pre-diabetes prevalence in Florida, we built a novel data visualization framework for ESDA.

Diabetes is a rapidly increasing global disease that is a major global health concern with significant implications for healthcare spending. Information about the relationship between diabetes and geographical sociodemographic characteristics could assist public health programs better targeting those who are at risk. We show the regional prevalence of disease in Florida and its relationship to …

Chitransformer: Towards Reliable Stereo From Cues, Qing Su

Computer Science Theses

Current stereo matching techniques are challenged by restricted searching space, occluded regions and sheer size. While monocular depth estimation is spared from these challenges and can achieve satisfactory results with monocular cues, the lack of stereoscopic relationship renders the monocular prediction less reliable on its own especially in highly dynamic or cluttered environments. To address these issues in both scenarios, an optic-chiasm-inspired self-supervised binocular depth estimation method is proposed in thesis, wherein vision transformer with gated positional cross-attention layer is designed to enable feature-sensitive pattern retrieval between views, while retaining the extensive context information aggregated through self-attentions. This crossover design …

Feasibility And Effectiveness Analysis Of Deep Learning Vision Classification Models For Camera Communication, Abdulhaseeb Ahmed

Computer Science Theses

This thesis studies and evaluates Deep Neural Network models for data demodulation and decoding in a camera-based Visible Light Communication system. Camera communication is an emerging technology that enables communication using light beams, where information is modulated through optical transmissions from light-emitting diodes. This work conducts empirical studies to identify the feasibility and effectiveness of using Deep Learning models to improve signal reception in camera communication. The key contributions of this work include the investigation of transfer learning and customization of existing models to demodulate transmitted signals at the receiver end. The work expounds from a binary quantized system to …

Enhancing The Quality Of Denoised Image Using Guided Image Filtering, Kiruthiga Sekar

Computer Science Theses

In the current generation, the transmission of visual information serves as an essential form of communication. Often during this transmission, the digital images are corrupted by noise, which is a perversion in data that degrades a neural network's performance. Thus, denoising plays a vital role in Image Processing and the reason for the never-ending quest for an effective denoising algorithm to remove or suppress the noise while preserving the image's essential information. This paper proposes an idea of applying the Guided Image Filtering technique on a denoised image. Guided Image Filtering is an edge-preserving smoothing technique that uses the second …

Automating Graphology Using Computer Vision, Yashaswini Hosaguthi Vishwanath

Computer Science Theses

Graphology is the science of studying an individual's personality traits through handwriting analysis. In this thesis, we have automated the graphology process, particularly automating the pattern analysis of the handwriting and inference of the personality traits. The thesis is based off computer vision techniques to build a pipeline for automated graphology using handwritten text, camera and a microcomputing device. In this work, we consider the intricate details of a handwriting sample, like the size and slant variations, the various patterns formed in the writing of the text as visual features for computer vision training and processing. Our experimental analysis on …

Cliquesnv: Gui And Deployment On Galaxy, Sakshitha Channadi

Computer Science Theses

Next Generation Sequencing (NGS) is a powerful technology that has enabled sequencing of thousands to millions of DNA molecules simultaneously. NGS can be applied to any species and source of DNA including viral genomes. CliqueSNV is a novel reference-based tool for reconstruction of viral variants from NGS data. In this thesis, we present a graphical user interface for CliqueSNV installed on Galaxy and as a standalone application such that microbiologists can use the tool without having to type any command line instructions and can also run multiple experiments and include it into pipelines.

Intelligent Vehicular Perception Of Non-Line-Of-Sight Environment Using Visible Light Communication With Stereo Camera, Vignesh Varadarajan

Computer Science Theses

This thesis explores the use of stereo cameras to perceive immediate and non-line of sight roadway environments of a vehicle. The proposed system enables a ``see-through-the-vehicle-in-front'', functionality by combining scene perception with vehicle-vehicle communication. The fundamental idea of this work is to develop robust scene perception outcomes that can be communicated to other vehicles in the vicinity, potentially using brake light to transmit and decode using cameras, conceptually similar to visible light communications. Through experimental evaluation of the prototype system, this work presents a proof-of-concept of non-line-of-sight (NLOS) perception in vehicles.

Active Noise Cancellation Of Drone Propeller Noise Through Waveform Approximation And Pitch-Shifting, Michael Narine

Computer Science Theses

The use of drones introduces the problem of noise pollution due to the audio noise generated from its propeller rotations. To mitigate the noise pollution from drone propellers, this thesis explores a method of using active noise cancellation ANC. This thesis hypothesizes that by analyzing the waveform of the drone propeller noise, an approximated wave function can be produced and used as an anti-noise signal that can effectively nullify the drone noise. In order to align the phase of the anti-noise signal to maximize drone noise reduction, this thesis presents a signal pitch-shifting approach, to guide areas of destructive interference …

Data Preprocessing For Haplotype Calling From Viral Ngs Data, Sai Sudheep Reddy Kaidapuram

Computer Science Theses

For viral outbreaks like the recent COVID-19 outbreak, medical professionals in many areas require to know who infected whom, which variants are drug resistant and what therapy should be selected. To answer these questions, it is necessary to identify viral variants (haplotypes and SNP’s) in patients. A haplotype refers to a combination of alleles or a set of single nucleotide polymorphisms (SNPs) found on the same chromosome. This thesis describes the development and assessment of several pipelines and tools for viral NGS and read data analysis and the effect on the accuracy of the haplotype identification.

Machine Learning And Deep Learning To Predict Cross-Immunoreactivity Of Viral Epitopes, Zahra Tayebi

Computer Science Theses

Due to the poor understanding of features defining cross-immunoreactivity among heterogeneous epitopes, vaccine development against the hepatitis C virus (HCV) is trapped. The development of vaccines against HCV and human immunodeficiency virus, which are highly heterogeneous viruses (HIV) is significantly vulnerable due to variant-specific neutralizing immune responses. The novel vaccine strategies are based on some assumptions such as immunological specificity which is strongly linked to the epitope primary structure, by increasing genetic difference between epitopes cross-immunoreactivity (CR) will decline [1]. In this study first, we checked the hamming distance and statistic evaluation associating HVR1 sequence and CR based on the …

Privacy Recommendations For Future Distributed Control Systems, Wasfi Momen

Computer Science Theses

As the role of privacy becomes more established in research, new questions and implementations trickle into the Distributed Control Systems (DCS) space focusing on privacy-preserving tools. In the near future, standards will have to include measures to protect the privacy of various objects, people, and systems in DCS plants. Building a privacy framework capable of meeting the needs of DCS applications and compatible with current standards to protect against intellectual theft and sabotage is the primary aspect for DCS. By identifying the lack of privacy protections in the current standards, detailing requirements for the privacy, and proposing suitable technologies we …

Deep Neural Networks To Denoise Images, Nikhita Kokkirala

Computer Science Theses

Deep Neural Networks have the tendency to be easily fooled and research has shown that these neural networks consider unrecognizable images as recognizable. And, essentially this could lead to a lot of problems in secure systems based on image recognition. As, a solution to this problem, this paper a denoising architecture that extracts the noise from an image thus enabling the neural network to accurately label an image.

Connected-Dense-Connected Subgraphs In Triple Networks, Dhara Shah

Computer Science Theses

Finding meaningful communities - subnetworks of interest within a large scale network - is a problem with a variety of applications. Most existing work towards community detection focuses on a single network. However, many real-life applications naturally yield what we refer to as Triple Networks. Triple Networks are comprised of two networks, and the network of bipartite connections between their nodes. In this paper, we formulate and investigate the problem of finding Connected-Dense-Connected subgraph (CDC), a subnetwork which has the largest density in the bipartite network and whose sets of end points within each network induce connected subnetworks. These patterns …

Deep Learning In Chemistry And Computer-Go, Mengyuan Zhu

Computer Science Theses

Deep learning a research field in artificial intelligence and also a fast-growing technology in helping human in different directions. This thesis will focus on two of its usages: chemistry and computer go. In the two fields, deep learning achieves state of art accuracy in prediction and game playing ability.

Exploring Parallel Efficiency And Synergy For Max-P Region Problem Using Python, Viney Sindhu

Computer Science Theses

Given a set of n areas spatially covering a geographical zone such as a province, forming contiguous regions from homogeneous neighboring areas satisfying a minimum threshold criterion over each region is an interesting NP-hard problem that has applications in various domains such as political science and GIS. We focus on a specific case, called Max-p regions problem, in which the main objective is to maximize the number of regions while keeping heterogeneity in each region as small as possible. The solution is broken into two phases: Construction phase and Optimization phase. We present a parallel implementation of the Max-p problem …

Stock Marketing Prediction Using Narx Algorithm, Enas Alkhoshi

Computer Science Theses

Computational technologies have offered faster and efficient solutions to financial sector. In the financial market, the advancements in computational field have been achieved by the use of neural networks and machine learning that delivered a number of financial tools. Thus, in this thesis, we aim to predict the stock index marketing for the “Dow Jones” index by using deep learning algorithms. We propose a model based on an adaptive NARX neural network to predict the closing price of a moderately stable market. In our model, non-linear auto regressive exogenous input model inserts delays into the input as well as the …

Autonomous Uav Path Planning For Wildfire Data Collection, John Bent

Computer Science Theses

Teams of unmanned aerial vehicles (UAVs) can be employed in the real-time collection of data for dynamic data-driven simulations of wildfires to provide more accurate simulation predictions. The goal of this thesis was to identify and simulate path planning algorithms and inter-UAV negotiation protocols that best govern and divide responsibility among a team of autonomous UAVs continuously observing the perimeter of a wildfire containing sections of varying importance – ranging from low-value to high-value target areas. Several benchmark algorithms were implemented as well to measure the effectiveness of the new algorithms. The performance of each algorithm was measured by simulating …

Identifying Mavens In Social Networks, Hussah Albinali

Computer Science Theses

This thesis studies social influence from the perspective of users' characteristics. The importance of users' characteristics in word-of-mouth applications has been emphasized in economics and marketing fields. We model a category of users called mavens where their unique characteristics nominate them to be the preferable seeds in viral marketing applications. In addition, we develop some methods to learn their characteristics based on a real dataset. We also illustrate the ways to maximize information flow through mavens in social networks. Our experiments show that our model can successfully detect mavens as well as fulfill significant roles in maximizing the information flow …