MS, University of Cincinnati, 2024, Engineering and Applied Science: Computer Science

The continuous threat of deepfakes, cleverly crafted deceptions masquerading as reality, necessitates cutting-edge detection methods. While there are many methods available, this project dives into the realm of fine-tuning the VGG16 convolutional neural network (CNN) and synergistically integrating Natural Language Processing (NLP) to unveil deepfake images effectively. By using the Keras API and machine learning principles, we empower the model to discern authentic images from their manipulated counterparts, drawing inspiration from real-world cases like the notorious Jennifer Aniston deepfake scam. Firstly, we establish a robust foundation for feature extraction by pre-training the VGG16 architecture on vast image datasets. Subsequently, we meticulously curate a comprehensive deepfake image dataset encompassing diverse manipulation techniques and real-world scenarios. This tailor-made dataset fuels the fine-tuning of specific VGG16 layers, accurately crafting a model with exceptional generalizability. Intriguingly. The project rigorously evaluates the fine-tuned VGG16 model's performance on unseen deepfakes through a battery of meticulous metrics, including accuracy, and loss while detecting the deepfakes. We delve into a comprehensive comparison, carefully analyzing these results not only against the baseline performance of a model I created from scratch, the untrained VGG16, the VGG16 after I applied transfer learning. This project aspires to make a significant contribution to the ongoing battle against deepfakes by showcasing the remarkable potential of fine-tuning the VGG16 that helps us in achieving superior detection accuracy. By thoroughly incorporating real-world examples and harnessing the synergistic power of CNNs, we strive to develop a robust and adaptable solution capable of combating the ever-evolving landscape of deepfakes. Ultimately, this endeavor aims to safeguard online safety and trust, mitigating the detrimental effects of deepfakes on (open full item for complete abstract)

Committee: Yizong Cheng Ph.D. (Committee Chair); William Hawkins Ph.D. (Committee Member); Jun Bai Ph.D. (Committee Member) Subjects: Computer Science

PhD, University of Cincinnati, 2023, Engineering and Applied Science: Mechanical Engineering

By integration of artificial intelligence (AI) and prognostic and health management (PHM) techniques, condition monitoring, health assessment, and maintenance of industrial systems have changed dramatically. Due to the successful implementation of intelligent data-driven approaches, these methods are gaining remarkable attention in PHM applications. With the advent of industrial AI, a new perspective has been introduced to the value and application of AI in industry. Within this scope, domain knowledge plays an important role. In the cases where the domain knowledge is low, transfer learning approaches are exploited to transfer the obtained knowledge from the source domain data to the target domain data. Due to the different working regimes and operating conditions, there exists a discrepancy between the data distribution of source and target domain datasets. Domain adaptation techniques are deployed to tackle the data distribution discrepancy. Mostly it is assumed that during the training process, a complete dataset for all health condition of the system is available. However, in practical real scenarios, providing a complete dataset, which consists of all health conditions with enough samples, is not practical. There are different factors such as cost, sensor, and maintenance limitations that prohibit the collection of complete industrial dataset. In this dissertation a novel methodology is proposed that addresses data-driven model training while domain knowledge is low. The basis of the proposed methodology is transfer learning-based using deep learning approaches. The effectiveness of the proposed architecture is evaluated on different case studies. The case studies are designed for different PHM applications. First, a consistency-based regularization approach is introduced to serve as a machine-to-machine communication term. This term helps the network find the optimum feature sub-space for fault prognosis problems. This methodology is implemented in industrial (open full item for complete abstract)

Committee: Jay Lee Ph.D. (Committee Chair); Manish Kumar Ph.D. (Committee Member); Jonathan Tan Ph.D. (Committee Member); Jing Shi Ph.D. (Committee Member); Laura Pahren Ph.D. (Committee Member) Subjects: Mechanical Engineering

Doctor of Education, Miami University, 2023, Educational Leadership

Micro-credentials have been used in the education setting for many years. Adult learners have taken micro-credential courses to improve their skills to perform their jobs successfully. The federal technology transfer professional has limited professional development opportunities to perform their jobs better. This dissertation study examined the development and evaluation of technology transfer training through micro-credentials. The results of this study concluded that, although government agencies conduct technology transfer differently, the Federal Laboratory Consortium can provide micro-credentials needed as a foundation in the form of career pathways. Surveys given to the participants revealed that they were drawn to the interactive components of the course. The study results were aligned to previous studies on adult learning, and we must take into consideration the time adults have to learn in relation to their current work obligations.

Committee: Lucian Szlizewski (Committee Co-Chair); Joel Malin (Committee Co-Chair) Subjects: Adult Education; Technology

PhD, University of Cincinnati, 2023, Engineering and Applied Science: Computer Science and Engineering

Users' privacy and data security are under unprecedented threat because of the growing use of the Internet and digital devices, one example of such threat is side-channel attacks. The side-channel attacks are a class of cyber-attacks in which the attacker tries to exploit physical side-channel information leakage to recover critical information of a user. The research on side-channel attacks has made significant progress and remains a hot topic. Deep learning based side-channel attack methods have shown advantages in many aspects. But these methods are facing critical limitations -- insufficient labeled training data and data distribution shifts, which will cause side-channel attack failure. Our research focuses on these problems and investigates two specific side-channel attacks: 1) side-channel attack over encrypted network traffic (also called website fingerprinting); 2) side-channel attack to power consumption on micro-controllers (also called side-channel attack). My main focus and contributions are 3 fold: Firstly, we studied website fingerprinting in a more real-world scenario: the attacker and the user have different network setups and website content updates frequently, which causes the well-trained model outdated in a few days, but the collection of labeled data could take more than 2 weeks. Due to this reason, the attacker cannot obtain enough labeled training data to perform the attack. Facing this challenge, we designed a novel website fingerprinting attack method based on the adversarial domain adaption technique, which can enable the attacker to perform the attack with less than 20 traces per website and achieve over 80\% accuracy when the network setup is different. Secondly, we studied the side-channel attack when the attacker cannot obtain a sufficient number of training traces, which will cause the traditional deep learning based methods to fail to recover the key. To address this limitation, we proposed a novel side-channel attack b (open full item for complete abstract)

Committee: Boyang Wang Ph.D. (Committee Chair); Wen-Ben Jone Ph.D. (Committee Member); Wenhai Sun Ph.D. (Committee Member); Nan Niu Ph.D. (Committee Member); Seokki Lee Ph.D. (Committee Member) Subjects: Computer Engineering

Doctor of Philosophy, The Ohio State University, 2023, Mechanical Engineering

Traditional design tasks and manufacturing systems often require a multitude of manual efforts to fabricate sophisticated artifacts with desired performance characteristics. Engineers typically iterate on a first principles-based model to make design decisions and then iterate once more by manufacturing the artifacts to take manufacturability into design consideration. Such manual decision-making is inefficient because it is prone to errors, labor intensive and often fails to discover process-structure-property relationships for novel materials. As robotics is an integral part of modern manufacturing, building autonomous robots with decision-making capabilities is of crucial importance for this application domain. Unfortunately, most of the robots and manufacturing systems in the industries lack such cognitive abilities. We argue that this whole process can be made more efficient by utilizing machine learning (ML) approaches, more specifically by leveraging sequential decision-making, and thus making these robots, design processes and manufacturing systems autonomous. Such data-driven decision-making has multiple benefits over traditional approaches; 1) machine learning approaches may discover interesting correlations in the data or process-structure-property relationship, 2) ML algorithms are scalable, can work with high dimensional unstructured problems and learn in highly nonlinear systems where a model is not available or feasible, 3) thousands of man-hour and extensive manual labor can be saved by building autonomous data-driven methods. Due to the sequential nature of the problem, we consider reinforcement learning (RL), a type of machine learning algorithm that can take sequential decisions under uncertainty by interacting with the environment and observing the feedback, to build autonomous manufacturing systems (AMS). Unfortunately, traditional RL is not suitable for such hardware implementation because (a) data collection for AMS is expensive and (b) tradit (open full item for complete abstract)

Committee: David Hoelzle (Advisor); Parinaz Naghizadeh (Committee Member); Jieliang Luo (Committee Member); Kira Barton (Committee Member); Michael Groeber (Committee Member) Subjects: Artificial Intelligence; Design; Mechanical Engineering; Robotics

Master of Sciences, Case Western Reserve University, 2023, EECS - Computer and Information Sciences

A central problem in applying planning to new environments is to quickly discover the characteristics of those environments. In practice, this means identifying the behaviors and properties of objects within the environment. In this thesis I investigate how to approach this problem with transfer learning for deterministic planning tasks within the Object-Oriented Markov Decision Process (OO-MDP) framework. I extend this framework with additional logical relations and describe an efficient rule learning algorithm for these domains. I then present two approaches of using the learned rules in new environments to discover the types and properties of objects under different assumptions. I show empirically that these approaches allow an agent to learn and plan in new domains with better sample complexity than if it had started from scratch.

Committee: Soumya Ray (Advisor); M. Cenk Cavusoglu (Committee Member); Michael Lewicki (Committee Member) Subjects: Computer Science

Doctor of Philosophy, The Ohio State University, 2023, Electrical and Computer Engineering

Recent decades saw a huge increase in the number of personal devices, wearables, edge devices, etc which led to increased data collection and increased connectivity at the edge. This collected data can be used to make insights about health, the economy, and business and help us make better decisions at the individual, organizational and global levels. With the proliferation of these devices, there are also numerous challenges associated with making use of these devices and the data to train useful models. The challenges could be due to privacy regulations or other constraints determined by the particular learning setup. These constraints make it difficult to extract the required insights from the data and the edge systems. The goal of this thesis is to understand these challenges or resource constraints and develop efficient algorithms that enable us to train models while adhering to the constraints. This thesis makes the following contributions: 1. Propose an efficient algorithm FedCMA for model heterogeneous Federated Learning under resource constraints, showed the convergence and generalization properties, and demonstrated the efficacy against state-of-the-art algorithms in the model heterogeneity setting. 2. Proposed a two-timescale aggregation algorithm that does not require the knowledge of the number of adversaries for defending against Byzantine adversaries in the distributed setup, proved the convergence of the algorithm, and demonstrated the defense against state-of-the-art attacks. 3. We highlight the challenges posed by resource constraints in the Offline Reinforcement Learning setup where the observation space during inference is different from the observation space during training. We propose a simple algorithm STPI (Simultaneous Transfer Policy Iteration) to train the agent to adapt to the changes in the observation space and demonstrated the effectiveness of the algorithm on MuJoCo environments against simple baselines.

Committee: Abhishek Gupta (Advisor); Ness Shroff (Advisor) Subjects: Computer Engineering; Computer Science; Electrical Engineering; Engineering

Doctor of Philosophy, The Ohio State University, 2022, Political Science

My dissertation comprises (1) the development of a machine learning framework that combines verbal and visual features together, models the intricate web of relationships between them, and extracts visual semantics, and (2) the application of a deep learning and a transfer learning framework to extract emotions from social media posts. This dissertation consists of three papers as follows. The first paper introduces a machine-learning visual framing analysis to examine the visual and verbal patterns of online news reporting and explore image-text relations in news stories. The second paper presents a machine-learning multimodal framing analysis to integrate the various types of data (e.g., image, text, and metadata) simultaneously and extract the semantic meaning from them together. The third paper is an application of a deep learning and a transfer learning to show the power of Twitter in providing fine-grained measures of real-time emotions and thereby offer a comprehensive overview of the role of emotions in voting participation. My dissertation can take into account various types of data simultaneously and extract politically meaningful semantics using computer vision, NLP, graph theory, high-dimensional statistics, and transfer learning.

Committee: Skyler Cranmer (Committee Chair); Janet Box-Steffensmeier (Committee Member); Robert Bond (Committee Member) Subjects: Communication; Computer Science; Political Science

Doctor of Philosophy, The Ohio State University, 2022, Engineering Education

This work describes the design and implementation of an online education program designed for working engineers. The program is offered through a university research center and covers topics in Finite Element Analysis (FEA). The relatively recent digital transformation of the engineering workplace requires the use of advanced computational engineering tools such as FEA, and the engineering workforce is being challenged to learn how to adapt to new tools and methods to fully realize the benefits of this digital transformation. Universities are uniquely positioned to develop and offer continuing education programs to support engineers in learning and adapting to these new tools and methods. The conceptual framework underlying this research is the Theory of Reasoned Action, to frame the understanding of how engineers engage with a continuing professional education program using a novel online delivery design. This framework relates beliefs, attitudes, and intentions to understand behaviors, in this case, participation in the online course. A learning transfer framework was used to describe the initial course design, and the Learning Transfer System Inventory (LTSI) was used to preliminarily evaluate learning transfer for the learners. The initial offering of the course was studied using a naturalistic inquiry design, investigating the experiences of the first four learners as described in their own discussion post writings. In a subsequent study which considered several course offerings in the program, course evaluation results were analyzed to determine sentiments of the learners on various aspects of the course including assignments and program participation. The course design is further described using the framework of Project Based Learning to describe the approach taken to present relevant theoretical topics which can be directly applied to modeling decision making and application. Based on the research data collected, the engineers participating in this progra (open full item for complete abstract)

Committee: Ann Christy (Advisor); Julie Aldridge (Committee Member); Prasad Mokashi (Committee Member); David Stein (Advisor) Subjects: Education; Engineering

Master of Science (M.S.), University of Dayton, 2022, Electrical and Computer Engineering

Remote sensing has seen significant attention due to advances in technology and access to data. A current challenge is classifying land regions by their usage – residential, industrial, forest, etc. Scope is very important, too large of an area would lead to multiple classes being present in one scene, and too small of an area would not contain enough contextual information to accurately determine a scene. To further complicate matters, there are multiple similar objects all present in different classes, for example trees are found in residential, forest, and park classes. Deep learning is a current technology that is successful with problems at this level of ambiguity. The most straight-forward approach to address this level of complexity is to use remote sensing images to classify land regions. However, deep learning using 2D images has its downsides, especially when analyzing aerial data, namely, it lacks 3-dimensional information such as depth. Similarly, there are also 3D deep learning architectures with different weaknesses, i.e., longer processing times and lack of intensity information. As access to processing hardware and remote sensing data continues to increase, there is a pressing need to leverage the strengths of both modalities. This can be done in one of three ways: (1) a data-level fusion, where data modalities are fused together directly; (2) a feature-level fusion, where features are fused after data modalities are processed individually; or (3) a decision-level fusion, where predictions are made using each modality independently, until, ultimately, they are fused into one final decision. In this work, we utilize a feature-level fusion because our dataset (comprised of lidar and RGB scenes) have very different types of information; after analysis, we found that each modality was better suited to different sections of our data, which we could harness using a feature-level fusion. Furthermore, to improve on these results, an accurate regist (open full item for complete abstract)

Committee: Vijayan Asari (Advisor); Andrew Stokes (Committee Member); Theus Aspiras (Committee Member); Eric Balster (Committee Member) Subjects: Artificial Intelligence; Computer Engineering; Computer Science; Remote Sensing

Doctor of Philosophy, The Ohio State University, 2022, Computer Science and Engineering

With the continuous increase of time series data, more and more research is focused on using these data to improve people's lives. On the one hand, the Markov Decision Process (MDP) is used widely in decision-making. An agent can decide the best action based on its current state. When the agent is applied to time series data, the model will help people make more informed decisions. However, state identification, which is very important in obtaining an optimal decision, has received less attention. On the other hand, with the development of deep learning, identifying the category of a time series has become more and more precise. As a result, the recognition of complex time series sequences has become the hub of public attention. In this dissertation, we focus on developing an automatic state selection using MDP and investigate the application of deep learning in recognizing time series data. We propose a method that combines decision-tree modeling and MDP to permit automatic state identification in a way that offers desirable trade-offs between simplicity and Markovian behavior. We first create a simplified definition of the host state, which becomes the response measure in our decision-tree model. Then, we fit the model in a way that weighs accuracy and interpretability. The leaves of the resulting decision-tree model become the system states. This follows, intuitively, because these are the groupings needed to predict (approximately) the system evolution. Then, we generate and apply an MDP control policy. Our motivating example is cyber vulnerability maintenance. Using the proposed methods, we predict that a Midwest university could save more than four million dollars compared to the current policy. Prechtl's general movements assessment (GMA) allows visual recognition of movement patterns in infants that, when abnormal (cramped synchronized, or CS), have very high specificity in predicting later neuromotor disorders. However, training req (open full item for complete abstract)

Committee: Rajiv Ramnath (Advisor); Ping Zhang (Committee Member); Theodore Allen (Advisor) Subjects: Artificial Intelligence; Bioinformatics; Business Costs; Computer Engineering; Computer Science

Ph.D., Antioch University, 2021, Leadership and Change

School districts spend millions of dollars each year to provide training and learning to staff working in direct and indirect service to students (National Council on Teacher Quality, 2021). This financial commitment says nothing about what is even more important: the need for school employees and the systems in which we work to serve students more effectively. Despite vast allocations of time and money and presumably best intentions for better social and academic outcomes for students, very little data exist that reflect regular transfer and application of training/learning into professional practice (Nittler et al., 2015). By and large, schools and school systems look the same today as they did 50+ years ago despite the fact that the world looks very different and so much more is known about the cognitive process and contextual contributors involved in erudition development. Teacher application of critical competencies such as cultural responsiveness, trauma informed practices, social emotional learning and basic neuroscience in the ways they conceptualize and implement instructional practices may not be easily apparent during casual observation, yet they are inextricably linked to positive academic and social outcomes for students, thus imperative to effective professional practice. This study investigates the ways in which professional educators make decisions about the transfer and application of professional learning centered on critical competencies (soft skills) in their daily work. Narrative Inquiry (NI) provided the methodological frame for this exploratory study that through thematic analysis surfaced five key factors influencing learning transfer: Instructor/Presenter/Facilitator; Connection to Lived Experience; Relevance to Job Assignment; Alignment with Self-Identity; and COVID–19. This dissertation is available in open access at AURA (https://aura.antioch.edu ) and OhioLINK ETD Center (https://etd.ohiolink.edu).

Committee: Jon Wergin PhD (Committee Chair); J. Beth Mabry PhD (Committee Member); Leann Kaiser PhD (Committee Member) Subjects: Continuing Education; Education; Educational Evaluation; Educational Leadership; Inservice Training; Instructional Design; Organizational Behavior; School Administration; Teacher Education; Teaching

MS, University of Cincinnati, 2021, Education, Criminal Justice, and Human Services: Information Technology

The world is still overwhelmed by the spread of the COVID-19 virus. With a total of 118,154,964 infected cases as of the ninth of March 2021 and affecting 219 countries and territories, the world is still in the pandemic. The detection of COVID-19 using the deep learning method on CT scan images can play a vital role in assisting medical professionals in current pandemic times. To control the spread of disease as well as to support the decision-making process faster by medical professionals, contribution to this area of research is crucial. The current method RT-PCR o diagnosing COVID-19 is time consuming and not available universally. The convolution neural network is widely used in the field of large-scale image recognition. This research aims to propose a deep learning-based approach that classifies COVID-19 pneumonia patients from bacterial pneumonia, viral pneumonia, and healthy (normal cases) on open-source data available. In this study, deep transfer learning is used to classify the data using Inception-ResNet-V2 neural network architecture. In this thesis multi-class open-source dataset is used to perform the binary classification and multi-class classification experiment. The dataset has (i) Collection of 8699 CT scan images including 4001 CT images of confirmed positive COVID-19 pneumonia disease and 4698 CT images with no disease (normal lungs CT) (ii) Collection of 11931 CT scan images including 1255 CT images of viral pneumonia, 4698 CT images of normal lungs with no disease, 4001 CT images of confirmed COVID-19 pneumonia disease and 1977 CT images of bacterial pneumonia. Experimental results show that the pre-trained model Inception-ResNet-V2 achieves sensitivity rate of 0.97 for binary class and 0.89 for multiclass classification.

Committee: Nelly Elsayed Ph.D. (Committee Chair); Zaghloul Elsayed Ph.D. (Committee Member); M. Murat Ozer Ph.D. (Committee Member) Subjects: Artificial Intelligence

Doctor of Philosophy, Case Western Reserve University, 2021, Biomedical Engineering

Cervical level spinal cord injuries often result in paralysis of all four limbs - a condition known as tetraplegia. Tetraplegia severely limits patient independence and quality of life. Previous studies have demonstrated that coordinated functional electrical stimulation (FES) of the neuromuscular system can restore limited motor function to people with tetraplegia. However, to fully restore upper-limb motor function, controllers for FES systems must be able to coordinate the many actuators and many mechanical degrees of freedom of the human upper extremity. Several FES controller architectures have already been identified. However, most of these architectures required patient-specific manual tuning, which may not be practical to perform for all 175,000 people in the United States living with tetraplegia due to spinal cord injuries. Here, I propose an FES controller for the human upper extremity that learns automatically via reinforcement learning without the need for patient-specific manual tuning. I demonstrate that the reinforcement learning controller can quickly learn to control a horizontal planar model of the human arm with high accuracy. In the future, I hope that reinforcement learning controllers will enable efficient and efficacious restoration of motor function to people with spinal cord injuries.

Committee: Robert Kirsch PhD (Advisor); Jonathan Miller MD (Committee Member); Dustin Tyler PhD (Committee Member); Antonie van den Bogert PhD (Committee Member) Subjects: Biomechanics; Biomedical Engineering; Engineering

PhD, University of Cincinnati, 2021, Engineering and Applied Science: Mechanical Engineering

Using Artificial Intelligence (AI) for different applications is gaining more popularity. Financial industries have long been using AI to predict customer behavior and enhance their market share. Using AI in manufacturing imposes unique challenges to data security and privacy and therefore in the early years, it had a very slow adoption rate. The advent of technologies such as cloud computing and server-less architecture with their "pay as you go"pricing strategy has significantly changed the data access and security landscape provingubiquitous data and computation access throughout the entire business. Nowadays, advanced analytic methods such as deep learning are changing the traditional machine learning-based AI frameworks because of their better performance and significant ease of use. Among them, Convolutional Neural Networks (CNN) have drastically changed the conventional image and signal processing by providing automatic feature extraction, self-learning, and better pattern recognition. New applications scenarios such as self-driving cars and autonomous manufacturing are emerging and expected to significantly change the way humans work and interact with each other. Although promising, labeled data as the key component of the AI system is not available in all applications. For manufacturing systems, this is more challenging due to the high-volume data, multi-sensor access, and data integrity across processes, machines, and systems. In addition, data is often noisy with missing values and inconsistency that further hinders developments. The performance of deep learning systems highly depends on the availability of high-quality datasets. In order to address this issue, we adopted transfer learning as a viable solution that uses labeled and unlabeled data to increase the performance of the models. Our focus is on CNN-based transfer learning for fault diagnosis in manufacturing systems; where, data shift due to maintenance, degradation, change in the operat (open full item for complete abstract)

Committee: Jay Lee Ph.D. (Committee Chair); Manish Kumar Ph.D. (Committee Member); Jing Shi Ph.D. (Committee Member); Jonathan Tan Ph.D. (Committee Member) Subjects: Engineering

Master of Sciences, Case Western Reserve University, 2021, EECS - Computer and Information Sciences

Fundus photography-based eye disease prediction attracted great attention since breakthroughs in deep convolutional neuron networks (DCNNs). However, the performance of existing studies focusing on identifying the right disease among several candidates,which is close to clinical diagnosis in practice,is at most mediocre. Moreover, obtaining large labeled dataset is difficult due to privacy concerns, resulting in the infeasibility to train huge DCNNs. Hence, we propose to utilize a lightweight deep learning architecture (MobileNetV2) and transfer learning to distinguish four eye diseases from normal controls using a small dataset. A visualization approach is also applied to highlight the loci for the predicted label, which may give some hints for further fundus image studies. Our experimental results show that our system achieves an average accuracy of 96.2%, sensitivity of 90.4%, and specificity of 97.6% via five independent runs, and outperforms two other deep learning based algorithms both in accuracy and efficiency.

Committee: Jing Li (Advisor); Guiyun Wu (Committee Member); Shuai Xu (Committee Member) Subjects: Artificial Intelligence; Bioinformatics; Computer Science

Master of Arts, The Ohio State University, 2020, Psychology

Transfer of learning refers to how learning in one context influences performance in a different context. A well-versed theory of transfer is paramount to understanding learning. Yet, a thorough understanding of transfer has been frustratingly elusive, with some researchers arguing that meaningful transfer rarely occurs or attempts to detect transfer are futile. In spite of this pessimism, we explore a model-based account of transfer. Building on the laws of practice, we develop a scalable, quantitative framework to detect transfer (or lack thereof). We perform a simulation study to explore under what conditions under which transfer can be detected and model parameters can be faithfully recovered. We then use our modeling framework to explore a large-scale game-play dataset from Lumosity. Our results suggest there are conditions in which transfer is more easily detected and there is evidence of specific game-to-game transfer in the Lumosity data.

Committee: Brandon Turner (Advisor); Patricia Van Zandt (Committee Member); Vladimir Sloutsky (Committee Member) Subjects: Psychology

PhD, University of Cincinnati, 2020, Engineering and Applied Science: Mechanical Engineering

Aiming to achieve near-zero down time for machinery by transforming the traditional preventive maintenance into predictive maintenance, the research on prognostic and health management (PHM) has attracted a great deal of attention over the past two decades. Prognostics is the process of assessing the current health condition and next predicting the future health condition, which includes incipient failure detection and remaining useful life (RUL) prediction. The outputs of the prognostics, which is the health status of the machine, play a key role in the decision-making process of the subsequent health management plan. The research that has been conducted thus far in the PHM area has mainly focused on prognostics, meaning that a large number of papers that have been published focus on the development of prognostic models in the current literature. However, the application of these developed prognostic models in industry is still limited. My theory, which will be discussed throughout this dissertation, is that this gap is caused by three main leading issues. The first issue is that when faced with complex raw data coming from complicated machines, the features used to establish the prognostic models may not accurately capture the health status of machine. The second issue is that when faced with a distribution shift caused by machine variance, the performance of the machine learning-based prognostic models would deteriorate significantly, since the assumption that the testing data and training data follow the same distribution is incorrect. The third issue is the high cost associated with the prognostic model deployment, which is a key factor preventing industry from adapting the prognostic model. In order to address these issues and promote the process of industrial application, this research proposes 5-stage framework to guide the development of robust prognostic models suitable for large-scale deployment. More specifically, within the proposed framework, a met (open full item for complete abstract)

Committee: Jay Lee Ph.D. (Committee Chair); Chao Jin Ph.D. (Committee Member); Milind Jog Ph.D. (Committee Member); Jay Kim Ph.D. (Committee Member) Subjects: Engineering

Master of Sciences (Engineering), Case Western Reserve University, 2020, EECS - Computer and Information Sciences

Complex sclerosing lesions and radial scars are rare benign architectural distortions (AD), which present a difficult diagnostic problem due to the visual similarity they share with malignant AD; practice standards state that patients found to have AD should be biopsied. 3D digital breast tomosynthesis (DBT) has shown improved cancer detection and lower recall rate compared to traditional 2D digital mammography, as well as greater sensitivity to the presence of AD. Computer-aided diagnosis systems (CAD) can alleviate unnecessary biopsies through Deep Learning (DL) approaches, involving the training of a convolutional neural network to recognize disease patterns. DBTs of AD are commonly acquired multiple times, from more than one angle. Effectively combining the acquisitions, such as methods combining the predictions from the different angles, can boost diagnostic performance. We explore baseline methods for combining the acquisitions and propose a novel sequence-based CAD that may provide means to differentiating malignant and benign AD.

Committee: Anant Madabhushi (Committee Chair); Michael Lewicki (Committee Member); Jing Li (Committee Member) Subjects: Artificial Intelligence; Comparative; Computer Science; Medical Imaging

Doctor of Philosophy, The Ohio State University, 2020, Biomedical Sciences

There are trillions of cells, which make up hundreds of different cell types, found in the human body. These cells make up not only tissues but dictate the functions of those tissues. In diseased tissues, cell types can have a profound impact on the outcome of a patient. For these reasons, having a comprehensive understanding of cell types is important. In the past 10 years, single cell RNA sequencing has profoundly impacted our understanding of known and previously unknown cell types. Along with the numerous single cell datasets, a multitude of bulk expression datasets, multi-omic datasets, and curated information also exist. All of these data sources must be leveraged together to most improve our understanding of human tissues and diseases at the single cell level. We developed methodologies, frameworks, and algorithms that leverage multiple diverse datasets simultaneously to better understand single cell RNA sequencing data and as a result tissue heterogeneity as a whole.

Committee: Yan Zhang (Advisor); Kun Huang (Advisor); Jeffrey Parvin (Committee Member); Christopher Bartlett (Committee Member) Subjects: Bioinformatics; Biomedical Research