Master of Science (MS), Wright State University, 2024, Computer Science

In today's technological landscape, hardware devices are integral to critical applications such as industrial automation, autonomous vehicles, and medical equipment, relying on advanced platforms like FPGAs for core functionalities. However, the multi-stage manufacturing process, often distributed across various foundries, introduces substantial security risks, notably the potential for hardware Trojan insertion. These malicious modifications compromise the reliability and safety of hardware systems. This research addresses the detection of hardware Trojans through side-channel analysis, utilizing power and electromagnetic signal data, combined with meta-learning techniques, specifically model stacking. By employing diverse base models and a meta-model to consolidate predictions, this non-invasive approach effectively identifies Trojans without requiring direct access to internal circuitry. The methodology demonstrates robust classification capabilities, achieving an accuracy of 88.0\%, precision of 81.0\%, and recall of 95.0\%, even on previously unseen data. The results highlight the superior performance of meta-learning over traditional detection methods, offering an efficient and reliable solution to enhance hardware security.

Committee: Fathi Amsaad Ph.D. (Advisor); Junjie Zhang Ph.D. (Committee Member); Huaining Cheng Ph.D. (Committee Member); Nitin Pundir Ph.D. (Committee Member); Thomas Wischgoll Ph.D. (Other); Subhashini Ganapathy Ph.D. (Other) Subjects: Computer Engineering; Computer Science; Electrical Engineering

Master of Science (MS), Wright State University, 2024, Computer Science

Semiconductor microelectronics integrated circuits (ICs) are increasingly integrated into modern life-critical applications, from intelligent infrastructure and consumer electronics to the Internet of Things (IoT) and advanced military and medical systems. Unfortunately, these applications are vulnerable to new hardware security attacks, including microelectronics counterfeits and hardware modification attacks. Physical Unclonable Functions (PUFs) are state-of-the-art hardware security solutions that utilize process variations of integrated circuits for device authentication, secret key generation, and microelectronics counterfeit detection. The negative impact of aging on Static Random \linebreak Access Memory Physical Unclonable Functions (SRAM PUFs) has significant consequences for microelectronics authentication, security, and reliability. This research thoroughly \linebreak examines the effect of aging on the reliability of SRAM PUFs used for secure and trusted microelectronics integrated circuit applications. It initially provides an overview of SRAM PUFs, highlighting their significance and essential features while addressing encountered challenges. The study then covers mitigation techniques, including multi-modal PUFs, that already exist to boost the resilience of SRAM PUFs against aging impacts, highlighting their advantages and the gap in the research addressed in this research. This work proposes a novel AI-enabled security for reliable SRAM PUFs. The proposed approach aims to study and countermeasure the impact of aging on SRAM PUF by analyzing data samples, including Bias Temperature Instability (BTI), Bit Flips, Accelerated aging, and Hot Carrier Injection (HCI) and to study their effects on SRAM PUF cell properties and output. Accelerated aging is a direct result of a change in the environmental temperature and voltage for a few hours. We aim to mitigate the impact of accelerated aging on the reliability authentication and encryption keys of (open full item for complete abstract)

Committee: Fathi Amsaad Ph.D. (Advisor); Wen Zhang Ph.D. (Committee Member); John Emmert Ph.D. (Committee Member) Subjects: Computer Engineering; Computer Science; Electrical Engineering; Information Technology; Technology

Doctor of Philosophy, Miami University, 2024, Cell, Molecular and Structural Biology (CMSB)

Long non-coding RNAs (lncRNAs) play critical roles in diverse biological processes. The extensive availability of public RNA-Seq data offers valuable resources for identifying novel lncRNAs. Here, we introduce LncDC (Long non-coding RNA detection), a machine learning-based tool designed to detect lncRNAs from RNA-Seq data. LncDC utilizes an XGBoost model incorporating features derived from primary sequences, secondary structures, and translated proteins to differentiate between lncRNAs and mRNAs. Notably, sequence and secondary structure k-mer score features, along with various open reading frame-related features, contribute to the classification of lncRNAs and mRNAs. Benchmarking experiments have shown that LncDC surpasses six state-of-the-art tools in several performance metrics. Applying LncDC to 180 RNA-Seq datasets from osteosarcoma patients led to the discovery of 97 novel osteosarcoma-specific lncRNAs. Additionally, the role of lncRNA in Oryza sativa RNase P protein 30 (OsRpp30)-mediated disease resistance in rice remains largely unexplored. OsRpp30 is known as a positive regulator of rice immunity against various pathogens. To further understand this mechanism, we conducted RNA-Seq and small RNA-Seq profiling of lncRNAs, miRNAs, and mRNAs in wild type, OsRpp30 overexpression, and OsRpp30 knockout rice plants. Our comprehensive transcriptome analysis identified 91 differentially expressed lncRNAs, 1671 differentially expressed mRNAs, and 41 differentially expressed miRNAs across these rice lines. We also explored interactions between differentially expressed lncRNAs and mRNAs, uncovering 10 trans- and 27 cis-targeting pairs specific to the OsRpp30 overexpression and knockout conditions. Furthermore, we constructed a competing endogenous RNA network comprising differentially expressed lncRNAs, miRNAs, and mRNAs to elucidate their interactions in rice immunity. Our findings reveal that lncRNAs participate in OsRpp30-mediated disease resistance in rice by regula (open full item for complete abstract)

Committee: Chun Liang (Advisor); Haifei Shi (Committee Chair); Philippe Giabbanelli (Committee Member); Richard Moore (Committee Member); Tereza Jezkova (Committee Member) Subjects: Bioinformatics; Biology

Master of Computing and Information Systems, Youngstown State University, 2023, Department of Computer Science and Information Systems

Additive Manufacturing (AM), particularly LASER Powder Bed Fusion (LPBF), has gained prominence for its flexibility and precision in handling complex metal structures. However, optimizing L-PBF for intricate designs involves analyzing over 130 process parameters, leading to prolonged duration and increased costs. This thesis proposes a novel approach by harnessing statistical and machine learning algorithms to predict manufacturability issues before the printing process. By performing a comparative analysis of the intended design with the machine produced result, the study introduces two machine learning and one artificial neural network (ANN) algorithm to forecast the printability of new designs accurately. This innovative method aims to reduce or eliminate the need for iterative printing, reducing productivity costs and optimizing the LPBF additive manufacturing process.

Committee: Alina Lazar PhD (Advisor); John R. Sullins PhD (Committee Member); Hunter Taylor PhD (Committee Member) Subjects: Computer Engineering; Computer Science; Engineering; Information Science; Information Systems; Information Technology; Materials Science; Mechanical Engineering

Master of Science (MS), Wright State University, 2023, Computer Science

The Industrial Internet of Things (IIoT) refers to a set of smart devices, i.e., actuators, detectors, smart sensors, and autonomous systems connected throughout the Internet to help achieve the purpose of various industrial applications. Unfortunately, IIoT applications are increasingly integrated into insecure physical environments leading to greater exposure to new cyber and physical system attacks. In the current IIoT security realm, effective anomaly detection is crucial for ensuring the integrity and reliability of critical infrastructure. Traditional security solutions may not apply to IIoT due to new dimensions, including extreme energy constraints in IIoT devices. Deep learning (DL) techniques like Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) have been the focus of recent research to increase the precision and effectiveness of anomaly identification. This Thesis initially investigates a unique hybrid DL-enabled approach that provide the needed security analysis before successful attacks are launched against IIoT infrastructure. For that, different hybrid models are developed, trained, tested, and validated using Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), Short-Term Memory (LSTM), Autoencoder, and XGBoost algorithms. Experimental results show that the proposed XGBoost ML model exhibits the highest performance, as compared to other models, and excels across multiple metrics, including recall, precision, F1-score, and false alarm rate (FAR). The results also show that hybrid CNN+GRU model is closely behind, which exhibited strong performance in accurately detecting anomalies in encrypted IoT traffic. Specifically, Our experimental results show that the developed hybrid CNN+GRU model outperforms the others, achieving an accuracy of 94.94%, a recall of 92.29%, a precision of 98.49%, an F1 score of 95.24%, and a low false alarm rate of 0.001. However, it is (open full item for complete abstract)

Committee: Fathi Amsaad Ph.D. (Advisor); Lingwei Chen Ph.D. (Committee Member); Michael L. Raymer Ph.D. (Committee Member); Anton Netchaev Ph.D. (Committee Member) Subjects: Computer Science

Master of Science, The Ohio State University, 2023, Mechanical Engineering

This thesis presents and evaluates two Machine Learning models to model the degradation of an EV's battery capacity: Extreme Gradient Boosting (XGB) and Neural Basis Expansion (N-BEATS). XGB was used to estimate incremental losses in battery capacity as a function of temperature, state of charge, amp-hour throughput, and time. As a performance comparison, a semi-empirical model was developed and calibrated using Particle Swarm Optimization. The XGB model test results produced a mean average percentage error 61\% lower than the semi-empirical model, with significantly lower implementation costs. N-BEATS was used to predict long-term capacity losses. Using a 13-week lookback window, along with temporal and thermal covariate data, the model was able to predict 10 year capacity loss with 2\% error. This thesis aims to demonstrate the potential for Machine Learning in battery capacity modeling. It closes with a discussion of model limitations, suggested improvements, and possible applications.

Committee: Matilde D'Arpino (Committee Member); Giorgio Rizzoni (Advisor) Subjects: Computer Science; Mechanical Engineering

Master of Arts, Miami University, 2022, Geography

The accurate attribution of fire detections to a specific fuel source type is of critical importance for the assessment of emissions, tracking energy and industry expansion, and informing wildfire management. The Arctic presents a particular set of challenges to accurate source attribution owing primarily to its remoteness, diversity, and the difficulty inherent in imaging at extreme latitudes. This study describes the development of a novel machine learning application for the identification of persistent sources of fire in high northern latitudes, especially those persistent sources which represent non-biomass burns. Emphasis is placed on the integration of existing data alongside measures of a fire detections' persistence in space and time, as well as other fire characteristics. The results provide important contextual information about the probability of a given fire detection being associated with a persistent non-biomass burning source. Assessment of the model at various persistence probability cutoffs is given to provide guidance for its application. Future improvements to the model are discussed as well as the limitations of the current iteration which estimates all persistent pan-Arctic fires for 2018.

Committee: Jessica McCarty (Advisor); Mary Henry (Committee Member); John Maingi (Committee Member) Subjects: Earth; Environmental Science; Forestry; Geographic Information Science; Geography

Master of Science, The Ohio State University, 2021, Mechanical Engineering

Poor outcomes for patients with trauma result from many non-linear dependent risk factors, including patient demographics, injury characteristics, medical care provided, and characteristics of medical facilities; yet traditional approaches attempted to capture these relationships using rigid regression models. We hypothesized that neural networks and gradient-boosted trees could deeply understand a trauma patient's condition and accurately identify individuals at high risk for mortality or admission to an intensive care unit without relying on restrictive regression model criteria. Deidentified patient visit data were obtained from the years 2007-2014 of the National Trauma Data Bank and 2007-2015 of the National Hospital Ambulatory Medical Care Survey. All patient visits occurred in U.S. hospitals, with only a small minority of the more than 2 million encounters resulting in mortality or admission to an intensive care unit. We designed, trained, and evaluated different models' performance on patients with complete data and those with missing features independently. The models were evaluated on their sensitivity, specificity, positive and negative predictive value, and Matthews Correlation Coefficient. When working with complete data only, gradient-boosted tree methods and neural networks perform similarly and models developed in recent papers, with a slightly more favorable Matthews Correlation Coefficient with the neural net. However, the tree-based methods had much stronger performance characteristics once incomplete data was introduced, as they did not require any form of imputation in order to process the data. While testing for confounding factors, we discovered that excluding fall-related injuries boosted performance for adult trauma patients; however, it reduced performance for children. The models described here demonstrate similar performance to contemporary machine intelligence models without requiring restrictive regression model criteria or extensive (open full item for complete abstract)

Committee: Satyanarayana Seetharaman (Committee Member); Herman Shen (Advisor) Subjects: Mechanical Engineering

Master of Science, University of Toledo, 2020, Engineering (Computer Science)

In this study, an Intrusion Detection system (IDS) is designed based on Machine Learning classifiers and its performance is evaluated for the set of attacks entailed in the UNSW-NB15 dataset. This dataset is comprised of 2,540,226 realistic network data instances as well as 49 features. Most studies reported in the literature employ a representative subset of this dataset with predefined training and testing subsets, and containing a total of 257,673 records which this study also used. In light of relatively lower than expected performance of Machine Learning or Statistical classification algorithms tested on this dataset and as reported by others in the literature, this dataset was subjected to visual data analysis to explore potential reasons or issues which likely challenge Machine Learning classifiers. The consequent observations demonstrated the presence of class representation imbalance with respect to pattern counts and class overlap in feature space, which makes preprocessing strategies indispensable before this dataset can be meaningfully employed for data-driven model development for intrusion detection. For preprocessing, we implemented min-max scaling in the normalization phase followed by the application of Elastic Net and Sequential Feature Selection (SFS) algorithms. We employed ensemble methods using three base classifiers, namely Balanced Bagging, XGBoost, and RF-HDDT, augmented to address the imbalance issue. Parameters of Balanced Bagging and XGBoost are tuned for the imbalanced data, and Random Forest is supplemented by the Hellinger distance metric to address the limitations of default distance metric. Two new algorithms are proposed to address the class overlap issue in the dataset and applied during training. These two algorithms are leveraged to help improve the performance on the testing dataset by affecting the final classification decision made by three base classifiers as part of the ensemble classifier which employsa majority vote combi (open full item for complete abstract)

Committee: Gursel Serpen (Committee Chair); Ahmad Y. Javaid (Committee Member); Mohammed Niamat (Committee Member); Richard G. Molyet (Committee Member) Subjects: Computer Engineering; Computer Science; Engineering; Mathematics; Statistics