Digital Commons Network^™

Deficient Executive Control In Transformer Attention, Suketu Chandrakant Patel, Hongbin Wang, Jin Fan

Publications and Research

Although transformers in large language models (LLMs) effectively implement a self-attention mechanism that has revolutionized natural language processing, they lack an explicit architecture for the executive control of attention found in humans, which is essential for resolving conflicts and selecting relevant information in the presence of competing computations and is critical for adaptive behavior. To investigate the impact of this limitation in LLMs, we employed the classic color Stroop task, widely regarded as the gold standard, to test the executive control of attention in these models. Our results revealed a typical conflict effect of underperformance in terms of accuracy in …

Deep Learning Approaches For Ctenophore Identification And Tracking, Anagha Bharadwaj

Theses

Ctenophores are translucent marine organisms with nearly invisible tentacles and pose significant challenges due to their transparent morphology and ambiguous structural features. This research addresses the classification and tracking of these organisms and evaluates the performance of current computer vision models under sparse-data environments.

A dataset from the NJIT Life History Lab consisting of microscopic laboratory videos and photographs of different growth stages is used to train and assess a number of convolutional neural network designs, including VGG16, ResNet, BioCLIP2, YOLO, and DeepLabCut. Additionally, a web-based interface is developed to evaluate expert-labeled ground truth with the model's performance.

The findings …

Adaptive Multimodal Smart Home Control On A Raspberry Pi 5 Using Hand Gestures, Voice Cues, And User Feedback, Vaibhav Bora

Theses

A real time multimodal smart home control system deployed on a Raspberry Pi 5 is presented. The system combines hand gestures, short voice cues, and proximity aware interaction to execute household commands such as light brightness control, fan speed adjustment, and stop or kill switch actions. Lightweight gesture and keyword spotting voice classifiers were trained offline and exported to TensorFlow Lite for efficient on device inference. For more natural spoken phrases, the system additionally integrates a locally deployed pretrained offline ASR component rather than a speech recognizer trained from scratch. Using a USB camera and microphone, the system operates fully …

Retinomorphic Mid-Wave Infrared In-Sensor Processing Engine: Device-To-Architecture Co-Design, Hemalatha Nagaraju

Theses

Conventional frame-based CMOS image sensors acquire full-frame pixel data at discrete time intervals, resulting in substantial spatial redundancy and loss of temporal information between frames. The repeated conversion and transfer of redundant pixel data increases bandwidth and power consumption in machine vision systems. Retinomorphic sensing architectures address these limitations by enabling programmable, analog-domain processing directly at the sensor interface. A compact behavioral model of the PbSe device is developed in HSPICE based on calibrated TCAD simulation data to capture gate-controlled photocurrent modulation under varying illumination and gate bias conditions. Error analysis is performed to quantify the deviation between TCAD-generated photocurrent …

Omama-Db: The Oregon-Massachusetts Mammography Database, Avanih Kanamarlapudi

Graduate Masters Theses

Public datasets for training AI models in breast cancer screening are limited in size and quality, making it difficult to develop reliable systems. We introduce OMAMA-DB, an extensive publicly available collection of 2D mammograms and 3D tomosynthesis volumes. Starting from 967,991 images, we created a curated set of 231,080 images us ing a multi-stage filtering process that removes missing labels, uncommon dimensions, rare scanner types, duplicate studies, and invalid DICOM files. All 2D images then undergo additional outlier detection using histogram filtering and a variational autoen coder to remove low-quality outliers. OMAMA-DB includes pathology-based cancer labels and automated lesion annotations …

Reliable Offline Evaluation Under Exposure Bias: A Cross-Scale Study Of Counterfactual Estimators In Recommender Systems, Lakshmi Pranathi Vutla

Graduate Masters Theses

Offline evaluation underpins model selection in recommender systems, yet historical interaction logs are shaped by prior recommendation policies. Because users only provide feedback on exposed items, logged data entangles user preferences with exposure mechanisms, leading to exposure bias and potentially misleading model comparisons. Counterfactual estimators such as IPS, SNIPS, CRM, and DR offer principled corrections, but their empirical reliability across datasets and exposure regimes remains insufficiently under- stood. We present a systematic, cross-scale study of counterfactual evaluation in recommender systems. Comparing IPS, SNIPS, CRM, and DR on datasets with randomized exposure (Yahoo! R3, Coat, and KuaiRec), we analyze estimator behavior …

Enabling Ml/Ai In 6g And Future Wireless Communication With Privacy Preservation, Mec Offloading And Quantum Computing, Changshi Zhou

Dissertations

The forthcoming sixth-generation (6G) and future wireless networks are envisioned to support an unprecedented range of services, delivering ultra-low latency, massive connectivity, and intelligent real-time responsiveness. These capabilities will enable emerging applications such as extended reality (XR), autonomous vehicles (AVs), industrial robotics, and the Internet of Things (IoT) to reach their full potential. Achieving this vision requires the integration of enabling technologies such as artificial intelligence and machine learning (AI/ML) and quantum computing, which are poised to play central roles in shaping the landscape of wireless communication systems.

In AI-native, data-driven, and computing-centric 6G networks, ML models will be deeply …

Holistic Dram Enhancements: From Intrinsic In-Memory Operations To Robust Security Mechanisms, Ranyang Zhou

Dissertations

Dynamic Random-Access Memory (DRAM) is both the performance bottleneck and a critical security boundary of modern computing systems. Its physical properties make it an attractive substrate for near-data computation—yet those same properties expose it to disturbance-based hardware attacks. This dissertation argues that these two dimensions are not independent: the architectural choices that make DRAM efficient also reshape its threat landscape. Addressing both requires a unified approach to memory architecture and security co-design.

The first part of this dissertation attacks the memory wall through four processing-in-DRAM (PIM) frameworks. ReD-LUT and LT-PIM unify lookup-table arithmetic with charge-sharing logic, achieving up to 37.8x …

Towards Interpretable Transfer Learning With Limited Data, Youxiang Zhu

Graduate Doctoral Dissertations

Modern foundation models are typically trained with large-scale data to ensure good performance. However, certain tasks cannot scale with large amounts of data due to cost and practical constraints, resulting in limited performance. To address this, in this dissertation, I introduce model-task alignment, a general methodology for making a foundation model work well with tasks with limited data. The methodology comprises two parts: aligning downstream tasks with foundation models and aligning foundation models with downstream tasks. To study and validate this methodology, I first focus on speech-based dementia detection, a representative task with limited data, and then extend to general …

Ai-Augmented Financial Advisors: Comparing Ai And Human Analyst Investment Recommendations In Agreement, Performance, And Firm Size Effects, Crystal Chen

Honors College Theses

This study examines the level of agreement and performance between artificial intelligence (AI) generated investment recommendations and human analyst recommendations across U.S. publicly traded firms. Using a sample of twelve companies categorized by firm size (large, mid, and small), the study collects buy, hold, or sell recommendations from generative AI systems and human analysts. Agreement between AI-to-AI and AI-to-human recommendations is measured using Cohen’s Kappa agreement. Portfolio performance is evaluated by constructing equal-weighted portfolios for each recommendation source and size category. Risk-adjusted returns are measured using the Sharpe ratio over 1-, 2-, and 3-month periods. Furthermore, the study tests whether …

Debatrix, Vivienne Lu, Huy Ngo, Luke Ponssen, Jonathan Preiss, Ryan Rani

Computer Science and Engineering Senior Theses

Developing public-speaking skills remains a persistent challenge in formal education, constrained by limited instructional time and the lack of scalable, individualized feedback. Existing automated tools address only narrow aspects of this problem, offering text-based coaching against rigid rubrics that fail to capture argument quality, evidence use, or real-time rebuttal skill. This thesis presents Debatrix, an AI-powered platform that enables K-12 students, university learners, and independent self-studiers to debate an intelligent opponent. The system combines automatic speech recognition, large language model-driven rebuttal generation, and a multi-dimensional rhetorical analysis engine that evaluates argument structure, evidence integration, and persuasive technique. Users receive explainable …

Evaluating Soil Health And Crop Yield In Louisiana Agricultural Systems: Impacts Of Best Management Practices And Prediction Models, Hector J. Mendoza Lagos

LSU Doctoral Dissertations

The adoption of conservation management practices is critical for improving soil health, enhancing nutrient use efficiency, and sustaining crop productivity in row crop systems in Louisiana. This study evaluated the role of conservation agronomic practices, soil biochemical indicators, and machine learning predictive models to improve soil nutrient dynamics, soil health indicators, microbial communities (MC), and crop productivity on a corn (Zea mays L.) research plot scale and in a commercial forty-hectare cotton (Gassypium hirsutum L.)-corn-soybean (Glycine max L.) rotation system in northeast Louisiana. The objectives of the study were to evaluate soil nutrient dynamics and MCs under …

An Ai-Integrated Methodology For Secure Software And System Development, Ian Matthew Campbell Coston

Electronic Theses and Dissertations 2020 - Present

Securing interconnected software systems requires more than layering existing frameworks on top of each other. Most current Secure Software and System Development Lifecycle (S-SDLC) models treat security as a phase rather than a design condition, leaving real gaps in governance, access control, and automated enforcement that become critical failure points in Internet of Things (IoT) environments where devices are resource-constrained, long-lived, and frequently insecure by default.

This dissertation introduces the Automated Zero Trust Risk Management with DevSecOps Integration (AZTRM-D) methodology, a novel approach that unifies DevSecOps automation, the National Institute of Standards and Technology (NIST) Risk Management Framework (RMF), and …

Wake-On-Anomaly Federated Architecture For Privacy-Preserving Poultry Health Early Warning, Mahmoud Aziz Louati

Masters Theses

Highly pathogenic avian influenza outbreaks, respiratory disease, heat stress, and silent equipment failures share one operational reality: they are detected too late because today’s poultry-health workflow is reactive, manual, and dependent on producers volunteering commercially sensitive data. This thesis presents a wake-on-anomaly federated architecture that addresses both the detection-latency problem and the privacy–adoption deadlock that has so far prevented cross-farm collaboration. The architecture is organized in two tiers. Tier 1 is a lightweight LSTM autoencoder that continuously screens four routine telemetry channels (water, feed, house temperature, activity proxy) and emits a per-window reconstruction-error score. A debounced k-of-m trigger with cooldown …

Decision Making At Triage Classification Using Svm With Smote Technique, Mehanas Shahul, Pushpalatha Kp

Northeast Journal of Complex Systems (NEJCS)

The efficient functioning of triage gates in overcrowded emergency departments (EDs) occurs in the context of the complex adaptive system (CAS) framework, where diverse system elements – patients, medical personnel, resources, patients’ inflow patterns, and patients themselves – simultaneously and dynamically influence the decision process. This study addresses the automated incorporation of machine learning triage algorithms as part of the system triage process to support automated classified risk-level recognition based on a limited set of vital signs. Patients are dynamically subsumed under high and low-risk categories enhanced by sensitivity, which enables optimal diagnosis and triage response to the critical clinician …

Power Consumption Prediction And Energy Forecasting Using Machine Learning Models, Sheik Mohideen Shah S Mr

Theses and Dissertations

Power consumption trends are essential to be identified in the energy grid areas to analyze the utilization, deficiency, and the measures to be taken for an effective and comfortable usage of energy. There are two scenarios in which the power consumption can be analyzed, namely identification and prediction. Identification deals with the post-utilization analysis of energy trends, whereas prediction deals with prior analysis of various factors of energy utilization, including the cost, supply details, shortages, and the need for new energy resources. In the existing models, the power consumption-related data are collected through smart meters, and the energy forecasting methods …

Real-Time Fraud Detection, Haidi Aly Fahmy, Abdelhadi Nait-Zerrad, Shiva Krishana Reddy Ravuula, Sangwhan Cha

Harrisburg University Other Works

Financial fraud detection is a high-volume, high-velocity analytics problem. Traditional rule-based systems are often easy to deploy, but they are limited by static thresholds, delayed response, high false-positive rates, and weak explainability. This report presents a formalized end-to-end Big Data architecture for real-time fraud and anomaly detection in financial transaction streams.

The proposed architecture ingests transaction events through AWS Kinesis, enriches them through an Apache Flink stream-processing layer, scores them with an XGBoost classifier, explains model outputs using SHAP, and converts structured evidence into human-readable summaries through a controlled GPT explanation layer. Results are persisted through a hybrid storage strategy …

Automatic Speed Guidance Method And Simulation Evaluation For Trams At Intersections, Jing Teng, Wencong Tong, Zhongjie Zhang, Xing Yao, Junxian Li

Journal of System Simulation

To address the lack of speed regulation mechanism and the high dispersion of operational time in the current manual driving mode, this paper proposes an automatic speed guidance method for trams at intersections. Considering the speed disturbances caused by potential traffic conflicts at intersections, an initial decision point for safe passage speed at an intersection is established, dividing the operational curve into deterministic segments and disturbance-response segments.To verify the effectiveness of the method, a case study of Songjiang Tram Line 1 is conducted.Driving simulation experiments are performed to obtain manual driving trajectories, and a dynamic trajectory simulation model …

Capacity Market Trading Strategies Of Generators Based On Per-Maddpg Algorithm, Yanbin Li, Zhaolun Pan, Xinyue Ma, Minghao Song, Yujie Hu, Xiaoda Xue

Journal of System Simulation

Considering the issue of how power generators trade off their quantity and price bidding strategies to maximize profits in different capacity market environments, a capacity market bidding equilibrium model is constructed. Recognizing the limitations of traditional solution methods, which rely on the assumption of complete information and have low utilization of historical trading strategy information, a capacity market trading simulation method based on prioritized experience replay multi- agent deep deterministic policy gradient (PER-MADDPG) is proposed. The action space is constructed using quantity bidding strategy and price bidding strategy, and the state space is constructed using historical transaction strategies and winning …

An Automated Generation Method For Combat Simulation Scenarios Based On Large Language Models, Zhiming Dong, Zhongqi Hu, Haoran Dai, Jiancheng Gao

Journal of System Simulation

To address the issue of low efficiency in generating traditional army tactical combat simulation scenarios, an automated generation method based on large language models is proposed. The large language model invokes a semantic segmentation algorithm to parse and restructure the combat scenario, forming semantic modules. Utilizing a multi-agent collaborative framework based on the model contextual protocol, the large language model drives each agent to extract simulation elements from the corresponding semantic modules, constructing a knowledge graph of scenario elements. Using this knowledge graph as a retrieval medium, the method employs a dense retrieval algorithm to achieve precise matching between simulation …

Robust Identification Of Black-Box Nonlinear Ssm Using Expectation-Maximization, Xiaonan Li, Tao Chao, Ping Ma, Ming Yang, Yuxuan Wang

Journal of System Simulation

To address the robust identification problem of nonlinear state space models (SSM) with outliers, missing observations, and unknown state equations, this paper proposes a modeling method based on eigenfunction expansion, Gaussian-process state-space models (GP-SSM), and Student-t distribution. The proposed approach consists of: modeling the state transition function using eigenfunctions and pre-encoding the priors of basis function coefficients via GP-SSM to enhance flexibility; modeling observations as a Student-t distribution with unknown parameters to enhance robustness against outliers; proposing the enhanced particle Gibbs with ancestor sampling (EPGAS) algorithm to adapt to state estimation in scenarios with missing observations; and deriving unknown model …

Multi-Agent Cbs Path Planning Algorithm Based On Minimum Planning Margin First, Longxiao Liang, Jianlin Mao, Niya Wang, Chengyuan Fang, Wenna Zhou

Journal of System Simulation

To address the problems of chain effect and insufficient solving efficiency in the conflict tree (CT) expansion of the traditional conflict-based search (CBS) framework, a minimum-margin-first CBS algorithm based on planning margin was proposed. The calculation of planning margin was introduced into the underlying A* search, and the robots with the minimum margin were prioritized in the high-level conflict resolution, to suppress the chain expansion of the CT while ensuring path optimality.Simulation experiments show that the proposed algorithm significantly reduces the amount of CT node expansion and the number of root node conflicts and effectively improves the solving efficiency, …

Topology Identification Of Complex Dynamical Networks Under Dynamical Saturation Inputs, Haoyu Wang, Yayong Wu, Guoping Jiang, Ying Zheng, Xuanxin Zhou

Journal of System Simulation

In view of the problem that the controller inputs in actual engineering systems are vulnerable to the constraints of dynamical saturation and delayed dynamical saturation, which makes it difficult for the topology identification of complex dynamical networks to adapt to real physical scenarios, a topology identification method based on the drive-response mechanism was proposed. A response network with the same dynamical characteristics and node scale as the original network was constructed, and the dynamical equation of synchronization error between the drive-response networks was established. A controller with dynamical saturation and delayed dynamical saturation and a topology identifier were designed, and …

Robust Two-Stage Mimo-Ofdm Channel Estimation Method Against Sensing Errors, Yi Peng, Jun Wang, Qingqing Yang, Jianming Wang, Hui Li

Journal of System Simulation

To address the challenges of performance degradation, high pilot overhead, and high computational complexity in traditi onal channel estimation methods for integrated sensing and communication (ISAC) assisted MIMO-OFDM systems when radar sensing information contains errors, this paper proposes a robust two-stage sparse channel estimation framework designed to be tolerant of sensing errors. In the first stage, a residual energy weighted simultaneous orthogonal matching pursuit (REW-SOMP) algorithm is designed. Leveraging locally adaptive dictionary expansion and a residual- weighted path selection mechanism, it accurately captures communication-associated paths even under sensing errors. The second stage introduces an adaptive penalty factor alternating direction method …

Dynamic Task Planning For Wargaming Based On Large Language Models, Yingang Liu, Ming Ma, Ronghua Zhang

Journal of System Simulation

To address the problems of great difficulty in intelligent decision-making and insufficient dynamism in task planning caused by the complex adversarial environment and strong uncertainty in wargaming tasks, this paper proposed a hierarchical Agent collaborative decision-making framework based on large and small model synergy.Through a multi-level structure, the hierarchical decoupling and dynamic coordination of battlefield tasks were achieved. A memory management module was constructed, and a query optimization mechanism driven by large language models was introduced to dynamically perceive the decision-making process and query intent, completing the semantic reconstruction and context completion of raw queries. A time-driven two-stage task …

Ultra-Short-Term Photovoltaic Power Prediction Based On Improved Patchtst Considering Data Drift, Huawei Mei, Penghui Yang, Yang Yu

Journal of System Simulation

Existing PV power prediction methods often suffer from limited accuracy and robustness due to three key shortcomings: relying on single-point mapping that cannot fully extract local temporal patterns; inadequate exploration of the global temporal dependencies in PV output, and failure to account for prevalent data drift phenomena. To overcome these limitations,an improved patch time series transformer (PatchTST) based approach is proposed for ultra-short-term PV power prediction. The methodology applies rough set theory for feature dimensionality reduction, effectively preserving critical decision information by analyzing both feature-label relationships and inter-feature correlations. An enhanced PatchTST model with a modified channel-independent mechanism extracts …

Sos Effectiveness Evaluation Method Based On Fuzzy Functional Dependency Network Analysis, Hongjia Su, Cheng Zhang, Fei Liu

Journal of System Simulation

Functional dependency network analysis (FDNA) enables modeling functional dependencies among equipment in a system of systems (SoS) and then computing the whole SoS effectiveness based on the effectiveness of each equipment, thus overcoming the deficiency of traditional SoS effectiveness evaluation based on tree-like index systems. However, critical parameters such as strength/criticality of dependency in this methodology currently rely on subjective empirical assignments, where the deviations resulting from subjectivity may compromise the accuracy of effectiveness evaluation. To address this limitation, this paper proposes a fuzzy FDNA (FFDNA)-based SoS effectiveness evaluation method. This method constructs a functional dependency network (FDN) model …

Multi-Agent Reinforcement Learning Method For Wargame Simulation Based On Suboptimal Demonstration Guidance, Zicong Zhou, Junjie Zeng, Yue Hu, Zhengqiu Zhu, Quanjun Yin

Journal of System Simulation

To address issues such as fixed behavior patterns and insufficient adaptability in complex adversarial environments exhibited by traditional wargame agent decision-making models, this paper proposes a multi-agent reinforcement learning method based on suboptimal demonstrations (MARLSD). The proposed method integrates reward relabeling with a self-imitation learning mechanism, effectively improving the training efficiency of multi-agent reinforcement learning algorithms in environments with large state-action spaces and sparse rewards, even when only a small number of suboptimal demonstrations are available, while encouraging agents to explore better strategies. Experimental results show that, compared with baselines such as QMIX and MAGAIL, MARLSD significantly improves performance and …

Research On Calculation Model Of Excavation Resistance Under Heterogeneous Soil Conditions, Xin Zhang, Ping Zhang, Chen Zhang, Wei Liu, Boyang Han

Journal of System Simulation

To address the problem of insufficient prediction accuracy of excavation resistance in heterogeneous cohesive soil, a spatial calculation model of excavation resistance at each excavation stage under heterogeneous soil conditions was proposedbased on the cutting wedge model, comprehensively considering multi-dimensional factors such as bucket geometry, side plate effect, lateral force, and inertia. By taking a small crawler hydraulic excavator as the research object, a coupled simulation model of boom multi-body dynamics and soil-rock particle discrete element was established, and the theoretical model was validated through the co-simulation of excavation operations. The simulation results indicate that under the working …

Research On Completion Method For Trajectory Based On Image Representation And Collaborative Feature Perception, Ye Tao, Jinhui Tang, Chen Zhou, Chong Wang

Journal of System Simulation

To address the constraints imposed by missing trajectory data in surveillance systems on the efficacy of civil aviation safety monitoring, as well as the limitations on the development and application of advanced technologies within trajectory-based operational frameworks, a completion method for trajectory based on image representation and collaborative feature perception was proposed. A conversion strategy for trajectory image representation was designed to reformulate the trajectory completion task as a deterministic image completion problem, effectively circumventing the cumulative error problem of traditional time-series data caused by the limitation of recurrent neural network inference mechanisms.A regression model fusing a multi-kernel hybrid …