Digital Commons Network^™

Nexus: Network Exploration For Exploiting Unsafe Sequences In Multi-Turn Llm Jailbreaks, Javad Rafiei Asl, Sidhant Narula, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi

School of Cybersecurity Faculty Publications

Large Language Models (LLMs) have revolutionized natural language processing, yet remain vulnerable to jailbreak attacks—particularly multi-turn jailbreaks that distribute malicious intent across benign exchanges, thereby bypassing alignment mechanisms. Existing approaches often suffer from limited exploration of the adversarial space, rely on hand-crafted heuristics, or lack systematic query refinement. We propose NEXUS (Network Exploration for eXploiting Unsafe Sequences), a modular framework for constructing, refining, and executing optimized multi-turn attacks. NEXUS comprises: (1) ThoughtNet, which hierarchically expands a harmful intent into a structured semantic network of topics, entities, and query chains; (2) a feedback-driven Simulator that iteratively refines and prunes these chains …

A Review On Knowledge And Information Extraction From Pdf Documents And Storage Approaches, Salvador D. Atagong, Henri Tonnang, Kennedy Senagi, Mark Wamalwa, Komi M. Agboka, John Odindi

All Peer-Reviewed Publications

Introduction: Automating the extraction of information from Portable Document Format (PDF) documents represents a major advancement in information extraction, with applications in various domains such as healthcare, law, or biochemistry. However, existing solutions face challenges related to accuracy, domain adaptability, and implementation complexity. Methods: A systematic review of the literature was conducted using the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) methodology to examine approaches and trends in PDF information extraction and storage approaches. Results: The review revealed three dominant methodological categories: rule-based systems, statistical learning models, and neural network-based approaches. Key limitations include the rigidity of rule-based …

Ai Characterisations And Their Legal Implications, Jerrold Tsin Howe Soh

Research Collection Yong Pung How School Of Law

This chapter examines the difficult legal characterisation problems that artificially intelligent systems raise and explores how different characterisations of artificial intelligence (AI) shape practical legal outcomes. Three reasons are offered for the legal difficulty with characterising AI. First, answers to characterisation problems are inherently subjective and perspective-driven, particularly when the subject is an intangible technological system. Second, AI technology is especially difficult to define since the field typically proceeds on inexact anthropomorphic metaphors. Third, AI characterisation problems raise difficult sub-problems, particularly in determining how autonomous an AI system is. The chapter thus argues that a range of plausible AI characterisations …

Governing Intelligence: Singapore’S Evolving Ai Governance Framework, Jason G. Allen, Jane Loo, Jose Luna

Research Collection Yong Pung How School Of Law

This paper provides an outline analysis of the evolving governance framework for Artificial Intelligence (AI) in Singapore. Across the Singapore government, AI solutions are being adopted in line with Singapore’s “Smart Nation Initiative” to leverage technology to make impactful changes to the nation and the economy. In tandem, Singaporean authorities have been assiduous to release a growing number of governance documents, which we analyse together to chart the city-state’s approach to AI governance in international comparison. Characteristics of Singapore’s AI governance approach include an emphasis on consensusbuilding between stakeholders (particularly government and industry but also citizens) andvoluntary or “quasi” regulation, …

Toward Embodied Navigation Through Vision And Language, Muraleekrishna Gopinathan

Theses: Doctorates and Masters

Embodied AI is a challenging but exciting field in which a robot learns to interact with human-living spaces to perform various tasks. This thesis studies the embodied navigation problem in which a robotic agent navigates in a previously unseen indoor environment based on a challenging task. In particular, the Vision-and-Language Navigation (VLN) task requires a robot to navigate based on a descriptive human-language instruction. This thesis aims to improve VLN agents on four key aspects - their understanding of the environment, training via additional data, correcting navigational errors, and predicting the layout of the environment for better planning.

First, we …

Financial Named Entity Recognition: How Far Can Llm Go?, Yi-Te Lu, Yintong Huo

Research Collection School Of Computing and Information Systems

The surge of large language models (LLMs) has revolutionized the extraction and analysis of crucial information from a growing volume of financial statements, announcements, and business news. Recognition for named entities to construct structured data poses a significant challenge in analyzing financial documents and is a foundational task for intelligent financial analytics. However, how effective are these generic LLMs and their performance under various prompts are yet need a better understanding. To fill in the blank, we present a systematic evaluation of state-of-the-art LLMs and prompting methods in the financial Named Entity Recognition (NER) problem. Specifically, our experimental results highlight …

Insights In Adaptation: Examining Self-Reflection Strategies Of Job Seekers With Visual Impairments In India, Akshay Kolgar Nayak, Yash Prakash, Sampath Jayarathna, Hae-Na Lee, Vikas Ashok

Computer Science Faculty Publications

Significant changes in the digital employment landscape, driven by rapid technological advancements and the COVID-19 pandemic, have introduced new opportunities for blind and visually impaired (BVI) individuals in developing countries like India. However, a significant portion of the BVI population in India remains unemployed despite extensive accessibility advancements and job search interventions. Therefore, we conducted semi-structured interviews with 20 BVI persons who were either pursuing or recently sought employment in the digital industry. Our findings reveal that despite gaining digital literacy and extensive training, BVI individuals struggle to meet industry requirements for fulfilling job openings. While they engage in self-reflection …

Can Llms Beat Humans On Discerning Human-Written And Llm-Generated Science News, Dominik Soós, Meng Jiang, Jian Wu

Computer Science Faculty Publications

Science news is increasingly important in connecting scientists and the public by sharing discoveries and innovations. With the rise of large language models (LLMs), there is potential to automate science news creation, but concerns exist about the quality of LLM-generated news versus human-written news. This paper explores whether LLMs can outperform humans in distinguishing between human-written and LLM-generated news. Inspired by the Chain-of-Thought prompting method, we designed a simple yet effective variant called Guided Few-shot (GFS), which encodes the characteristics of news of two types with examples. Our experiments indicated that GFS with just a single example effectively boosted the …

From Philosophy To Nlu: Evolving Definitions With Research Hypotheses, Jian Wu, Sarah Rajtmajer

Computer Science Faculty Publications

Over the past decades, alongside advancements in natural language processing, significant attention has been paid to training models to automatically extract, understand, test, and generate hypotheses in open and scientific domains. However, interpretations of the term hypothesis for various natural language understanding (NLU) tasks have migrated from traditional definitions in the natural, social, and formal sciences. Even within NLU, we observe differences defining hypotheses across literature. In this paper, we overview and delineate various definitions of hypothesis. Especially, we discern the nuances of definitions across recently published NLU tasks. We highlight the importance of well-structured and well-defined hypotheses, particularly as …

From Philosophy To Nlu: Evolving Definitions Of Research Hypotheses, Jian Wu, Sarah Rajtmajer

Computer Science Faculty Publications

Over the past decades, alongside advancements in natural language processing, significant attention has been paid to training models to automatically extract, understand, test, and generate hypotheses in open and scientific domains. However, interpretations of the term hypothesis for various natural language understanding (NLU) tasks have migrated from traditional definitions in the natural, social, and formal sciences. Even within NLU, we observe differences defining hypotheses across literature. In this paper, we overview and delineate various definitions of hypothesis. Especially, we discern the nuances of definitions across recently published NLU tasks. We highlight the importance of well-structured and well-defined hypotheses, particularly as …

Neural Topic Modeling Via Contextual And Graph Information Fusion, Jiyuan Liu, Jiaxing Yan, Chunjiang Zhu, Xingyu Liu, Qing Li, Yanghui Rao

Computer Science Faculty Publications

Topic modeling is a powerful unsupervised tool for knowledge discovery. However, existing work struggles with generating limited-quality topics that are uninformative and incoherent, which hindering interpretable insights from managing textual data. In this paper, we improve the original variational autoencoder framework by incorporating contextual and graph information to address the above issues. First, the encoder utilizes topic fusion techniques to combine contextual and bag-of-words information well, and meanwhile exploits the constraints of topic alignment and topic sharpening to generate informative topics. Second, we develop a simple word co-occurrence graph information fusion strategy that efficiently increases topic coherence. On three benchmark …

Humans Vs. Llms On Open Domain Scientific Claim Verification: A Baseline Study, Benjamin Curtis, Stefania Dzhaman, Matthew Maisonave, Jian Wu

Computer Science Faculty Publications

Verifying scientific claims is challenging for the general public because most people lack domain knowledge. Manual verification by subject domain experts is accurate, but it is obviously not scalable to meet the rising number of scientific claims on the Web. Whether the emerging large language models and large reasoning models can be used for scientific claim verification, and how their performances compare to humans, are still research questions. To this end, we developed a new benchmark MSVEC2 that consists of 138 claims from credible fact verification websites and science news outlets. Two tasks were given to both human and LLM …

An Analytical Review Of Preprocessing Techniques In Bengali Natural Language Processing, Sovon Chakraborty, Protiva Das, Shakib Mahmud Dipto, Md Aktaruzzaman Pramanik, Jannatun Noor

Computer Science Faculty Publications

Research in Bengali Natural Language Processing (BNLP) is rapidly expanding. Despite being one of the most widely spoken languages in the world, BNLP research remains insufficient, particularly in Bengali speech recognition. The languages rich morphology, agglutinative structure, and diverse dialects make text and speech processing especially challenging. However, these challenges can be addressed with effective preprocessing techniques. Various organizations in Bangladesh and West Bengal are integrating Natural Language Processing (NLP) into their services, but without a thorough understanding of preprocessing, these implementations remain incomplete. Applying proper preprocessing techniques to the Bengali language will serve as a foundation for developing robust …

Position: Benchmarking Is Broken - Don't Let Ai Be Its Own Judge, Zerui Cheng, Stella Wohnig, Ruchika Gupta, Samiul Alam, Tassallah Abdullahi, João Alves Ribeiro, Christian Nielsen-Garcia, Saif Mir, Siran Li, Jason Orender, Seyed Ali Bahrainian, Daniel Kirste, Aaron Gokaslan, Carsten Eickhoff, Pramod Viswanath, Ruben Wolff

Computer Science Faculty Publications

The meteoric rise of Artificial Intelligence (AI), with its rapidly expanding market capitalization, presents both transformative opportunities and critical challenges. Chief among these is the urgent need for a new, unified paradigm for trustworthy evaluation, as current benchmarks increasingly reveal critical vulnerabilities. Issues like data contamination and selective reporting by model developers fuel hype, while inadequate data quality control can lead to biased evaluations that, even if unintentionally, may favor specific approaches. As a flood of participants enters the AI space, this "Wild West" of assessment makes distinguishing genuine progress from exaggerated claims exceptionally difficult. Such ambiguity blurs scientific signals …

Quickque: Enabling Quick Access To Information In User Reviews For Screen Reader Users, Mohan Sunkara, Akshay Kolgar Nayak, Sandeep Kalari, Yash Prakash, Sampath Jayarathna, Hae-Na Lee, Vikas Ashok

Computer Science Faculty Publications

Efficiently perusing online customer reviews is presently challenging for blind users, who rely on a screen reader that supports predominantly one-dimensional narration of content via keyboard shortcuts. To address this, with restaurant reviews as seminal case study, we developed QuickCue, a browser extension prototype that enables screen reader users to quickly obtain the positives and negatives regarding different aspects of a restaurant (e.g., food quality, hygiene, ambiance), without having to sift through numerous reviews containing redundant information. At its core, QuickCue utilizes a large language model to perform aspect and sentiment-based joint classification of reviews to group them based on …

Accessmenu: Enhancing Usability Of Online Restaurant Menus For Screen Reader Users, Nithiya Venkatraman, Akshay Kolgar Nayak, Suyog Dahal, Yash Prakash, Hae-Na Lee, Vikas Ashok

Computer Science Faculty Publications

Online food ordering has become commonplace due to its convenience. The wide variety of culinary choices, combined with fast and economical door-delivery services, encourages more people to order food online. To facilitate this process, food vendors, including restaurants, often provide full menus on their websites, typically in visual formats such as images or PDFs. While this is convenient for sighted users, blind and visually impaired (BVI) individuals face significant challenges accessing these visual menus with their screen reader assistive technology. An interview study with 12 BVI screen reader users revealed that present assistive tools do not adequately satisfy the needs …

Characterizing Language Use In Online Accessibility Discussion Forums, Nithiya Venkatraman, Anand Ravi Aiyer, Yash Prakash, Sampath Jayarathna, Hae-Na Lee, Vikas Ashok

Computer Science Faculty Publications

Discussion forums are one of the favored platforms for knowledge sharing. Given their popularity, copious research exists on understanding the linguistic and behavioral characteristics of forum conversations, so as to inform the design of many downstream applications including discourse visualization, sentiment analysis, and question answering. However, prior investigations have mainly focused on general forums designed primarily for sighted users, and as such the applicability of their findings to dedicated accessibility discussion forums frequented by blind screen reader users remains unanswered. To bridge this knowledge gap and facilitate the development of better-informed assistive technologies for blind people, we investigated language use …

Nexus: Network Exploration For Exploiting Unsafe Sequences In Multi-Turn Llm Jailbreaks, Javad Rafiei Asl, Sidhant Narula, Mohammad Ghasemigol, Eduardo Blanco, Daniel Takabi

Computer Science Faculty Publications

Large Language Models (LLMs) have revolutionized natural language processing, yet remain vulnerable to jailbreak attacks—particularly multi-turn jailbreaks that distribute malicious intent across benign exchanges, thereby bypassing alignment mechanisms. Existing approaches often suffer from limited exploration of the adversarial space, rely on hand-crafted heuristics, or lack systematic query refinement. We propose NEXUS (Network Exploration for eXploiting Unsafe Sequences), a modular framework for constructing, refining, and executing optimized multi-turn attacks. NEXUS comprises: (1) ThoughtNet, which hierarchically expands a harmful intent into a structured semantic network of topics, entities, and query chains; (2) a feedback-driven Simulator that iteratively refines and prunes these chains …

A Proposed Ehrenfeucht-Fraïssé Game Model For Natural Language Processing Generative Adversarial Networks, Don Li

Anthós

Large Language Models (LLM’s) (e.g., ChatGPT) constitute both a significant research area and commercial application of AI. Current major LLM’s are built on Generative Pre-Trained Transformer (GPT) neural network architecture to perform natural language processing (NLP) tasks. Generative Adversarial Network (GAN) is another popular neural network architecture, which leverages a zero-sum game between constituent neural networks within the architecture to train the GAN, and is widely used for visual data applications. This article proposes a new GAN architecture for NLP: an EF-GAN whose underlying algorithm uses Ehrenfeucht–Fraïssé (EF) games, a game-theoretic approach from model theory to determine elementary equivalence of …

Performance Of Standard Medical Mllms On Ecg Image Data, Prisha Anil

Masters Theses

This work presents a structured benchmarking study of multimodal large language models (MLLMs) applied to electrocardiogram (ECG) interpretation tasks. We evaluate three representative architectures: MedGemma, HuatuoGPT-Vision, and LLaVA-Med, across progressive experimental stages involving text-only structured prompt normalization, text–image fusion with ECG plots, and full multimodal fusion incorporating time-series signals. A standardized five-section cardiology prompt was designed to enforce consistent output structure and SCP-code alignment, enabling reproducible metric computation across models. Quantitative evaluation using BERTScore, token-level F1, and diagnostic accuracy demonstrates that HuatuoGPT-Vision achieves the highest semantic and diagnostic alignment, while MedGemma exhibits superior formatting stability and reproducibility. In contrast, LLaVA-Med …

Relationship Extraction Using Retrieval Augmented Generation For Biomedical Dataset, Jannat Jannat

Theses and Dissertations

With the increasing number of structured and unstructured data, obtaining reliable information effectively has become crucial. In the biomedical domain, extracting information from the scientific papers is crucial in order to stay up-to-date with accurate information, given the increased pace by which new research studies are published. This work focuses on identifying relationships between entities that are extracted from the abstracts and titles of biomedical research papers. In this work, we developed a Retrieval Augmented Generation (RAG) based system to automatically identify relations between biomedical entities. We evaluate multiple open source Large Language Models (LLMs) and the number of examples …

Pushing The Boundaries Of Large Language Models: Innovations And Limitations In Nlp, Finance, And Mathematics, A M Muntasir Rahman

Dissertations

Large Language Models (LLMs) have emerged as transformative tools across a spectrum of domains, yet their practical deployment reveals a blend of remarkable potential and notable limitations. This research explores innovative methodologies to extend the capabilities of LLMs while addressing critical challenges in their evaluation and application. By leveraging rule-based approaches, the in-context learning capabilities of LLMs, and human-in-the-loop validation across three focused studies, this research introduces robust strategies for dataset synthesis, model enhancement, and model assessment in three distinct domains: natural language processing, financial sentiment analysis, and mathematical reasoning

The first study proposes an efficient data augmentation framework, EASE, …

Measuring And Improving Api Usability And Quality: A Comprehensive Framework And Empirical Study, Sultan Alanazy

Computer Science and Engineering Theses and Dissertations

Cloud computing provides on-demand access to flexible computing resources, enabling rapid application deployment without substantial infrastructure investment. Application Programming Interfaces (APIs) play an important role in ensuring the success of cloud applications. The primary users of APIs are the extensive community of application programmers who search, read, and understand APIs before integrating them into their applications or systems. In addition, developers often turn to online API support when seeking help. Problems in such support can result in incorrect API usage and integration problems. There is an urgent need to measure API usability and support issues to identify, characterize, and assess …

Enhancing Graph Neural Networks By Editing Graphs, Jiayu Li

Dissertations - ALL

Graphs are pervasive in both the natural world and various domains of science and engineering. Numerous advanced classifiers, such as Graph Neural Networks (GNNs), have been developed to perform node classification on these graphs. However, as graphs become denser with an increasing number of edges, GNNs often suffer from suboptimal generalization performance due to the presence of task-irrelevant connections. These redundant connections can introduce noise, consume excessive computational resources, and degrade performance. Identifying and preserving critical connections in large-scale graphs, while pruning unnecessary ones, is crucial for enhancing the efficiency and accuracy of GNNs in node classification, particularly for GCNs. …

Privacy Protection In Machine Learning Via Exploitation Of Constrained Adversarial Evasion, Brian Testa

Dissertations - ALL

Deep neural networks are extensively applied to real-world tasks. In many cases, the data feeding these tasks are human generated content, which emphasizes the criticality of privacy and data protection. The diverse modalities of this user data provide fertile ground for 3rd parties to monetize a user’s data. This work considers two such modalities: user speech when interacting with smart speaker voice assistants (VAs) and images shared with online service providers. In these cases, the user would like to support some form of machine learning (ML) inference without allowing others. A user interacting with a smart speaker would like the …

Improving Clinical Information Extraction From Electronic Health Records: Leveraging Large Language Models And Evaluating Their Outputs, Kriti Bhattarai

McKelvey School of Engineering Graduate Student Theses & Dissertations

Accurate extraction of clinical entities and phenotypes from unstructured electronic health record (EHR) text is crucial for various clinical research tasks, including cohort identification, tracking temporal patterns in disease progression and deciding treatment course. However, this task remains challenging due to the complexity and ambiguity of medical language. This dissertation explores the application of advanced generative pre-trained transformer (GPT) models, such as GPT-4, GPT-3.5-turbo, Llama-3.1, Llama-3 and Flan-T5, for clinical entity and phenotype extraction from EHRs. Building upon these findings, this dissertation also investigates a hybrid approach where integration of external knowledge sources, such as Unified Medical Language System (UMLS) …

Implication Of Generative Ai On Education And Research, Riddhi Gupta

The Journal of Purdue Undergraduate Research

No abstract provided.

Utilizing Large Language Models To Synthesize Product Desirability Datasets, John D. Hastings, Sherri Weitl-Harms, Joseph Doty, Zachary L. Myers, Warren Thompson

Research & Publications

This research explores the application of large language models (LLMs) to generate synthetic datasets for Product Desirability Toolkit (PDT) testing, a key component in evaluating user sentiment and product experience. Utilizing gpt-4o-mini, a cost-effective alternative to larger commercial LLMs, three methods, Word+Review, Review+Word, and Supply-Word, were each used to synthesize 1000 product reviews. The generated datasets were assessed for sentiment alignment, textual diversity, and data generation cost. Results demonstrated high sentiment alignment across all methods, with Pearson correlations ranging from 0.93 to 0.97. Supply-Word exhibited the highest diversity and coverage of PDT terms, although with increased generation costs. Despite minor …

Advanced Models For Linking Process In Data Washing Machine, Bushra Sajid

Theses and Dissertations

Entity Resolution (ER) is a critical process in data integration and quality improvement that identifies and links multiple records referring to the same real-world entity. As data volumes and heterogeneity increase, traditional ER methods face new challenges, prompting research into more advanced techniques. The Proof-of-Concept Data Washing Machine (DWM), developed under the NSF DART Data Life Cycle and Curation research theme, aims to automatically detect and correct data quality errors through unsupervised entity resolution. Recent research focuses on enhancing DWM's effectiveness by replacing rule-based methods with machine learning and deep learning approaches, particularly in the linking process. Deep learning models, …

La Creatividad En Peligro: Como La Inteligencia Artificial Es Un Reto Para Los Artistas., Nathaly Cisneros

Capstones

Los artistas digitales han creado obras maestras que nos han dejado sin aliento con sus pinceles digitales, lápices y pinturas. Desde retratos que parecen saltar de la pantalla hasta paisajes que nos transportan a mundos desconocidos, su arte ha sido una fuente constante de inspiración.

Pero en los últimos años, una nueva fuerza ha comenzado a cambiar el juego. La inteligencia artificial ha estado avanzando a pasos agigantados y ahora se perfila como una amenaza para el futuro de los artistas digitales. ¿Qué significa esto para el arte y la creatividad?

Link: https://docs.google.com/document/d/1xe8UxDMekX_SwiIppyt_JppK8M-lB-YWNWGyeyShlJM/edit?usp=sharing