Digital Commons Network^™

A New Model For Educational Program Assessments Using Automated Collective Concept Maps, Andrew R. Paullin

Master's Theses (2009 -)

This paper presents: Epistemological principles; the origins and theoretical foundations of concept maps; the current state of related works; the contributions, methods, results, and future road map of this research; and A New Model for Educational Program Assessments. An understanding that knowledge results from an evolution of cognitive structures enables a scientific approach to enhance the efficiency of learning. Building on prior works, this research provides a novel algorithm and tool to automatically create Collective Concept Maps from Individual Concept Maps by utilizing a dictionary, thesaurus, Natural Language Processing, Machine Learning / Artificial Intelligence, and domain expertise. Abstraction Filters enable …

Automated Grain-Matrix Segmentation In Photomicrographs Of Clastic Sedimentary Rocks Using Neuro-Visual Algorithms, Rajdeep Das

Doctoral Theses

Sediments and sedimentary rocks, covering almost sixty-five percent of exposed earth’s surface, contain features that allow us to interpret ancient depositional environments, and their role in the evolution of life on earth. Study of sedimentary rocks is also relevant to exploration of geo-economic resources including petroleum, ore minerals, and groundwater as well as research in environmental geology, anthropological studies, and the bi-directional relation of man and environment. One of the important and crucial steps in this study of the detrital sedimentary rocks like sandstone, is to know the size and shape of the grains/clasts (fragments of different minerals or rocks), …

Virtual Assistant For Pest Management, Viswanada Chakravarthy Karri

Master’s Dissertations

Effective pest identification and management are essential for ensuring agricultural productivity, especially in regions with limited expert access. This work proposes a virtual assistant based on a Retrieval-Augmented Generation (RAG) [1] framework to support pest management tasks. The system utilizes a multimodal dataset consisting of pest images and annotated textual interactions, adapted from the AgriLLaVA corpus [2]. The assistant combines retrieval mechanisms with generative language models to generate contextually grounded responses. It is designed for deployment on local hardware with limited computational resources, integrating open-source models for both retrieval and generation. Preliminary results suggest that this approach can provide accurate, …

2.5d Dual-Encoder U-Net For Lesion Segmentation In Chest Ct Scans, Jagannath Mukkara

Master’s Dissertations

Accurate segmentation of lesions in chest CT scans plays a vital role in diagnosing and monitoring pulmonary diseases such as COVID-19. In this, we introduce a novel 2.5D[1] dual-encoder U-Net model[2] that utilizes both the central slice and its neighboring slices to improve segmentation accuracy while keeping computational demands manageable. Our model incorporates residual connections[3] and feature fusion[4] to effectively merge multi-slice contextual information, overcoming the limitations found in traditional 2D and 3D methods. To ensure a reliable evaluation and avoid data leakage, we used patient-level data splitting. We validate our approach on a carefully curated chest CT dataset, showing …

Understanding Batch-Normalization In Deep Neural Networks, Pendyala Sai Srujan

Master’s Dissertations

Batch Normalization (BN) is a commonly used technique in various deep learning architectures for tasks such as image classification and object detection. It stabilizes and accelerates training by normalizing the activations of intermediate layers using mean and variance of the batch, allowing the use of higher learning rates and often improving generalization through implicit regularization. During inference, BN uses running estimates of batch statistics accumulated during training. However, if individual batches are not representative of the overall data distribution, these accumulated statistics may not accurately approximate the population statistics. This discrepancy can lead to a phenomenon known as **estimation shift**, …

A Study On Planar 2-Center Problem, Shince K. Baby

Master’s Dissertations

The Planar-k-Centre problem is an important problem in the class of Optimal Facility Location problems, where given n points in the planar, the objective is finding the smallest k congruent discs such that their union encompasses all points. This dissertation examines a variant of this problem in which the L1 metric is used to determine the distances rather than the standard Euclidean metric, which we call L1P2C and a closely related problem which we call Undirected k Square Coverage where there is no directional constraint for the k squares which contain the given set of points. We have found two …

Multi-Level Differentiable Moving Particles With Partition Of Unity, Jinjin He

Dartmouth College Master’s Theses

Representing implicit geometry with intricate features has long been a challenge. Recent advances in Implicit Neural Representations (INRs) have shown great promise in applications such as 3D reconstruction, inverse rendering, and dynamic surface evolution. These methods leverage neural networks to model complex shapes continuously, offering advantages in resolution and flexibility over traditional discrete representations. Despite their success, efficiently handling fine geometric details and evolving dynamic scenes remains an open problem.

We introduce a differentiable moving particle representation based on the multi-level partition of unity (MPU) to model dynamic implicit geometries efficiently. Our approach employs two types of particles—feature particles and …

Neat Vs. Scruffy: How Early Ai Researchers Classified Epistemic Cultures Of Knowledge Representation, Lindsay Poirier

Statistical and Data Sciences: Faculty Publications

Accounts of the history of symbolic artificial intelligence (AI) often categorize researchers and their programs through binary oppositions: logicist versus procedural, or alternatively, neat versus scruffy. These terms were often referenced throughout the 1970s and 1980s by AI researchers at institutions in the United States and Europe to document epistemic and esthetic differences in approaches to knowledge representation. In this article, I historicize and situate these binary oppositions, showing how researchers leveraged them in discourse to demarcate the beliefs and commitments underpinning certain AI approaches. I argue that the labels helped knowledge representation scholars identify, trouble, and grapple with the …

A Framework For Scalable And Controlled Hallucination Data Collection, Lin Ting Liang

Computer Science Senior Theses

This thesis addresses a key bottleneck in hallucination research: the scarcity and limitations of hallucination benchmark datasets. Existing datasets typically focus on a single type of hallucination and are expensive to produce due to the need for manual prompt creation and annotation. To overcome these challenges, we propose a novel mixture-of-experts (MoE) adversarial framework that actively induces hallucinations. Our framework employs three large language model (LLM) agents that iteratively and adversarially revise prompts to provoke hallucinated responses from a target question-answering model. It automates the generation of both intrinsic hallucinations (logical inconsistencies) and extrinsic hallucinations (inclusion of unverifiable external information). …

K-Mshc: Unmasking Minimally Sufficient Head Circuits In Large Language Models With Experiments On Syntactic Classification Tasks, Pratim Chowdhary, Peter Chin, Deepernab Chakrabarty

Computer Science Senior Theses

Understanding which neural components drive specific capabilities in mid-sized language models ($\leq$10B parameters) remains a key challenge. We introduce the $(\bm{K}, \epsilon)$-Minimum Sufficient Head Circuit ($K$-MSHC), a methodology to identify minimal sets of attention heads crucial for classification tasks as well as Search-K-MSHC, an efficient algorithm for discovering these circuits. Applying our Search-K-MSHC algorithm to Gemma-9B, we analyze three syntactic task families: grammar acceptability, arithmetic verification, and arithmetic word problems. Our findings reveal distinct task-specific head circuits, with grammar tasks predominantly utilizing early layers, word problems showing pronounced activity in both shallow and deep regions, and arithmetic verification demonstrating a …

Evaluating Vision Language Model Capabilities For Time Series Interpretation: An Empirical Study With Conversation Duration And Psychological Flourishing Data From The Studentlife Dataset, Jusung Park

Computer Science Senior Theses

This research investigates the capability of Vision Language Models (VLMs), specifically ChatGPT‑4o, to interpret and predict psychological outcomes based on visual representations of time series data. Leveraging conversation duration metrics and psychological flourishing scores from the StudentLife dataset, this study rigorously evaluates the predictive accuracy of VLMs using various methods, including zero‑shot on raw data, zero‑shot on graph data, few‑shot learning, qualitative labeling, and chain‑of‑thought reasoning. Despite multiple methodological enhancements, predictive performance remains modest, revealing significant challenges in quantitative interpretation of visualized temporal data by current multimodal models. We demonstrate that standardizing input tokens by using graph images rather than …

Character Relationship Prediction In Movies: Toward Emotionally-Aware Automatic Audio Descriptions, Seung Hyun Hahm

Computer Science Senior Theses

Automatic audio description (AD) systems support visually impaired audiences by narrating visual content, but they often fail to capture the interpersonal dynamics that underpin narrative understanding. In this work, we introduce a novel framework for character relationship prediction as a means of enriching audio descriptions with socially grounded context. Our contributions are threefold: (1) we propose the Character Relationship Module (CRM), which extends identity-aware video captioning with directed sentiment inference between character pairs; (2) we develop a scalable weak supervision pipeline that uses large language models to generate 669,520 relationship annotations across 202 films; and (3) we construct a complementary …

Bayesian Segmentation–Driven Informative Path Planning For Uav-Based Water Orthomosaic Generation, Phuc Dai Tran

Computer Science Senior Theses

This paper presents a comprehensive implementation

study of an informative path planning (IPP) algorithm

for autonomous water body detection and mapping using

unmanned aerial vehicles (UAVs). We propose a hybrid IPP

framework that seamlessly integrates Bayesian probabilistic

classification and real-time uncertainty quantification to achieve

superior flight efficiency and mapping accuracy compared

to conventional systematic coverage methods. Our approach

employs the state-of-the-art SegFormer deep learning segmentation

model in conjunction with log-odds-based orthomosaic

generation to produce high-fidelity water body maps under

diverse environmental conditions. Through random sampling of

the FloodNet dataset, we demonstrate that our IPP algorithm

maintains flight distance while achieving …

Motionteller: Multi-Modal Integration Of Wearable Time-Series With Llms For Health And Behavioral Understanding, Aiwei Zhang, Arvind Pillai, Andrew Campbell, Nicholas C. Jacobson

Computer Science Senior Theses

As wearable sensing becomes increasingly pervasive, a key challenge remains: how can we generate natural language summaries from raw physiological signals such as actigraphy - minute-level movement data collected via accelerometers? In this work, we introduce MotionTeller, a generative framework that natively integrates minute-level wearable activity data with large language models (LLMs). MotionTeller combines a pretrained actigraphy encoder with a lightweight projection module that maps behavioral embeddings into the token space of a frozen decoder-only LLM, enabling free-text, autoregressive generation of daily behavioral summaries.

We construct a novel dataset of 54,383 ⟨actigraphy, text⟩ pairs derived from real-world NHANES recordings, and …

An Efficient Algorithm For Finding High Harmonic Centrality Vertices In Graphs, John Balson

Computer Science Senior Theses

In this thesis we consider the problem of harmonic centrality in graphs. This measure is used widely in the study of real-world complex networks. In particular, we consider the problem of finding a high harmonic centrality vertex in time much faster than $O(mn)$, the time required to calculate the exact harmonic centrality of all vertices in a graph. The problem of calculating centrality measures faster has received much attention in recent years, since calculating the exact harmonic centrality of all vertices can be infeasible for large graphs; hence, faster algorithms are needed. This thesis proposes a new algorithm for finding …

An Approach To Stylometry Using Causal Language Models, Harrison F. Stropkay

Computer Science Senior Theses

We present a novel stylometric approach using large language models. By training separate models on individual authors' works, we find that each model achieves lower cross-entropy loss when predicting text from its training author compared to other authors' texts. Moreover, for any given text, the model trained on its true author’s corpus yields the lowest loss. We suggest that, in this way, a model trained on one author's works embodies the unique writing style of that author. We demonstrate our approach on works by eight known authors. This approach also confirms that R. P. Thompson wrote the well-studied 15th book …

Enhancing Iot Decentralization With Iota 2.0: A Dag-Based Fast Probabilistic Consensus Framework, Ayat N. Kadhum, Ahmed M. Al-Salih

Journal of Intelligent Informatics, Networking, and Cybersecurity

The integration of IoT and blockchain enhances security, trust, and data integrity but is hindered by security attacks, scalability, and high latency. In this work, a more efficient method of consensus using Directed Acyclic Graph (DAG)-based Fast Probabilistic Consensus (FPC) and Edwards-Curve Digital Signature Algorithm (EdDSA) is proposed to yield better security, efficiency, and decentralization. By eliminating mining, resource use is optimized, and consensus is hastened. Experimental results show a high throughput of 7228.05 Transactions Per Second (TPS), rapid consensus formation in just 7.62 rounds on average, and high adversary resilience, with the system successfully mitigating 89% of adversarial attacks. …

Cross-Modality Learning For Predicting Ihc Biomarkers From H&E-Stained Whole-Slide Images, Amit Das

Computer Science Senior Theses

Hematoxylin and Eosin (H&E) staining is a cornerstone of pathological analysis, offering reliable visualization of cellular morphology and tissue architecture for cancer diagnosis, subtyping, and grading. Immunohistochemistry (IHC) staining, an important ancillary study, provides molecular insights by detecting specific proteins within tissues, enhancing diagnostic accuracy, and improving treatment planning. However, IHC staining is costly, time-consuming, and resource-intensive, requiring specialized expertise. To address these limitations, this study proposes HistoStainAlign, a novel deep learning framework that predicts IHC staining patterns directly from H&E whole-slide images (WSIs) by learning joint representations of morphological and molecular features. The framework integrates paired H&E and IHC …

Machine Verification Of Correctness For A Concurrent Union-Find Object, Karun Ram

Computer Science Senior Theses

We consider a family of near-optimal randomized multiprocessor implementations for the union-find problem due to Jayanti and Tarjan–known as the Jayanti-Tarjan Randomized-Linking (JT-RL) union-find objects–and provide the first formal and fully machine-verified proof of their strong linearizability (i.e., correctness). Their algorithms are efficient both in theory and in practice: numerous benchmarking works demonstrate that they perform faster, or as fast, as all other known implementations for the union-find object on both CPUs and GPUs.

The correctness of the JT-RL algorithms is subtle, which motivates the need for formal verification. To this end, we first specify the JT-RL objects in TLA+, …

Efficient Signal Synthesis For Data Augmentation Using Generative Ai, Yagna Veera Narayan Kaasaragadda

Theses and Dissertations

Modern signal processing AI applications face increasing demands for diverse training data while operating under computational constraints. State-of-the-art generative models, though effective, often require prohibitive resources, limiting their deployment in real-time or embedded systems. This thesis proposes a computationally efficient framework for synthetic signal generation using a two-stage architecture that combines a Vector Quantized Variational Autoencoder (VQ-VAE) with either a decoder-only transformer or a discrete diffusion model. The VQ-VAE encodes high-dimensional signals into discrete latent tokens, significantly reducing model complexity while enabling symbolic sequence modeling. These discrete representations are then modeled using transformer-based autoregressive models or Score Entropy Discrete Diffusion …

From Lists To Infinite Scroll: A Comparative Analysis Of Youtube’S Two Recommendation Algorithms, Selimhan Dagtas

Theses and Dissertations

As social media platforms increasingly dominate information consumption, the role of recommendation algorithms in determining user experience has grown both in complexity and impact. This thesis investigates the behavioral patterns and algorithmic preferences embedded within YouTube’s recommendation systems, comparing long- form videos with the rapidly growing category of short-form content, YouTube Shorts. Through a combination of automated data collection, engagement metric analysis, emotional sentiment detection, and toxicity assessment, this study analyzes the evolution of content over successive recommendation depths. Using a controlled digital environment, the research explores how content recommendations change in response to user behavior, including varying watch times …

Video Game Hacking, A General Problem With Generalized Solutions, Luke Rowe

Master's Theses

As video games continue to get more popular and lucrative, the number of malicious actors seeking to exploit them grows with it. As this industry expands, so does the importance of securing games against cheating and abuse. This thesis aims to educate developers to help mitigate the abuse of video games by these malicious actors. The goal of this thesis is to provide a foundational framework for thinking like a hacker and how to make games harder to abuse once a hacker bypasses conventional anti-cheat software.

This thesis outlines some of the most common cheating methods and provides general context …

Ai Assisted Literature Review Tools For Undergraduates: Restrict, Embrace Or Curate?, Aaron Tay

Research Collection Library

As AI-driven literature review tools become widespread, academic librarians must grapple with a fundamental question—should we ban these tools, selectively curate their use, or embrace them fully? This keynote explores the three competing schools of thought shaping AI’s role in undergraduate literature reviews.

The Restrict school argues that students who have not proven capable of writing quality literature review should be restricted from use of such tools. Much like handing a preschooler a calculator before they understand basic arithmetic will affect the learning of arithmetic, premature use of such tools has the potential to shortcut the research learning process. If …

Enabling Automatic Solar Pv Array Identification Using Big Satellite Imagery, Qi Li, Keyang Yu, Carson Snow, Dong Chen

Computer Science Faculty Research and Publications

Recently, there has been a growing interest in automatically collecting distributed solar photovoltaic (PV) installation information in smart grid systems, including the quantity and locations of solar PV deployments, as well as their profiling information across a given geospatial region. Most recent approaches are still suffering low detection accuracy due to insufficient sample and principal feature learning when building their models and also separation of rooftop object segmentation and identification during their detection processes. In addition, they cannot report accurate multi-deployment results. To address these problems, we design a new system-SolarDetector+, which can automatically and accurately detect and profile distributed …

Runtime Backdoor Detection For Federated Learning Via Representational Dissimilarity Analysis, Xiyue Zhang, Xiaoyong Xue, Xiaoning Du, Xiaofei Xie, Yang Liu, Meng Sun

Research Collection School Of Computing and Information Systems

Federated learning (FL), as a powerful learning paradigm, trains a shared model by aggregating model updates from distributed clients. However, the decoupling of model learning from local data makes FL highly vulnerable to backdoor attacks, where a single compromised client can poison the shared model. While recent progress has been made in backdoor detection, existing methods face challenges with detection accuracy and runtime effectiveness, particularly when dealing with complex model architectures. In this work, we propose a novel approach to detecting malicious clients in an accurate, stable, and efficient manner. Our method utilizes a sampling-based network representation method to quantify …

Nexusgs: Sparse View Synthesis With Epipolar Depth Priors In 3d Gaussian Splatting, Yulong Zheng, Zicheng Jiang, Shengfeng He, Yandu Sun, Junyu Dong, Huaidong Zhang, Yong Du

Research Collection School Of Computing and Information Systems

Neural Radiance Field (NeRF) and 3D Gaussian Splatting (3DGS) have noticeably advanced photo-realistic novel view synthesis using images from densely spaced camera viewpoints. However, these methods struggle in few-shot scenarios due to limited supervision. In this paper, we present NexusGS, a 3DGS-based approach that enhances novel view synthesis from sparse-view images by directly embedding depth information into point clouds, without relying on complex manual regularizations. Exploiting the inherent epipolar geometry of 3DGS, our method introduces a novel point cloud densification strategy that initializes 3DGS with a dense point cloud, reducing randomness in point placement while preventing over-smoothing and overfitting. Specifically, …

Anonymous Communication In Quantum Networks, Anish Majumdar

Master’s Dissertations

In this dissertation, we explore the protocols enabling anonymous communication in quantum networks i.e. transmission of qubits from sender to receiver by creating or distributing Entangled states between them without disclosing their identities as sender or receiver to the other parties in the network. Existing methods uses classical, as well as quantum subprotocols to achieve anonymity.The Quantum sub-protocols include protocols for anonymous entanglement distribution using GHZ states, verification (not device-independent) of GHZ state that requires secure private classical channels, techniques for ϵ-anonymity(i.e. the other parties can at most guess a little ϵ amount better than a random guess about who …

Binary Document Filtering For Retrieval-Augmented Generation, Sreyan Saha

Master’s Dissertations

Retrieval-Augmented Generation (RAG) has become a popular technique to enhance Large Language Models (LLMs) with access to external information sources. However, the success of RAG systems critically depends on the relevance and quality of the retrieved documents. In particular, supplying irrelevant or noisy context can lead to degraded downstream generation quality. To address this, our project focuses on improving the document filtering stage in a RAG pipeline through binary relevance classification — deciding whether a retrieved document is suitable to include in the final context window based on its usefulness in directly answering the user query. We explore a wide …

Spos: An Attestation Solution For The Detection And Mitigation Of Point Of Sale Malware, Damian Singh Dhesi

Master's Theses

Securing 95% of card present transactions, accounting for billions of transactions a year, has made EMV the premier protocol for card-based payment. Created by and named after Europay, Mastercard, and Visa, the EMV protocol provides multiple solutions to resolve security concerns with the outdated, swipe-based, magnetic stripe payment. Such solutions are Chip and PIN which provides a more secure transaction at a significant time cost and EMV contactless which provides improved security to Chip and PIN at greater ease of use with its quick, tap-to-pay based payment. However, regardless of how secure the EMV protocol makes the card side of …

Adversarial Deep Reinforcement Learning For Tank Duel Simulation Using Lidar-Based Observations, Braedan Kennedy

Master's Theses

Previous research has demonstrated that reinforcement learning agents can learn to steer differential-drive robots around obstacles using 2D lidar scans as observations. However, these studies typically treat all range returns as undifferentiated obstacles—objects to avoid—without distinguishing between different object types. This thesis builds upon previous research by introducing an adversarial task in which an agent must interpret raw range readings to both avoid static obstacles and identify, pursue, and engage a hostile target.

To investigate this problem, this thesis introduces TankGame, a novel, lightweight 2D tank duel simulator. Each agent receives a 360° lidar scan, controls its motion via tread …