Digital Commons Network^™

Generalizable Adaptation For Vision-Language Models, Niloufar Alipour Talemi

All Dissertations

Vision-Language Models (VLMs) and Multimodal Large Language Models (MLLMs) have recently emerged as powerful frameworks for learning joint representations across visual and textual modalities. These models enable a wide range of applications, including visual recognition, multimodal reasoning, and visual question answering. However, adapting large pre-trained VLMs to downstream tasks while preserving their strong generalization ability remains a significant challenge, particularly under domain shifts or limited supervision. This dissertation focuses on developing methods for generalizable adaptation of VLMs, aiming to improve robustness, efficiency, and applicability across diverse tasks and environments.

First, this work introduces novel prompt learning strategies for adapting CLIP-style …

Detecting Doubt In Reflective Learning: A Learning Analytics Study With Large And Small Language Models, Eng Lieh Ouh, Kar Way Tan, Siaw Ling Lo, Yuhao Zhang

Research Collection School Of Computing and Information Systems

Reflective learning enhances understanding, especially when instructors promptly address difficulties raised in student reflections. Automated doubt detection can reduce time for instructors, yet existing classification approaches take substantial time for manual annotation and model training. This paper investigates whether large and small language models (LLMs, SLMs) can automate doubt detection without time-consuming training. Using a dataset of anonymized student reflections, we evaluate zeroshot, few-shot prompting, and multi-step reasoning against prior supervised classification baselines. We show that LLMs (GPT-4o, Claude-4, Gemini-2.5) surpass earlier F1 scores without prompting, while prompting further improves their performance. However, using proprietary LLMs can raise cost and …

Market Reactions To Deceptive Language In Fake News: Implications From Language Expectancy Theory And Transfer Learning, Ka Chung Ng, Ping Fan Ke, Ping Fan, Mike So, Tam, Kar Yan

Research Collection School Of Computing and Information Systems

The advent of generative artificial intelligence (AI) has heightened the proliferation of fake news. A key challenge is the limited real-world data to investigate the societal impact of fake news produced by generative AI. In this paper, we examine stock market reactions to financial news articles that exhibit stylometric similarity to human-crafted and AI-crafted fake financial news. Grounded in language expectancy theory, we employ a style-based transfer learning model, pre-trained to recognizing deceptive language employed in various types of fake news intricacies. We then apply this model to a comprehensive dataset of financial news, assigning a “veracity style score” to …

Dementia Detection In Low-Resource Languages: Evaluating Translation-Assisted Transfer Learning For Multilingual Clinical Assessment, Kylar A. Deloach

Honors Theses

Alzheimer's disease (AD) is a growing global health concern, with millions of people affected worldwide and cases expected to rise significantly in the coming decades. Early detection is critical for patient treatment and care, and recent advances in natural language processing (NLP) have shown promise in identifying linguistic markers associated with AD. However, most existing work has focused on English, leaving speakers of other languages with limited access to such tools. This study investigates how effective AD detection models trained on English data are at transferring to Greek, a low-resource language with limited dementia-related speech data available. We propose a …

Uncertainty Handling In Stock Market Prediction: A Fuzzy Markov Chain Approach, Alpa Singh Rajput, Arpan Singh Rajput

Neutrosophic Systems with Applications

Predicting the stock market is never easy because it is influenced by many uncertain and constantly changing factors such as economic conditions, investor behaviour, and global events. Traditional models like the Crisp Markov Chain (CMC) try to predict market movements by using fixed probabilities for different states like bullish, bearish, or stagnant. However, real markets do not behave in such a strict way—they often move gradually between states, which these models fail to capture. To overcome this limitation, this study introduces a Fuzzy Markov Chain (FMC) model, where fuzzy logic is used to handle uncertainty and allow smoother transitions between …

Modeling Agentic Artificial Intelligence Uncertainty In Agriculture Based 6generation: A Hybrid Q-Rung Orthopair Fuzzy Mcdm Methodology, Zekra Sakr, Mona Mohamed

Neutrosophic Systems with Applications

In era of advanced intelligent revolutions, the collaboration between intelligent technologies became imperative. For instance, integrating 6G communications with agentic artificial intelligence considered a catalyst to shift agriculture sector into optimized and intelligence sector. This integration resulted in transitioning the sector from static automation to autonomous, agent-based ecosystems. Accordingly, the efficiency roles for artificial intelligence agents (AIAs), deploying and selecting optimal AIA is important. Yet, selection process is still difficult because agricultural criteria are multifaceted and there are inherent environmental uncertainties. To address these challenges and bolster the selection process, this paper suggests a hybrid multi-criteria decision-making (MCDM) that bolstered …

A Deep Learning-Based Approach For Bot Detection In Trending Hashtags On X, Mehboob Hussain, Muhammad Rizwan Rashid Rana, Muhammad Imran, Muhammad Shoaib, Muhammad Hasaan Mujtaba

Makara Journal of Technology

The widespread presence of bots on social media platforms, such as X (formerly Twitter), poses a significant threat to the integrity of online information by facilitating the dissemination of misinformation and manipulating public discourse. This study proposes a robust deep learning-based framework, DeepBot, to detect bot participation in trending hashtags and discussions on X. The approach uses a dataset sourced from Kaggle, comprising user profile metadata, including follower count, tweet frequency, account verification status, and engagement metrics. The data were subjected to comprehensive preprocessing, including noise removal, part-of-speech (POS) tagging, and word embedding using the pre-trained GloVe model. RoBERTa is …

Smart Kitchen: Towards Real-Time Ai Systems For Cognitive Support In Daily Activities, Ruiqi Wang

McKelvey School of Engineering Graduate Student Theses & Dissertations

The rapid growth of the aging population and the rising prevalence of Subjective Cognitive Decline (SCD) highlight the need for continuous, unobtrusive assessment of functional cognition during everyday activities. Vision-based smart home systems offer a promising pathway for monitoring behavior and supporting independent living. However, enabling real-time cognitive assistance remains challenging. It requires not only accurately interpreting complex human behaviors to detect cognitive errors, but also supporting real-time deployment on resource-constrained edge devices and under dynamic wireless network conditions. This dissertation presents Smart Kitchen, an AI-driven system for real-time cognitive error detection through the monitoring of daily cooking activities. Under …

Drug Risk Knowledge Discovery For Western Medicines Based On Knowledge Graph Link Prediction, Jianxiang Wei, Ma Hengyuan Ma, Yuehong Sun, Wenwen Du, Letian Hu

Journal of Scientific Information Research

[Purpose/significance] The risk information contained in drug instructions is usually incomplete, and some new adverse reactions can only be discovered in actual clinical use. This paper proposes an information organization and knowledge discovery method for pharmacovigilance, in order to timely and accurately identify missing risk knowledge in drug instructions. [Method/process] Drug instructions of 8 152 Western medicines are collected as the research data; On the basis of ontology construction, data annotation, and model training, the UIE model is used to jointly extract entity and relationship triplets from the research data; A new knowledge graph link prediction method CompGCN-RotatE, is proposed, …

Research On Temporal Knowledge Graph Completion Method For Emergent Events Based On Bigru And Graph Contrastive Learning, Peng Wu, Zhenyu Lu, Xuechen Zhang

Journal of Scientific Information Research

[Purpose/significance] During emergencies, social media short texts contain critical information but are heavily interfered with by noise. Traditional static knowledge graph completion techniques struggle to effectively address their dynamic evolution and data sparsity issues, making it imperative to introduce temporal modeling methods. [Method/process] This study proposes a dynamic completion framework that combines the temporal feature capture capability of Bidirectional Gated Recurrent Units (BiGRU) with the noise-resistant representation learning advantages of Graph Contrastive Learning (GCL). At the completion level, the ConBiTE method is introduced, which captures temporal dependencies through self-attention mechanisms and BiGRU, while leveraging GCL to enhance the completion of …

Cognition Is Not Content: A Structural Account Of Processing Conditions, Griselda Poe

Publications and Research

Human cognition has been described in terms of content. This description becomes insufficient once artificial systems make output observable apart from subject attribution, intention, and relational context. Under this contrast condition, what becomes visible is a layered structure in which output, reconstruction, evaluation, and termination do not necessarily coincide. The same input may register as complete under one processing condition while remaining unresolved under another. This separability means that content-based description does not merely omit an additional variable: it can mislocate a processing difference as a difference in meaning, personality, intention, ability, or attitude. This is not a proposal for …

Ex-04-142 Modeling Distress And Evaluating Chatbot Safety For Suicide-Related Social Media Texts, Rehma Razzak

C-Day Computing Showcase

This project addresses the urgent need for transparent chatbot safety evaluations amid rising concerns about AI-facilitated self-harm. Using public social media datasets, we simulate two tasks: (1) detecting suicidal ideation via emotion-based risk scoring, and (2) stress-testing a support-style chatbot against 888 high-risk prompts, including euphemisms and “for a story” framing. A multi-label classifier trained on GoEmotions feeds emotion profiles into a logistic regression model to generate suicidality risk scores. These scores guide a local chatbot built with Ollama’s llama3, which analyzes user messages and steers responses toward safe, empathetic behavior. Evaluation shows ~90% of replies were safe or supportive. …

Grp-07-163 Smishguard: An Ai-Powered Framework For Sms Phishing Detection And Alert System For Vulnerable Users, Jiban Krisna Das

C-Day Computing Showcase

This research introduces SMISH-GUARD, a multi-layer framework for adaptive SMS phishing (smishing) detection that integrates language-aware semantic modeling, graph-theoretic campaign reasoning, and cost-sensitive decision calibration within a unified architecture. The framework integrates dual transformer encoders for multilingual semantic understanding with a heterogeneous temporal graph layer that captures relational attack signals such as shared URLs, sender reuse, and campaign propagation patterns. A cost-sensitive decision optimization module is further incorporated to translate probabilistic model outputs into risk-aware alert policies that explicitly balance false-positive inconvenience against the higher societal and financial cost of missed smishing attacks. The study evaluates four integrated datasets comprising …

Temperature-Induced Uncertainty In Fixed-Context Retrieval-Augmented Generation, Steven Zeng, Murat Kuzlu

Cybersecurity Undergraduate Research Showcase

This study examines how decoding temperature affects output uncertainty in a fixed-context retrieval-augmented generation (RAG) system. We define uncertainty as the semantic dispersion among repeated answers under the same fixed retrieved context, with greater dispersion interpreted as higher uncertainty. To isolate this answer-generation variability from retrieval drift, each question was paired with a fixed retrieved context, and repeated generations differed only in temperature. The experiment used nine questions drawn from a machine-learning textbook corpus, with three questions each at easy, moderate, and hard difficulty. Each question was evaluated at five temperatures (0.0, 0.25, 0.5, 0.75, and 1.0) over 30 iterations, …

Reading The Room: A Structural Account Of Constraint-Based Processing And Its Limits In Artificial Systems, Griselda Poe

Publications and Research

This paper does not introduce a new structure.

It makes explicit a structural relation implied but not directly stated in prior work.

Human interpersonal processing is grounded in the co-presence of internal constraint and dependency.

These do not exist as separable components.

They form a single structural condition.

This paper shows that what is commonly described as “reading the room” is not a unitary function.

It differs across configurations in how this condition is processed.

In EF configurations, the effect of this condition is expressed through translation into self-return: outputs are evaluated in terms of how they return to the …

Extreme Cf: A Structural Account Of Processing Absence And Category Absence In Cf-Foregrounded Configurations, Griselda Poe

Publications and Research

Existing frameworks of cognition and intervention assume that processing occurs, meaning is available, and evaluation can be applied.

This paper describes a configuration in which this assumption does not hold.

In CF-foregrounded processing where modulation is absent, processing occurs only when coherence is satisfied. When coherence is not satisfied, processing does not occur. Under this condition, no representation is generated, no evaluation applies, and no action selection is produced.

Within single-layer cognitive models, EF and CF are not distinguished. Within such models, CF does not exist as a condition.

This configuration has not been represented as a category within existing …

Pre-Experiential Constraint Reconstruction: A Structural Account Of The Prior Layer In Dialogue With Jung, Griselda Poe

Publications and Research

Experiences commonly described as pre-experiential memory—such as immediate recognition, familiarity without prior exposure, and the sense of "already knowing"—are typically interpreted as the retrieval of stored content. However, this storage-based account does not provide a structurally consistent explanation of how such content is preserved prior to experience or reactivated in a form that aligns with present input.

This paper extends the three-layer cognitive architecture developed in prior work in this series (Poe, 2026n), in which cognition operates across Core processing, Modulation, and an operationally inaccessible Prior layer that supplies constraints to all processing.

This paper proposes an alternative account in …

Dual History Enhancement With Hybrid Hypergraph-Graph Networks For Temporal Knowledge Graph Reasoning, Kailun Ye, Xiangjie Kong, Yuchao Zhang, Xuan Wang, Linan Zhu, Jiaxin Du, Guojiang Shen, Jianxin Li

Research outputs 2022 to 2026

Temporal Knowledge Graph (TKG) reasoning seeks to predict future events by analyzing historical data, where the effective leverage of both local and global historical facts proves crucial. Existing approaches employ graph neural networks (GNNs) and recurrent neural networks (RNNs) for local evolution patterns, complemented by statistical methods to enhance attention to global facts, demonstrating efficient predictive capabilities. However, traditional GNNs, constrained by their low-order neighborhood aggregation design, inherently fail to model potential high-order dependencies among facts. Furthermore, existing global history modeling approaches may introduce irrelevant historical information that interferes with prediction tasks. To address these limitations, we propose a Dual …

Does Culture Translate? A Case Study On Artificial Intelligencegenerated Text Concerning Japanese-Style Employment, Kanji Kitamura

Languages, Literatures, and Cultures Faculty Publications

Translation matters to international business when it has a discernible impact on firm performance or critical business processes. Current developments include machine translation, which is evolving in tandem with generative artificial intelligence (AI). On one hand, useful translation apps have become available to help the public communicate internationally. On the other hand, recent research concludes that even machine translation's best output still requires post-editing by a human. This argument implies that a machine-translated, grammatically accurate text may not always make sense to a human audience, possibly because of a semantic-pragmatic issue beyond the lexical level. This paper regards it as …

The Core-Modulation Architecture (Cma): A Structural Overview Of Hallucination As Structural Mismatch, Griselda Poe

Publications and Research

Hallucination is defined not as factual error but as a structural failure of alignment across target, layer, and constraint.

Within the Core-Modulation Architecture (CMA), cognition proceeds through layered processing and requires layer-specific termination conditions. Hallucination arises when Modulation-level termination is registered as completion while Core-level resolution has not occurred, producing structurally ungrounded but locally coherent outputs.

Detection is therefore structural rather than content-based, focusing on layer mismatch and termination failure.

This document presents a minimal structural account of hallucination within the CMA framework.

The Intricate Dance Of Emotions And Psychophysiology: Unveiling The Secrets Of Microexpressions, Jaiteg Singh, Deepika Sharma, Babar Shah, Sukhjit Singh Sehra, Farman Ali, Irfan Hussain

All Works

Background: Emotion recognition plays a pivotal role in behavioral analysis, mental health assessment, and human-computer interaction. Micro-expressions, which are brief and involuntary facial movements, offer valuable insights into concealed emotions. However, validating micro-expressions remains a challenge due to their subtlety and short duration. This study aims to enhance the validation and classification of micro-expressions by integrating electromyogram (EMG) signals with facial action units (AUs). Methods: EMG data was collected using the EMG Muscle Sensor Module V3.0, interfaced with an Arduino Mega 2560 microcontroller. To ensure signal clarity, various data filtration techniques were applied to eliminate noise, motion artifacts, and baseline …

The Expanding Digital Border: Ai, Surveillance, And The Fight For Justice, James Chesser

Immigration and Human Rights Law Review

As artificial intelligence transforms the mechanisms of immigration control, the modern border has become a digital filter—one governed less by geography and more by code. This Article examines the legal, technical, and ethical implications of AI-driven systems now central to global border enforcement, including biometric surveillance, algorithmic risk scoring, and predictive profiling. It explores how states use these technologies not only to manage irregular migration, but to compete for global talent—constructing migration regimes that reward capital and compliance while eroding transparency, due process, and equality.

Through an international and comparative lens, the piece highlights the expansion of algorithmic decision-making across …

Scalable Multi-Task Low-Rank Model Adaptation, Zichen Tian, Antoine Ledent, Qianru Sun

PhD Student’s Publications Collection

Scaling multi-task low-rank adaptation (LoRA) to a large number of tasks induces catastrophic performance degradation, such as an accuracy drop from 88.2% to 2.0% on DOTA when scaling from 5 to 15 tasks. This failure is due to parameter and representation misalignment. We find that existing solutions, like regularization and dynamic routing, fail at scale because they are constrained by a fundamental trade-off: strengthening regularization to reduce inter-task conflict inadvertently suppresses the essential feature discrimination required for effective routing. In this work, we identify two root causes for this trade-off. First, uniform regularization disrupts inter-task knowledge sharing: shared underlying knowledge …

Semantic Entanglement In Vector-Based Retrieval: A Formal Framework And Context-Conditioned Disentanglement Pipeline For Agentic Rag Systems, Nick Loghmani

iSchool - All Scholarship

Retrieval-Augmented Generation (RAG) systems deployed in agentic environments depend on the geometric properties of vector representations to retrieve contextually appropriate evidence for autonomous reasoning. When source documents conflate multiple topics within contiguous text regions, standard vectorization pipelines produce embedding spaces in which semantically distinct content occupies overlapping geometric neighborhoods — a condition we term semantic entanglement. This paper formalizes semantic entanglement as a model-relative measure of cross-topic overlap, defines an Entanglement Index (EI) as a quantitative proxy, and argues that higher EI is associated with reduced attainable Top-K retrieval precision under cosine similarity retrieval. We introduce the Semantic Disentanglement …

Stacked From One: Multi-Scale Self-Injection For Context Window Extension, Wei Han, Pan Zhou, Shuicheng Yan

Research Collection School Of Computing and Information Systems

The limited context window of contemporary large language models (LLMs) remains a primary bottleneck for their broader application across diverse domains. Although continual pre-training on long-context data offers a straightforward solution, it incurs prohibitive data acquisition and computational costs. To address this challenge, we propose SHAREDLLM, a novel framework based on multi-grained context compression and query-aware information acquisition. SHAREDLLM comprises two stacked short-context LLMs: a lower model serving as a compressor and an upper model acting as a decoder. The lower model compresses long inputs into compact, multi-grained representations, which are then forwarded to the upper model for context-aware processing. …

Thinktank-Me: A Multi-Expert Framework For Middle East Event Forecasting, Haoxuan Li, He Chang, Yunshan Ma, Yi Bin, Yang Yang, See-Kiong Ng, Tat-Seng Chua

Research Collection School Of Computing and Information Systems

Event forecasting is inherently influenced by multifaceted considerations, including international relations, regional historical dynamics, and cultural contexts. However, existing LLM-based approaches employ single-model architectures that generate predictions along a singular explicit trajectory, constraining their ability to capture diverse geopolitical nuances across complex regional contexts. To address this limitation, we introduce ThinkTank-ME, a novel Think Tank framework for Middle East event forecasting that emulates collaborative expert analysis in real-world strategic decision-making. To facilitate expert specialization and rigorous evaluation, we construct POLECAT-FOR-ME, a Middle East–focused event forecasting benchmark. Experimental results demonstrate the superiority of multi-expert collaboration in handling complex temporal geopolitical forecasting …

Study Of Output And Behavior Of Llms Using Confidence Framing In Prompt Engineering, Micah Parrilla

Doctoral Dissertations and Master's Theses

While prompt engineering is pivotal for shaping Large Language Model (LLM) outputs, the impact of confidence framing on behavioral calibration remains underexplored. This study investigates the ways in which psychological framing, utilizing techniques such as capability praise, role amplification, and doubt induction, affects linguistic tone, objective accuracy, and internal calibration. A 1,080-trial experimental matrix evaluated six diverse models across factual, logical, coding, and cyber security domains. Analysis using the Kruskal-Wallis H-test revealed highly significant behavioral shifts across all measured dimensions, providing conclusive evidence that the applied frames exert a substantial influence on model performance.

The findings identify a distinct cognitive …

Deep Learning Based Approaches For Low Cost Defense Detection, Adele J. Noel-Rickert

All NMU Master's Theses

Pulmonary fibrosis is a progressive interstitial lung disease characterized by the accumulation of fibrotic tissue within the lungs, leading to impaired respiratory function and reduced quality of life. Early detection is important for disease management; however, accurate diagnosis often relies on high-resolution computed tomography (CT), which may not be accessible in all clinical settings. Chest radiography provides a lower-cost and widely available imaging modality, but interpretation of chest X-rays for fibrotic disease can be challenging due to subtle radiographic patterns and overlapping anatomical structures. This thesis investigates the use of multimodal deep learning techniques to assist in pul- monary fibrosis …

From Spatial To Actions: Grounding Vision-Language-Action Model In Spatial Foundation Priors, Zhengshen Zhang, Hao Li, Yalun Dai, Zhengbang Zhu, Lei Zhou, Chenchen Liu, Dong Wang, Francis E. H. Tay, Sijin Chen, Ziwei Liu, Yuxiao Liu, Xinghang Li, Pan Zhou

Research Collection School Of Computing and Information Systems

Existing vision-language-action (VLA) models act in 3D real-world but are typically built on 2D encoders, leaving a spatial reasoning gap that limits generalization and adaptability. Recent 3D integration techniques for VLAs either require specialized sensors and transfer poorly across modalities, or inject weak cues that lack geometry and degrade vision-language alignment. In this work, we introduce FALCON (From Spatial to Action), a novel paradigm that injects rich 3D spatial tokens into the action head. FALCON leverages spatial foundation models to deliver strong geometric priors from RGB alone, and includes an Embodied Spatial Model that can optionally fuse depth, or pose …

Ai Interpretability In Healthcare Communication, Ananya Jeyappragash

Dartmouth College Master’s Theses

Artificial intelligence has increasingly been adopted in healthcare, largely for specialized tasks and under significant human oversight. The use of large black-box systems raises important concerns about transparency in high-stakes environments such as clinical decision-making. Clinical communication is fundamentally human-centered, and failures in judgment can have serious consequences for patient care. Overestimating the reasoning abilities of large language models may lead to undue trust in fabricated or “hallucinated” outputs, while rejecting AI-assisted tools altogether may preserve inefficient workflows and contribute to missed or delayed diagnoses. These concerns reflect a broader tradeoff between accuracy and interpretability: although more complex models may …