Physical Sciences and Mathematics Commons^™

Extracting Dnn Architectures Via Runtime Profiling On Mobile Gpus, Dong Hyub Kim

Masters Theses

Due to significant investment, research, and development efforts over the past decade, deep neural networks (DNNs) have achieved notable advancements in classification and regression domains. As a result, DNNs are considered valuable intellectual property for artificial intelligence providers. Prior work has demonstrated highly effective model extraction attacks which steal a DNN, dismantling the provider’s business model and paving the way for unethical or malicious activities, such as misuse of personal data, safety risks in critical systems, or spreading misinformation. This thesis explores the feasibility of model extraction attacks on mobile devices using aggregated runtime profiles as a side-channel to leak …

Time Series Anomaly Detection Using Generative Adversarial Networks, Shyam Sundar Saravanan

Masters Theses

"Anomaly detection is widely used in network intrusion detection, autonomous driving, medical diagnosis, credit card frauds, etc. However, several key challenges remain open, such as lack of ground truth labels, presence of complex temporal patterns, and generalizing over different datasets. In this work, we propose TSI-GAN, an unsupervised anomaly detection model for time-series that can learn complex temporal patterns automatically and generalize well, i.e., no need for choosing dataset-specific parameters, making statistical assumptions about underlying data, or changing model architectures. To achieve these goals, we convert each input time-series into a sequence of 2D images using two encoding techniques with …

Machine Learning Modeling Of Polymer Coating Formulations: Benchmark Of Feature Representation Schemes, Nelson I. Evbarunegbe

Masters Theses

Polymer coatings offer a wide range of benefits across various industries, playing a crucial role in product protection and extension of shelf life. However, formulating them can be a non-trivial task given the multitude of variables and factors involved in the production process, rendering it a complex, high-dimensional problem. To tackle this problem, machine learning (ML) has emerged as a promising tool, showing considerable potential in enhancing various polymer and chemistry-based applications, particularly those dealing with high dimensional complexities.

Our research aims to develop a physics-guided ML approach to facilitate the formulations of polymer coatings. As the first step, this …

Intrusion Detection: Machine Learning Techniques For Software Defined Networks, Jacob S. Rodriguez

Masters Theses

In recent years, software defined networking (SDN) has gained popularity as a novel approach towards network management and architecture. Compared to traditional network architectures, this software-based approach offers greater flexibility, programmability, and automation. However, despite the advantages of this system, there still remains the possibility that it could be compromised. As we continue to explore new approaches to network management, we must also develop new ways of protecting those systems from threats. Throughout this paper, I will describe and test a network intrusion detection system (NIDS), and how it can be implemented within a software defined network. This system will …

Interstice, Shravan Rao

Masters Theses

When I was about three years old, I distinctly remember being too small to see what was on top of the table. A couple of years later, when I could see those objects, I thought the world around me had grown smaller. In a way, it did, as I experienced, lived, captured, remembered, and shared the space repeatedly. This sense of the world shrinking was exaggerated during the Covid-19 pandemic, allowing new behaviours and modes of interaction to emerge. Continually shaping our modern lives, virtual technologies redefine how we access and share information and stories or even explore new places. …

Detecting Ai Generated Text Using Neural Networks, Jesus Guerrero

Masters Theses

For humans, distinguishing machine generated text from human written text is men- tally taxing and slow. NLP models have been created to do this more effectively and faster. But, what if some adversarial changes have been added to the machine generated text? This thesis discusses this issue and text detectors in general.

The primary goal of this thesis is to describe the current state of text detectors in research and to discuss a key adversarial issue in modern NLP transformers. To describe the current state of text detectors a Systematic Literature Review was done on 50 relevant papers to machine-centric …

Survey Of Input Modalities In The Western World, John Ezat Sadik

Masters Theses

Having your account compromised can lead to serious complications in your life. One
way accounts become compromised is through the security risks associated with weak
passwords and reused passwords [22,23]. In this thesis, we seek to understand how
entering passwords on non-PC devices contributes to the problems of weak and reused
passwords. To do so, we conducted a survey that was distributed to people in the
the Western World. In our survey results, we found that users commented about
how the current password model was not created with a variety of device types in
mind, which created frustrations and complexity …

A Machine Learning Approach For Predicting Clinical Trial Patient Enrollment In Drug Development Portfolio Demand Planning, Ahmed Shoieb

Masters Theses

One of the biggest challenges the clinical research industry currently faces is the accurate forecasting of patient enrollment (namely if and when a clinical trial will achieve full enrollment), as the stochastic behavior of enrollment can significantly contribute to delays in the development of new drugs, increases in duration and costs of clinical trials, and the over- or under- estimation of clinical supply. This study proposes a Machine Learning model using a Fully Convolutional Network (FCN) that is trained on a dataset of 100,000 patient enrollment data points including patient age, patient gender, patient disease, investigational product, study phase, blinded …

Interactive Data Analysis Of Multi-Run Performance Data, Vanessa Lama

Masters Theses

Multi-dimensional performance data analysis presents challenges for programmers, and users. Developers have to choose library and compiler options for each platform, analyze raw performance data, and keep up with new technologies. Users run codes on different platforms, validate results with collaborators, and analyze performance data as applications scale up. Site operators use multiple profiling tools to optimize performance, requiring the analysis of multiple sources and data types. There is currently no comprehensive tool to support the structured analysis of unstructured data, when holistic performance data analysis can offer actionable insights and improve performance. In this work, we present thicket, a …

Computer Vision In Adverse Conditions: Small Objects, Low-Resoltuion Images, And Edge Deployment, Raja Sunkara

Masters Theses

"Computer vision based on deep learning is an essential field that plays a significant role in object detection, image classification, semantic segmentation, instance segmentation, and other applications. However, these models face significant challenges in adverse conditions, such as small objects, low-resolution images, and edge deployment. These challenges limit the accuracy and efficiency of computer vision algorithms, making it difficult to obtain reliable results.

The primary objective of this thesis is to assess the performance of deep learning- based computer vision models in challenging conditions and provide viable solutions to overcome the obstacles. The study will specifically address three key challenges, …

Dynamic Discounted Satisficing Based Driver Decision Prediction In Sequential Taxi Requests, Sree Pooja Akula

Masters Theses

"Ridesharing platforms rely on connecting available taxi drivers to potential passengers to maximize their revenue. However, predicting the stopping decision made by every driver, i.e., the final task performed during a given day, is crucial to achieving this goal. Unfortunately, little research has been done on predicting drivers’ stopping decisions, especially when they deviate from expected utility maximization behavior. This research proposes a Dynamic Discounted Satisficing (DDS) heuristic to model and learn the task at which human agents will stop working for that day, assuming that the human agents are taking sequential decisions based on their preference order. We apply …

Mat: Genetic Algorithms Based Multi-Objective Adversarial Attack On Multi-Task Deep Neural Networks, Nikola Andric

Masters Theses

"Vulnerability to adversarial attacks is a recognized deficiency of not only deep neural networks (DNNs) but also multi-task deep neural networks (MT-DNNs) that attracted much attention in the past few years. To the best of our knowledge, all multi-task deep neural network adversarial attacks currently present in the literature are non-targeted attacks that use gradient descent to optimize a single loss function generated by aggregating all loss functions into one. On the contrary, targeted attacks are sometimes preferred since they give more control over the attack. Hence, this paper proposes a novel targeted multi-objective adversarial ATtack (MAT) based on genetic …

Maximising Social Welfare In Selfish Multi-Modal Routing Using Strategic Information Design For Quantal Response Travelers, Sainath Sanga

Masters Theses

"Traditional selfish routing literature quantifies inefficiency in transportation systems with single-attribute costs using price-of-anarchy (PoA), and provides various technical approaches (e.g. marginal cost pricing) to improve PoA of the overall network. Unfortunately, practical transportation systems have dynamic, multi-attribute costs and the state-of-the-art technical approaches proposed in the literature are infeasible for practical deployment. In this paper, we offer a paradigm shift to selfish routing via characterizing idiosyncratic, multiattribute costs at boundedly-rational travelers, as well as improving network efficiency using strategic information design. Specifically, we model the interaction between the system and travelers as a Stackelberg game, where travelers adopt multi-attribute …

Meta-Heuristic Optimization Techniques For The Production Of Medical Isotopes Through Special Target Design, Cameron Ian Salyer

Masters Theses

Medical isotopes are used for a variety of different diagnostic and therapeutic purposes Ruth (2008). Due to recent newly discovered applications, their production has become rapidly more scarce than ever before Charlton (2019). Therefore, more efficient and less time consuming methods are of interest for not only the industry’s demand, but for the individuals who require radio-isotope procedures. Currently, the primary source of most medical isotopes used today are provided by reactor and cyclotron irradiation techniques, followed by supplemental radio-chemical separations Ruth (2008). Up until this point, target designs have been optimized by experience, back of the envelope calculations, and …

Man-In-The-Middle Attacks On Mqtt Based Iot Networks, Henry C. Wong

Masters Theses

“The use of Internet-of-Things (IoT) devices has increased a considerable amount in recent years due to decreasing cost and increasing availability of transistors, semiconductor, and other components. Examples can be found in daily life through smart cities, consumer security cameras, agriculture sensors, and more. However, Cyber Security in these IoT devices are often an afterthought making these devices susceptible to easy attacks. This can be due to multiple factors. An IoT device is often in a smaller form factor and must be affordable to buy in large quantities; as a result, IoT devices have less resources than a typical computer. …

Accelerating Dynamical Density Response Code On Summit And Its Application For Computing The Density Response Function Of Vanadium Sesquioxide, Wileam Y. Phan

Masters Theses

This thesis details the process of porting the Eguiluz group dynamical density response computational platform to the hybrid CPU+GPU environment at the Summit supercomputer at Oak Ridge National Laboratory (ORNL) Leadership Computing Center. The baseline CPU-only version is a Gordon Bell-winning platform within the formally-exact time-dependent density functional theory (TD-DFT) framework using the linearly augmented plane wave (LAPW) basis set. The code is accelerated using a combination of the OpenACC programming model and GPU libraries -- namely, the Matrix Algebra for GPU and Multicore Architectures (MAGMA) library -- as well as exploiting the sparsity pattern of the matrices involved in …

Evaluation Of Robust Deep Learning Pipelines Targeting Low Swap Edge Deployment, David Carter Cornett

Masters Theses

The deep learning technique of convolutional neural networks (CNNs) has greatly advanced the state-of-the-art for computer vision tasks such as image classification and object detection. These solutions rely on large systems leveraging wattage-hungry GPUs to provide the computational power to achieve such performance. However, the size, weight and power (SWaP) requirements of these conventional GPU-based deep learning systems are not suitable when a solution requires deployment to so called "Edge" environments such as autonomous vehicles, unmanned aerial vehicles (UAVs) and smart security cameras.

The objective of this work is to benchmark FPGA-based alternatives to conventional GPU systems that have the …

Power System Stability Assessment With Supervised Machine Learning, Mirka Mandich

Masters Theses

Power system stability assessment has become an important area of research due to the increased penetration of photovoltaics (PV) in modern power systems. This work explores how supervised machine learning can be used to assess power system stability for the Western Electricity Coordinating Council (WECC) service region as part of the Data-driven Security Assessment for the Multi-Timescale Integrated Dynamics and Scheduling for Solar (MIDAS) project. Data-driven methods offer to improve power flow scheduling through machine learning prediction, enabling better energy resource management and reducing demand on real-time time-domain simulations. Frequency, transient, and small signal stability datasets were created using the …

Quantum Simulation Using High-Performance Computing, Collin Beaudoin, Christian Trefftz, Zachary Kurmas

Masters Theses

Hermitian matrix multiplication is one of the most common actions that is performed on quantum matrices, for example, it is used to apply observables onto a given state vector/density matrix.

ρ→Hρ

Our goal is to create an algorithm to perform the matrix multiplication within the constraints of QuEST [1], a high-performance simulator for quantum circuits. QuEST provides a system-independent platform for implementing and simulating quantum algorithms without the need for access to quantum machines. The current implementation of QuEST supports CUDA, MPI, and OpenMP, which allows programs to run on a wide variety of systems.

Neural Network Supervised And Reinforcement Learning For Neurological, Diagnostic, And Modeling Problems, Donald Wunsch Iii

Masters Theses

“As the medical world becomes increasingly intertwined with the tech sphere, machine learning on medical datasets and mathematical models becomes an attractive application. This research looks at the predictive capabilities of neural networks and other machine learning algorithms, and assesses the validity of several feature selection strategies to reduce the negative effects of high dataset dimensionality. Our results indicate that several feature selection methods can maintain high validation and test accuracy on classification tasks, with neural networks performing best, for both single class and multi-class classification applications. This research also evaluates a proof-of-concept application of a deep-Q-learning network (DQN) to …

Biochemical Assay Invariant Attestation For The Security Of Cyber-Physical Digital Microfluidic Biochips, Fredrick Eugene Love Ii

Masters Theses

“Due to the devastating global impact that infectious diseases have had, especially in developing countries, the demand for access to adequate resources to combat sickness continues to be a heavy burden. Reliable and affordable diagnostics is a vital first line of defense in fighting outbreaks and providing accurate treatment. Digital microfluidics biochips capable of running multiple diagnostic tests on a single platform are an emerging technology that are increasingly being evaluated as a viable platform for rapid diagnosis and point-of-care field deployment. Although these systems offer many benefits, processing errors are inherent. Therefore, cyber-physical digital biochips are being investigated that …

Values Of Trust In Ai In Autonomous Driving Vehicles, Ru Lian

Masters Theses

“Automation with artificial intelligence technology is an emerging field and is widely used in various industries. With the increasing autonomy, learning, and adaptability of intelligent machines such as self-driving cars, it is difficult to regard them as simple tools in human hands. At the same time, a series of problems and challenges such as predictability, interpretability, and causality arise. Trust in self-driving technology will impact the adoption and utilization of autonomous driving technology. A qualitative research methodology, Value-Focused Thinking, is used to identify the values of trust in autonomous driving vehicles and analyze the relationship between these values”--Abstract, page iii.

Metric Learning Via Linear Embeddings For Human Motion Recognition, Byoungdoo Kong

Masters Theses

We consider the application of Few-Shot Learning (FSL) and dimensionality reduction to the problem of human motion recognition (HMR). The structure of human motion has unique characteristics such as its dynamic and high-dimensional nature. Recent research on human motion recognition uses deep neural networks with multiple layers. Most importantly, large datasets will need to be collected to use such networks to analyze human motion. This process is both time-consuming and expensive since a large motion capture database must be collected and labeled. Despite significant progress having been made in human motion recognition, state-of-the-art algorithms still misclassify actions because of characteristics …

Random Search Plus: A More Effective Random Search For Machine Learning Hyperparameters Optimization, Bohan Li

Masters Theses

Machine learning hyperparameter optimization has always been the key to improve model performance. There are many methods of hyperparameter optimization. The popular methods include grid search, random search, manual search, Bayesian optimization, population-based optimization, etc. Random search occupies less computations than the grid search, but at the same time there is a penalty for accuracy. However, this paper proposes a more effective random search method based on the traditional random search and hyperparameter space separation. This method is named random search plus. This thesis empirically proves that random search plus is more effective than random search. There are some case …

Confronting Wicked Crypto: Wicked Problems, Encryption Policy, And Exceptional Access Technology, Kevin Nicholas Kredit

Masters Theses

Public debate has resumed on the topic of exceptional access (EA), which refers to alternative means of decryption intended for law enforcement use. The resumption of this debate is not a renege on a resolute promise made at the end of the 1990s “crypto war”; rather, it represents a valid reassessment of optimal policy in light of changing circumstances. The imbalance between privacy, access, and security in the context of constantly changing society and technology is a wicked problem that has and will continue to evade a permanent solution. As policymakers consider next steps, it is necessary that the technical …

Information Theory Problem Description Parser, Gary Brent Hurst

Masters Theses

Data corruption and data loss create huge problems when they occur, so naturally safeguards are usually in place to recover lost data. This often involves allowing less space for data in order to allow space for an encoding that can be used to recover any data that might be lost. The question arises, then, about how to most efficiently implement these safeguards with respect to storage, network bandwidth, or some linear combination of those two things. This work has two main goals for the information theory community: to produce an intuitive-to-use problem description parser that facilitates research in the area, …

Towards Development Of A Remote Charting System For Connected Healthcare, Alex Bodurka

Masters Theses

Health Care Providers play a crucial role in a patients well-being. While their primary role is to treat the patient, it is also vital to ensure that they can spend adequate time with the patient to create a unique treatment plan and build a personal relationship with their patients to help them feel comfortable during their treatment. Health Care Providers are frequently required to manually record patient data to track their healthcare progress during their hospital stay. However, with hospitals continuously trying to optimize their workflows, this crucial one-on-one time with the patient is often not practical.

As a solution, …

Improving Convolutional Neural Network Robustness To Adversarial Images Through Image Filtering, Natalie E. Bogda

Masters Theses

The field of computer vision and deep learning is known for its ability to recognize images with extremely high accuracy. Convolutional neural networks exist that can correctly classify 96\% of 1.2 million images of complex scenes. However, with just a few carefully positioned imperceptible changes to the pixels of an input image, an otherwise accurate network will misclassify this almost identical image with high confidence. These perturbed images are known as \textit{adversarial examples} and expose that convolutional neural networks do not necessarily "see" the world in the way that humans do. This work focuses on increasing the robustness of classifiers …

A Framework To Support Automatic Certification For Self-Adaptive Systems, Ioannis Nearchou

Masters Theses

Presently, cyber-physical systems are increasingly being integrated into societies, from the economic sector to the nuclear energy sector. Cyber-physical systems are systems that combine physical, digital, human, and other components, which operate through physical means and software. When system errors occur, the consequences of malfunction could negatively impact human life. Academic studies have relied on the MAPE-K feedback loop model to develop various system components to satisfy the self-adaptive features, such that violation of the safety requirements can be minimized. Assurance of system requirement satisfaction is argued through an industrial standard form, called an assurance case, which is usually applied …

A Privacy Evaluation Of Nyx, Savannah A. Norem

Masters Theses

For this project, I will be analyzing the privacy leakage in a certain DDoS mitigation system. Nyx has been shown both in simulation and over live internet traffic to mitigate the effects of DDoS without any cooperation from downstream ASes and without any modifications to current routing protocols. However it does this through BPG-poisoning, which can unintentionally advertise information. This project explores what the traffic from Nyx looks like and what information can be gathered from it. Specifically, Nyx works by defining a deployer/critical relationship whose traffic is moved to maintain even under DDoS circumstances, and I will be evaluating …