Digital Commons Network^™

Optimization Of Learning Algorithms In Neuromorphic Computing Systems., Oyinpere S. Ameli

Masters Theses

Spiking Neural Networks (SNNs) are a type of artificial neural network that aim to more closely mimic the data processing processes observed in biological neural systems. However, one major challenge in training these networks has been their non-differentiable nature, which makes it difficult to apply traditional gradient-based learning techniques. Different approaches have been proposed to address this challenge, ranging from supervised learning - largely inspired by error backpropagation in Deep Neural Networks - to unsupervised learning, which closely emulates biological learning approaches such as spike-timing dependent plasticity (STDP). Neuromorphic hardware platforms such as Intel's Loihi offer programmable plasticity that allows …

Incorporating Ai-Assisted Sensing Into The Metaverse: Opportunities For Interactions, Esports, And Security Enhancement, Yi Wu

Doctoral Dissertations

With the rapid growth and development of Virtual Reality (VR) and Augmented Reality (AR), extensive research has been carried out in the domain of the Metaverse, including immersive gaming, human-computer interaction, eSports, and the associated security & privacy concerns.

My research explores the potential of incorporating Artificial Intelligence (AI)-assisted sensing technologies to facilitate a more immersive, convenient, authentic, and secure virtual experience. This dissertation mainly focus on the following topics: (1) how to perform facial expression tracking to improve the users' awareness in the Metaverse; (2) fitness tracking for immersive eCycling; (3) running gait analysis for immersive indoor running, and …

Codont5: A Multi-Task Codon Language Model For Species-To-Species Translation, Ashley N. Babjac

Doctoral Dissertations

DNA (DeoxyriboNucleic Acid) carries the genetic information for the biological processes and function of all organisms. It is composed of nucleotides, which can be grouped into 3-mer triplets called codons. It is well known that codons encoding the same amino acid, referred to as "synonymous" codons, are selected with differing frequencies between organisms. Prior research has revealed there are codons used with much higher frequency than others, causing to them being "preferred" in highly expressed genes. This has led to the development of multiple computational models that do a good job predicting gene expression in some protein-coding genes; however, their …

Understanding Traits To Support Crowdworkers' Flexibility, Senjuti Dutta

Doctoral Dissertations

Crowdworkers are drawn to the profession in part due to the flexibility it affords. However, the current design of crowdsourcing platforms limits this flexibility. Therefore, it is important to support the overall flexibility of crowdworkers. Incorporating a variety of device types in the workflow plays an important role in supporting the flexibility of crowdworkers, however each device type requires a tailored workflow. The standard workflow of crowdworkers consists of stages of work such as managing and completing tasks. I hypothesize that different devices will have unique traits for task completion and task management. Therefore in this dissertation, I explore what …

Enabling Reproducibility, Scalability, And Orchestration Of Scientific Workflows In Hpc And Cloud-Converged Infrastructure, Paula Fernanda Olaya

Doctoral Dissertations

Scientific communities across different domains increasingly run complex workflows for their scientific discovery. Scientists require that these workflows ensure robustness; where workflows must be reproducible, scale in performance; and exhibit trustworthiness in terms of the computational techniques, infrastructures, and people. However, as scientists leverage advanced techniques (big data analytics, AI, and ML) and infrastructure (HPC and cloud), their workflows grow in complexity, leading to new challenges in scientific computing; hindering robustness.

In this dissertation, we address the needs of diverse scientific communities across different fields to identify three main challenges that hinder the robustness of workflows: (i) lack of traceability, …

General Relativistic Gravity In Core-Collapse Supernova Simulations, James Nicholas Roberts Ii

Doctoral Dissertations

Core-collapse supernovae (CCSNe) are some of the most extreme and complex phenomena in the universe. The toolkit for high-order neutrino-radiation hydrodynamics (thornado) is being developed to simulate CCSNe which will provide insight into the mechanisms underlying these events. The thornado framework is a collection of modules used to calculate the effects of gravity, hydrodynamics, neutrino transport, and nuclear physics through the Weaklib equation of state table. This dissertation will present the development of the Poseidon code, which provides the general relativistic gravity solver for the thornado framework.

The Poseidon code solves for the general relativistic metric using the xCFC formulation …

Koopman-Inspired Proximal Policy Optimization (Kippo), Andrei Cozma

Masters Theses

Reinforcement Learning (RL) has made significant strides in various domains, yet developing effective control policies for environments with complex, nonlinear dynamics remains a challenge, particularly for policy gradient methods. These methods often struggle due to high-variance in gradient estimates, non-convex optimization landscapes, and sample inefficiency, resulting in unstable learning, suboptimal policies, and trade-offs between performance and reproducibility. The quest for more robust, stable, and effective methods has led to numerous innovations and remains a critical area of research. Proximal Policy Optimization (PPO) has gained popularity in recent years due to its balance in performance, training stability, and computational efficiency. In …

Extending Application Runtime Systems For Effective Data Tiering On Complex Memory Platforms, Brandon Kammerdiener

Doctoral Dissertations

Computing platforms that package multiple types of memory, each with their own performance characteristics, are quickly becoming mainstream. To operate efficiently, heterogeneous memory architectures require new data management solutions that are able to match the needs of each application with an appropriate type of memory. As the primary generators of memory usage, applications create a great deal of information that can be useful for guiding memory tiering, but the community still lacks tools to collect, organize, and leverage this information effectively. To address this gap, this work introduces a novel software framework that collects and analyzes object-level information to guide …

Enhancing Code Portability, Problem Scale, And Storage Efficiency In Exascale Applications, Nigel Tan

Doctoral Dissertations

The growing diversity of hardware and software stacks adds additional development challenges to high-performance software as we move to exascale systems. Re- engineering software for each new platform is no longer practical due to increasing heterogeneity. Hardware designers are prioritizing AI/ML features like reduced precision that increase performance but sacrifice accuracy. The growing scale of simulations and the associated checkpointing frequency exacerbate the I/O overhead and storage cost challenges already present in petascale systems. Moving forward, the community must address performance portability, precision optimization, and data deduplication challenges to ensure that exascale applications efficiently deliver scientific discovery. In this dissertation, …

Neural-Network-Based Detection Of Radiopharmaceutical Extravasation In Pet/Ct Data, Elijah D. Berberette

Masters Theses

The immediate identification of PET/CT radiopharmaceutical extravasation can eliminate many adverse effects such as misdiagnosis and improper therapy. Radiopharmaceutical extravasation is the leakage of an injected radiotracer from the patient’s intended vein into surrounding tissues. The detection of this phenomenon often requires the use of an external monitoring device; due to a lack of robust visual features that can provide indication that it has occurred. In this thesis, the feasibility of using neural networks trained on PET/CT data to identify extravasation is explored. This approach begins with a novel preprocessing methodology that automatically extracts body weight normalized standard uptake values …

Enhancing Security And Usability In Password-Based Web Systems Through Standardized Authentication Interactions, Anuj Gautam

Doctoral Dissertations

Password-based authentication is the predominant method for securing access on the web, yet it is fraught with challenges due to the web’s lack of inherent design for authentication. Password managers have emerged as auxiliary tools to assist users in generating, storing, and inputting passwords more securely and efficiently. But both the browser and the server are oblivious of the password manager’s presence, leading to usability and security issues. However, because the web wasn’t originally built to accommodate password-based authentication, password managers serve as a temporary fix and encounter several usability and security problems that limit their widespread use. This dissertation …

Graph-Based And Anomaly Detection Learning Models For Just-In-Time Defect Prediction, Aradhana Soni

Doctoral Dissertations

Efficiently identifying and resolving software defects is essential for producing high quality software. Early and accurate prediction of these defects plays a pivotal role in maintaining software quality. This dissertation focuses on advancing software defect prediction methodologies and practical applications by incorporating graph-based learning techniques and generative adversarial-based anomaly detection techniques. First, we present a novel approach to software defect prediction by introducing a graph-based defect ratio (GDR). This innovative metric leverages the intricate graph structure that captures the interdependencies among developers, commits, and repositories, offering a promising alternative to standard traditional features. This study highlights the potential for graph-based …

Stability Of Quantum Computers, Samudra Dasgupta

Doctoral Dissertations

Quantum computing's potential is immense, promising super-polynomial reductions in execution time, energy use, and memory requirements compared to classical computers. This technology has the power to revolutionize scientific applications such as simulating many-body quantum systems for molecular structure understanding, factorization of large integers, enhance machine learning, and in the process, disrupt industries like telecommunications, material science, pharmaceuticals and artificial intelligence. However, quantum computing's potential is curtailed by noise, further complicated by non-stationary noise parameter distributions across time and qubits. This dissertation focuses on the persistent issue of noise in quantum computing, particularly non-stationarity of noise parameters in transmon processors. It …

Easier Air Alert Platform: A Design And Approach To Creating A Distributed Air Quality Monitoring And Alert System, Bryceton Bible

Masters Theses

This thesis presents the design approach, development and implementation of the Elders Alert System for Imminent Environmental Risk (EASIER) project, an air quality monitoring and alert system aiming to improve the health and wellness of under-served elder communities, as a part of the Tennessee Valley Authority Connected Communities initiative for Environmental Justice. The EASIER project provides homes with a fully integrated, connected system capable of real-time air quality monitoring, notifications and descriptions of potential air quality risks, and educational material to empower these community members to take charge of their own air health. Further, EASIER aims to inform relevant family/friends …

Mapping Arbitrary Spiking Neural Networks To The Ravens Neuroprocessor, Jongheon Park

Masters Theses

In neuromorphic computing, a hardware implementation of a spiking neural network is used to provide improved speed and power efficiency over simulations of the networks on a traditional Von Neumann architecture. These hardware implementations employ bio-inspired architecture usually consisting of artificial neurons and synapses implemented in either analog, digital, or mixed-signal circuits. Since these hardware spiking neural networks are designed to support arbitrary networks under the constraints imposed by the available hardware resource, they have to be programmed by off-chip software with awareness of those constraints. The TENNLab research group at the University of Tennessee, Knoxville has recently developed the …

Genetic Algorithm Optimization Of Experiment Design For Targeted Uncertainty Reduction, Alexander Amedeo Depillis

Masters Theses

Nuclear cross sections are a set of parameters that capture probability information about various nuclear reactions. Nuclear cross section data must be experimentally measured, and this results in simulations with nuclear data-induced uncertainties on simulation outputs. This nuclear data-induced uncertainty on most parameters of interest can be reduced by adjusting the nuclear data based on the results from an experiment. Integral nuclear experiments are experiments where the results are related to many different cross sections. Nuclear data may be adjusted to have less uncertainty by adjusting them to match the results obtained from integral experiments. Different integral experiments will adjust …

Advancing Compact Modeling Of Electronic Devices: Machine Learning Approaches With Neural Networks, Mixture Density Networks, And Deep Symbolic Regression, Jack Robert Hutchins

Masters Theses

This thesis pioneers the integration of deep learning techniques into the realm of compact modeling, presenting three distinct approaches that enhance the precision, efficiency, and adaptability of compact models for electronic devices. The first method introduces a Generalized Multilayer Perception Compact Model, leveraging the function approximation capabilities of neural networks through a multilayer perception (MLP) framework. This approach utilizes hyperband tuning to optimize network hyperparameters, demonstrating its effectiveness on a HfOx memristor and establishing a versatile modeling strategy for both single-state and multistate devices.

The second approach explores the application of Mixture Density Networks (MDNs) to encapsulate the inherent stochasticity …

Understanding Student Experiences With Tls Client Authentication, Clay A. Shubert

Masters Theses

This thesis presents a comprehensive investigation into student experiences with TLS client authentication, highlighting the usability challenges and learning curves associated with this long term key managament system. We designed a study that required future innovators in technology and security to use modern-day implementations of this certificate-based authentication system. From this study, we analyzed server logs, project reports, and survey responses from students enrolled in the applied cryptography course. We revealed significant hurdles in the initial setup and long-term key management of credentials used in TLS client authentication, emphasizing the gap between theoretical knowledge and practical implementation skills. Through quantitative …

Particle Classification Of Electromagnetic Clusters Using The Sphenix Detector, Fredrick J. Melhorn

Chancellor’s Honors Program Projects

No abstract provided.

Generalized Differentiable Neural Architecture Search With Performance And Stability Improvements, Emily J. Herron

Doctoral Dissertations

This work introduces improvements to the stability and generalizability of Cyclic DARTS (CDARTS). CDARTS is a Differentiable Architecture Search (DARTS)-based approach to neural architecture search (NAS) that uses a cyclic feedback mechanism to train search and evaluation networks concurrently, thereby optimizing the search process by enforcing that the networks produce similar outputs. However, the dissimilarity between the loss functions used by the evaluation networks during the search and retraining phases results in a search-phase evaluation network, a sub-optimal proxy for the final evaluation network utilized during retraining. ICDARTS, a revised algorithm that reformulates the search phase loss functions to ensure …

Towards Safer Code Reuse: Investigating And Mitigating Security Vulnerabilities And License Violations In Copy-Based Reuse Scenarios, David Reid

Doctoral Dissertations

Background: A key benefit of open source software is the ability to copy code to reuse in other projects. Code reuse provides benefits such as faster development time, lower cost, and improved quality. There are several ways to reuse open source software in new projects including copy-based reuse, library reuse, and the use of package managers. This work specifically looks at copy-based code reuse.

Motivation: Code reuse has many benefits, but also has inherent risks, including security and legal risks. The reused code may contain security vulnerabilities, license violations, or other issues. Security vulnerabilities may persist in projects that copy …

Exact Models, Heuristics, And Supervised Learning Approaches For Vehicle Routing Problems, Zefeng Lyu

Doctoral Dissertations

This dissertation presents contributions to the field of vehicle routing problems by utilizing exact methods, heuristic approaches, and the integration of machine learning with traditional algorithms. The research is organized into three main chapters, each dedicated to a specific routing problem and a unique methodology. The first chapter addresses the Pickup and Delivery Problem with Transshipments and Time Windows, a variant that permits product transfers between vehicles to enhance logistics flexibility and reduce costs. To solve this problem, we propose an efficient mixed-integer linear programming model that has been shown to outperform existing ones. The second chapter discusses a practical …

Towards Expressive And Versatile Visualization-As-A-Service (Vaas), Tanner C. Hobson

Doctoral Dissertations

The rapid growth of data in scientific visualization has posed significant challenges to the scalability and availability of interactive visualization tools. These challenges can be largely attributed to the limitations of traditional monolithic applications in handling large datasets and accommodating multiple users or devices. To address these issues, the Visualization-as-a-Service (VaaS) architecture has emerged as a promising solution. VaaS leverages cloud-based visualization capabilities to provide on-demand and cost-effective interactive visualization. Existing VaaS has been simplistic by design with focuses on task-parallelism with single-user-per-device tasks for predetermined visualizations. This dissertation aims to extend the capabilities of VaaS by exploring data-parallel visualization …

Generative Adversarial Game With Tailored Quantum Feature Maps For Enhanced Classification, Anais Sandra Nguemto Guiawa

Doctoral Dissertations

In the burgeoning field of quantum machine learning, the fusion of quantum computing and machine learning methodologies has sparked immense interest, particularly with the emergence of noisy intermediate-scale quantum (NISQ) devices. These devices hold the promise of achieving quantum advantage, but they grapple with limitations like constrained qubit counts, limited connectivity, operational noise, and a restricted set of operations. These challenges necessitate a strategic and deliberate approach to crafting effective quantum machine learning algorithms.

This dissertation revolves around an exploration of these challenges, presenting innovative strategies that tailor quantum algorithms and processes to seamlessly integrate with commercial quantum platforms. A …

Optimizing Collective Communication For Scalable Scientific Computing And Deep Learning, Jiali Li

Doctoral Dissertations

In the realm of distributed computing, collective operations involve coordinated communication and synchronization among multiple processing units, enabling efficient data exchange and collaboration. Scientific applications, such as simulations, computational fluid dynamics, and scalable deep learning, require complex computations that can be parallelized across multiple nodes in a distributed system. These applications often involve data-dependent communication patterns, where collective operations are critical for achieving high performance in data exchange. Optimizing collective operations for scientific applications and deep learning involves improving the algorithms, communication patterns, and data distribution strategies to minimize communication overhead and maximize computational efficiency.

Within the context of this …

Hashed Coordinate Sparse Tensor Storage With Matlab, Jama Meili Charles

Doctoral Dissertations

Tensors, or n-way arrays, are incredibly useful for storing indexable data in an arbitrary number of dimensions. Interest in tensor analysis using tensor decomposition has expanded to a variety of fields, including data mining, signal processing, computer vision, and machine learning. Tensors modelling interesting data may also be sparse, where the majority of its values are zero. These tensors can be extremely large and contain millions of entries that cannot be stored explicitly. To address this problem, various formats have arisen in the past decade to compress and compact such massive data. However, most of these existing structures are …

Reducing Communication In The Solution Of Linear Systems, Neil S. Lindquist

Doctoral Dissertations

There is a growing performance gap between computation and communication on modern computers, making it crucial to develop algorithms with lower latency and bandwidth requirements. Because systems of linear equations are important for numerous scientific and engineering applications, I have studied several approaches for reducing communication in those problems. First, I developed optimizations to dense LU with partial pivoting, which downstream applications can adopt with little to no effort. Second, I consider two techniques to completely replace pivoting in dense LU, which can provide significantly higher speedups, albeit without the same numerical guarantees as partial pivoting. One technique uses randomized …

Unsupervised Machine Learning Of Tornado-Producing Storms In The Southeastern United States, Morgan R. Steckler

Masters Theses

The east-southeastern US is uniquely affected by storm and tornado-related damages, costs, injuries, and deaths. Based on doppler radar, satellite, and modeled data, previous research sought to understand these different types of storms that produce strong tornadoes. Many approaches to storm classification are time intensive, complex, and vary significantly across the literature. The purpose of this work is to (1) explore the radar-derived data structure and spread of strong tornado-producing mesoscale storms in the east-southeastern US; (2) use K-Means unsupervised machine learning methods to elucidate clusters (storm types) and clustering attributes; and (3) assess the utility of K-Means as a …

Reduced Order Modeling And Analysis Of Cardiac Chaos, Tuhin Subhra Das

Doctoral Dissertations

Numerous physiological processes are functioning at the level of cells, tissues and organs in the human body, out of which some are oscillatory and some are non-oscillatory. Networks of coupled oscillators are widely studied in the modeling of oscillatory or rhythmical physiological processes. Phase-isostable reduction is an emerging model reduction strategy that can be used to accurately replicate nonlinear behaviors in dynamical systems for which standard phase reduction techniques fail. We apply strategies of phase reduction, or isostable reductions in biologically motivated problems like eliminating cardiac alternans to alleviate arrhythmia by applying energy-optimal, non-feedback type control solutions.

Cardiac fibrillation caused …

Fabrication, Measurements, And Modeling Of Semiconductor Radiation Detectors For Imaging And Detector Response Functions, Corey David Ahl

Doctoral Dissertations

In the first part of this dissertation, we cover the development of a diamond semiconductor alpha-tagging sensor for associated particle imaging to solve challenges with currently employed scintillators. The alpha-tagging sensor is a double-sided strip detector made from polycrystalline CVD diamond. The performance goals of the alpha-tagging sensor are 700-picosecond timing resolution and 0.5 mm spatial resolution. A literature review summarizes the methodology, goals, and challenges in associated particle imaging. The history and current state of alpha-tagging sensors, followed by the properties of diamond semiconductors are discussed to close the literature review. The materials and methods used to calibrate the …