Digital Commons Network^™

Retrieval-Augmented Feature Generation For Domain-Specific Classification, Xinhao Zhang, Jinghan Zhang, Fengran Mo, Dakshak Keerthi Chandra, Yu-Zhong Chen, Fei Xie, Kunpeng Liu

Computer Science Faculty Publications and Presentations

Feature generation can significantly enhance learning outcomes, particularly for tasks with limited data. An effective way to improve feature generation is to expand the current feature space using existing features and enriching the informational content. However, generating new, interpretable features usually requires domain-specific knowledge on top of the existing features. In this paper, we introduce a Retrieval-Augmented Feature Generation method, RAFG, to generate useful and explainable features specific to domain classification tasks. To increase the interpretability of the generated features, we conduct knowledge retrieval among the existing features in the domain to identify potential feature associations. These associations are expected …

“Lost-In-The-Later”: Framework For Quantifying Contextual Grounding In Large Language Models, Yufei Tao, Adam Hiatt, Rahul Seetharaman, Ameeta Agrawal

Computer Science Faculty Publications and Presentations

Large language models (LLMs) are capable of leveraging both contextual and parametric knowledge but how they prioritize and integrate these sources remains underexplored. We introduce CoPE, a novel framework for systematically quantifying contextual grounding in LLMs. CoPE distinguishes between contextual knowledge (CK) and parametric knowledge (PK), enabling fine-grained attribution across languages and tasks. Using our newly created MultiWikiAtomic dataset in English, Spanish, and Danish, we analyze how LLMs integrate context, prioritize information, and incorporate PK in open-ended question answering. We find that across models and languages, only around 50 to 76 percent of outputs are grounded in the given context, …

Scitopic: Enhancing Topic Discovery In Scientific Literature Through Advanced Llm, Pengjiang Li, Zaitian Wang, Xinhao Zhang, Ran Zhang, Lu Jiang, Pengfei Wang, Yuanchun Zhou

Computer Science Faculty Publications and Presentations

Topic discovery in scientific literature provides valuable insights for researchers to identify emerging trends and explore new avenues for investigation, facilitating easier scientific information retrieval. Many machine learning methods, particularly deep embedding techniques, have been applied to discover research topics. However, most existing topic discovery methods rely on word embedding to capture the semantics and lack a comprehensive understanding of scientific publications, struggling with complex, high-dimensional text relationships. Inspired by the exceptional comprehension of textual information by large language models (LLMs), we propose an advanced topic discovery method enhanced by LLMs to improve scientific topic identification, namely SciTopic. Specifically, we …

Binoculars To Bytes: Development And Field Validation Of An Ai-Driven System For Avian Monitoring, Christian J. Dupree

Graduate Student Theses, Dissertations, & Professional Papers

Autonomous camera-trap arrays coupled with artificial-intelligence (AI) vision can lift bird monitoring beyond the spatial, temporal, and labor limits of traditional field surveys. We present Binoculars to Bytes (B2B), an open-source pipeline that turns 180–360° time-lapse imagery into analysis-ready avian data. At Freezout Lake Wildlife Management Area (Montana, USA) the system ran four-hour morning deployments during spring and fall migrations (2023–2024). A YOLO-NAS detector, incrementally refined with a “Specialized Localized Iterative Model” workflow, quadrupled local accuracy and, after confidence-based species-binning, cut false-positive rates in half. Daily AI species lists were benchmarked against contemporaneous eBird citizen-science checklists and recovered ≥ …

Investigating The Impact Of Aerial Firefighting On Rate Of Wildfire Spread, Lindsay Ann Wiard

Graduate Student Theses, Dissertations, & Professional Papers

Aerial retardant drops are widely used in wildfire suppression, yet their effectiveness in slowing fire spread remains difficult to quantify at scale. This study evaluates the impact of aerial suppression on wildfire rate of spread (ROS) using a modeling framework that incorporates both observed (real) and counterfactual (synthetic) drop locations from a sample of 62 wildfires in Oregon. Synthetic drops were generated to simulate a no-suppression baseline, allowing us to compare changes in ROS in the presence and absence of suppression. We trained two random forest classifiers: one using both real and synthetic drops (the full model), and another using …

Adapting Installation Instructions In Rapidly Evolving Software Ecosystems, Haoyu Gao, Christoph Treude, Mansooreh Zahedi

Research Collection School Of Computing and Information Systems

files play an important role in providing installation-related instructions to software users and are widely used in open source software systems on platforms such as GitHub. Software projects evolve rapidly alongside their dependencies in dynamic software ecosystems, requiring frequent updates to installation instructions. These instructions are crucial for users to start with a software project. Despite their significance, there is a lack of systematic understanding regarding the documentation efforts invested in README files and the triggers behind them. To fill the research gap, we conducted a qualitative study, investigating 400 GitHub repositories with 1,163 README commits that focused on updates …

Fedart: A Neural Model Integrating Federated Learning And Adaptive Resonance Theory, Shubham Pateria, Budhitama Subagdja, Ah-Hwee Tan

Research Collection School Of Computing and Information Systems

Federated Learning (FL) has emerged as a promising paradigm for collaborative model training across distributed clients while preserving data privacy. However, prevailing FL approaches aggregate the clients’ local models into a global model through multi-round iterative parameter averaging. This leads to the undesirable bias of the aggregated model towards certain clients in the presence of heterogeneous data distributions among the clients. Moreover, such approaches are restricted to supervised classification tasks and do not support unsupervised clustering. To address these limitations, we propose a novel one-shot FL approach called Federated Adaptive Resonance Theory (FedART) which leverages self-organizing Adaptive Resonance Theory (ART) …

Bi-Objective Dynamic Tugboat Scheduling With Speed Optimization Under Stochastic And Time-Varying Service Demands, Xiaoyang Wei, Hoong Chuin Lau, Zhe Xiao, Xiuju Fu, Xiaocai Zhang, Zheng Qin

Research Collection School Of Computing and Information Systems

With the growing emphasis on green shipping to reduce the environmental impact of maritime transportation, optimizing fuel consumption with maintaining high service quality has become critical in port operations. Ports are essential nodes in global supply chains, where tugboats play a pivotal role in the safe and efficient maneuvering of ships within constrained environments. However, existing literature lacks approaches that address tugboat scheduling under realistic operational conditions. To fill the research gap, this is the first work to propose the bi-objective dynamic tugboat scheduling problem that optimizes speed under stochastic and time-varying demands, aiming to minimize fuel consumption and manage …

The Effectiveness Of Local Updates For Decentralized Learning Under Data Heterogeneity, Tongle Wu, Zhize Li, Ying Sun

Research Collection School Of Computing and Information Systems

We revisit two fundamental decentralized optimization methods, Decentralized Gradient Tracking (DGT) and Decentralized Gradient Descent (DGD), with multiple local updates. We consider two settings and demonstrate that incorporating local update steps can reduce communication complexity. Specifically, for  $\mu$-strongly convex and $L$-smooth loss functions, we proved that local DGT  achieves communication complexity {}{$\tilde{\mathcal{O}} \Big(\frac{L}{\mu(K+1)} + \frac{\delta + {}{\mu}}{\mu (1 - \rho)} + \frac{\rho }{(1 - \rho)^2} \cdot \frac{L+ \delta}{\mu}\Big)$}, where $K$ is the number of additional local update}, $\rho$ measures the network connectivity and $\delta$ measures the second-order heterogeneity of the local losses. Our results reveal the tradeoff between communication and …

An Agent-Based Computational Finance Simulation Model To Study Market Efficiency, Wei Feng, Keng Siau, Wee-Yeap Lau, Lim-Thye Goh, Haonan Chen

Research Collection School Of Computing and Information Systems

The advancement of computational modeling, data systems, and digital infrastructure has enabled the rise of agent-based computational finance (ACF). This study models interactions among heterogeneous investors. By embedding behavioral logics such as environmental, social, and governance (ESG) preferences and volatility thresholds, the model captures microstructural dynamics under different trading rules. Using ACF, the authors compare transaction plus 0 day (T+0) to transaction plus 1 day (T+1). Results show that T+0 improves price discovery, deepens liquidity, and reduces transaction costs. From a computational perspective, this research contributes to ACF by showing how policy logic and investor heterogeneity can be encoded and …

More Effective Javascript Breaking Change Detection Via Dynamic Object Relation Graph, Dezhen Kong, Jiakun Liu, Chao Ni, David Lo, Lingfeng Bao

Research Collection School Of Computing and Information Systems

JavaScript libraries are characterized by their widespread use, frequent code changes, and a high tolerance for backward incompatible changes. Awareness of such breaking changes can help developers adapt to version updates and avoid negative impacts. Several tools have been targeted to or can be used to detect breaking change detection in the JavaScript community. However, these tools detect breaking changes using different ways, and there are currently no systematic reviews of these approaches. From a preliminary study on popular JavaScript libraries, we find that existing approaches, including simple regression testing, model-based testing and type differencing cannot detect many breaking changes …

Don’T Complete It! Preventing Unhelpful Code Completion For Productive And Sustainable Neural Code Completion Systems, Zhensu Sun, Xiaoning Du, Fu Song, Shangwen Wang, Mingze Ni, Li Li, David Lo

Research Collection School Of Computing and Information Systems

Currently, large pre-trained language models are widely applied in neural code completion systems. Though large code models significantly outperform their smaller counterparts, around 70% of displayed code completions from Github Copilot are not accepted by developers. Being reviewed but not accepted, their help to developer productivity is considerably limited and may conversely aggravate the workload of developers, as the code completions are automatically and actively generated in state-of-the-art code completion systems as developers type out once the service is enabled. Even worse, considering the high cost of the large code models, it is a huge waste of computing resources and …

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 …

Analysis Of Bin Packing Variants, Kyle T. Ambrose

UNF Graduate Theses and Dissertations

The Bin Packing problem is a classic and widely studied optimization problem that arises naturally in applications like manufacturing, logistics, and memory allocation, where space and resource constraints are critical. In this thesis, we first demonstrate the NP-completeness of Bin Packing via a reduction from Three-Dimensional Matching, establishing its foundational complexity. We then survey core heuristics for the one-dimensional case and extend our analysis to two and three-dimensional variants, including both offline and online strategies. Special attention is given to stochastic bin packing, where item sizes are modeled as random variables drawn from distributions such as uniform, truncated normal, and …

Three-Sided Skyline Counting Queries, Suruchi Kushwaha

Dissertations, Master's Theses and Master's Reports

A two-dimensional point p=(p.x,p.y) dominates another point p'=(p'.x,p'.y) if p.x ≥ p'.x and p.y>p'.y or p.x>p'.x and p.y ≥ p'.y. The skyline of a point set P is a subset P' ⊆ P such that every point in P' is not dominated by any other point in P. An orthogonal skyline counting query Q on a set of points P asks for the number of points on the skyline of P ⋂ Q.

In this work we study data structures that support orthogonal skyline counting queries in the special case when the query range is bounded on three …

Finding Antipatterns Across Languages With Abstract Syntax Trees, Daniel T. Masker

Dissertations, Master's Theses and Master's Reports

Finding antipatterns in student code is a difficult task that is useful for helping beginner programmers. Antipatterns are common mistakes that students make while writing code. Code critiquers are tools that find antipatterns and provide rich, immediate feedback to students, even when professors aren’t available. WebTA is a code critiquer that finds antipatterns using regular expressions (regex), error messages, and language-specific abstract syntax trees (ASTs). Each of these tools has obstacles to antipattern searching that are difficult to overcome. Regex is without context, limiting the patterns it can recognize. Additionally, even experienced users have difficulty reading and debugging regex. Error …

Topics On Ai Fairness Preferences In Kidney Transplantation, Mukund Telukunta

Doctoral Dissertations

Modern kidney transplantation incorporates artificial intelligence (AI) decision-support systems which exhibit social discrimination due to biases inherited from training data. Although researchers have proposed various group-based fairness notions to assess biases in AI, it remains uncertain which criterion is most suitable for evaluating biases in such complex healthcare systems. This dissertation explores human perception of fairness to identify the most appropriate fairness criterion for assessing AI tools in kidney transplantation, focusing on the preferences of non-expert (e.g. public, patients) stakeholders. The study examines two distinct AI systems employed in kidney transplantation: a classification model and a regression model. Human subject …

Machine Learning Models For Location Prediction, Message Routing And Path Planning For Rescue Of Underground Miners, Abhay Goyal

Doctoral Dissertations

Self-rescue during underground mine disasters is vital for miner safety. Evolving hazards and post-disaster conditions demand solutions that enable navigation under severe communication and computational constraints. Centralized systems often fail in such rugged settings, while decentralized methods—particularly Delay Tolerant Networks (DTNs), proven in battlefields and space missions—offer distinct advantages for underground applications. This research addresses five core challenges: (i) predicting miners’ next locations on low-power devices using points of interest and movement sequences; (ii) delivering timely updates on safe routes, evacuation zones, and hazardous areas; (iii) evaluating energy efficiency and comparing graph-based approaches to existing methods; (iv) enabling edge-ready frameworks, …

Balancing Performance And Efficiency: An Autoencoder Approach To Api-Based Malware Detection, Selma Bouraoui

Graduate Theses/Dissertations

The constant evolution of malware presents a critical challenge to today's interconnected world. It poses an increasing threat on different scales, spanning from individuals, organizations to critical infrastructures such as government’s security. Hackers continuously develop new techniques to evade detection methods. When confronted with the high volume and variety of malware, conventional approaches tend to struggle to perform in robust, accurate and timely manner. This thesis explores the application of deep learning methods to improve malware detection and classification techniques. By analyzing API call sequences, the proposed approach leverages Autoencoders to compress high-dimensional malware data into more optimized representations that …

Pfedrag: A Personalized Federated Retrieval-Augmented Generation System With Depth-Adaptive Tiered Embedding Tuning, Hangyu He, Xin Yuan, Kai Wu, Ren Ping Liu, Wei Ni

Research outputs 2022 to 2026

Large Language Models (LLMs) can undergo hallucinations in specialized domains, and standard Retrieval-Augmented Generation (RAG) often falters due to general-purpose embeddings ill-suited for domain-specific terminology. Though domain-specific fine-tuning enhances retrieval, centralizing data introduces privacy risks. The use of federated learning (FL) can alleviate this to some extent, but faces challenges of data heterogeneity, poor personalization, and expensive training data generation. We propose pFedRAG, a novel Personalized Federated RAG framework, which enables efficient collaborative fine-tuning of embedding models to address these challenges. The key contribution is a new Depth-Adaptive Tiered Embedding (DATE) architecture, which comprises a Global Shared Layer, combined using …

Automated Infinite Combos Detection System For Collectible Card Game, Amornpong Trakarnkulphun

Chulalongkorn University Theses and Dissertations (Chula ETD)

In Collectible Card Games, resources management is one of the most effective strategy. However, there are some combinations of cards which produce unlimited resources called infinite combos. Too-cheap infinite combos that can be played early in the game break game environment, because the player who executes the combo immediately wins. To solve the problem, game designers usually restrict a part of the combos from being played. Due to the large number of cards in the card pool, it is difficult to search for these combos and solve this problem before a release date of a card set. This research aims …

Pipeline For Non-Self-Intersecting Racetrack Generation In Racing Games, Chatrtime Chatramornrat

Chulalongkorn University Theses and Dissertations (Chula ETD)

Procedural content generation (PCG) allows racing games to supply new racetracks automatically, but straightforward generators can produce layouts with unrealistically tight bends or self-crossings that compromise playability. This thesis introduces a racetrack generation pipeline that connects chain–code–based synthesis with a dedicated post-processing stage for resolving intersections. Candidate layouts are first assembled from image–based differential chain codes and adjusted so that each forms a closed loop. Regions where the centreline crosses itself are then identified and repaired by cutting and joining overlapping branches and applying locally constrained smoothing, so that the overall course geometry is kept intact. Finally, geometric descriptors such …

Gitcofl : Design And Implementation Of A Git-Based Federated Learning Framework Utilizing Container-Based Technology, Chokchai Faroongsarng

Chulalongkorn University Theses and Dissertations (Chula ETD)

The rapid proliferation of artificial intelligence (AI) necessitates training paradigms that are both robust and privacy-preserving. Traditional centralized federated learning (CFL) suffers from single points of failure and communication bottlenecks, while decentralized federated learning (DFL) introduces communication complexities and lacks integrated version control. This paper proposes GitCoFL, a novel framework that leverages Docker containers and a Git server as an independent, external medium for model weight communication and storage. This architecture provides inherent flexibility, enabling efficient operation in both CFL and DFL settings. Using CIFAR-10 benchmarks with MobileNet and using FedAvg for federation algorithms, GitCoFL achieves competitive F1 score when …

A Benchmarking Study Of Grover's Algorithm For Solving Boolean Sat With Quantum Circuits, Jirapas Unison Jipipob

Chulalongkorn University Theses and Dissertations (Chula ETD)

This study explores how well Grover's Algorithm performs in solving the Boolean Satisfiability Problem (SAT) using quantum circuits. The algorithm is implemented with IBM's Qiskit framework and compared to classical brute-force methods. Experiments focus on 3-SAT, 4-SAT, and 5-SAT problems, using quantum simulators and IBM quantum hardware. The results show that Grover's Algorithm is more efficient, offering a theoretical quadratic speedup over classical methods. However, practical issues like limited qubit availability, hardware noise, and optimization challenges impact its current performance. The data highlights the potential for quantum computing to scale and solve NP-complete problems. This research shows how quantum computing …

Out-Of-Domain Tuberculosis Detection From Chest X-Ray Images Using Meta Learning With Soft Labels, Kanokphat Jinanarong

Chulalongkorn University Theses and Dissertations (Chula ETD)

Automated tuberculosis screening from chest x-ray images could greatly benefit regions with limited medical resources. However, currently available methods could not handle this problem well due to the difference in chest x-ray images in each region caused by the difference in machine and demographics, the possible noise in the available labels, and other reasons. We therefore propose a new machine learning algorithm that could detect tuberculosis with great accuracy on any other data sources by learning from completely different data sources. We used a model-agnostic meta-learning together with a soft label approach, utilizing the meta-gradient to dynamically adjust the soft …

Enhancing Large Language Models For Legal Question Answering : A Case Study On The Land And Building Tax Act In Thailand, Nattapat Tantapong

Chulalongkorn University Theses and Dissertations (Chula ETD)

Thailand's Land and Buildings Tax Act requires interpreting multiple legal instruments. However, general-purpose large language models often produce answers that lack both accuracy and verifiable legal citations, which are essential for legal reasoning. This study investigates whether curriculum-structured fine-tuning enables small Thai-aligned models (8B parameters) to perform comparably to large commercial models. The research constructs a domain-specific corpus of 8,410 Q&A pairs with 30,612 hard-negative triplets, then evaluates four curriculum designs and four adapter ranks using retrieval completeness and answer quality metrics. The experimental results reveal two key findings. First, hard-negative retrieval training improves Multi-HitRate@5 by 2.6% and Multi-MRR@5 by …

Diffusion Model-Based Synthetic Image Augmentation For Enhancing Vehicle Detection Model Performance, Pawaris Parnphotong

Chulalongkorn University Theses and Dissertations (Chula ETD)

A critical challenge for modern object detection models is the significant performance degradation encountered in adverse weather or poor lighting conditions. This issue primarily stems from the imbalance of training datasets, which are predominantly composed of normal daytime scenes, resulting in reduced robustness when models face challenging scenarios. Although traditional data augmentation techniques exist, they are limited to pixel-level adjustments and inherently restricted in generating data diversity. Consequently, this research proposes an approach utilizing diffusion models to generate realistic synthetic images that simulate new atmospheric and environmental conditions. To maximize the efficiency of synthetic data generation, we investigated a crucial …

Large Language Model-Based Framework For Mental Health Support Conversation In Thai, Radchaneeporn Changpun

Chulalongkorn University Theses and Dissertations (Chula ETD)

As mental health support needs continue to rise in Thailand, access to professional care remains constrained. LLM-based chatbots present a scalable alternative, though most current systems are limited to single-turn, solution-focused interactions. This thesis introduces a Large Language Model-Based Framework for Mental Health Support Conversations in Thai, organizing dialogues around five fundamental mental health counseling stages: rapport building, problem identification, goal setting, working, and termination. Drawing from Person-Centered Therapy (PCT) and Acceptance and Commitment Therapy (ACT). The framework was assessed against three single-agent baseline models using LLM-simulated users, with positive user reaction rates evaluated by LLMs. Results demonstrated a 79.01% …

Music Generation Using Large Language Models : A Case Study On Traditional Thai Music, Sutusta Kunasorn

Chulalongkorn University Theses and Dissertations (Chula ETD)

This study presents a framework for music generation using Large Language Models (LLMs), focusing on the creation of symbolic compositions in the style of traditional Thai music. The framework integrates prompting techniques, retrieval-augmented generation (RAG), and multi-agent collaboration to guide the model in producing melodies that reflect Thai musical characteristics. ABC notation is used as the primary representation, allowing LLMs to process musical structure in a text-based format. The system incorporates curated Thai musical examples and theoretical knowledge to support culturally informed generation, while multiple agents work together to compose, evaluate, and refine musical ideas using rubric-based criteria adapted from …

A Hybrid Hardware Modeling, Simulation, And Generation Framework : A Case Study Of Risc-V Out-Of-Order Superscalar Cpu Design, Tanawin Devaveja

Chulalongkorn University Theses and Dissertations (Chula ETD)

We present Kathryn, a novel framework that simplifies hardware design through a hybrid design flow and an integrated hybrid simulation approach. As specialized hardware demand grows, managing design complexity, precise control flow, and parallelism control remains challenging. Kathryn alleviates these issues by abstracting control logic and still maintaining designers’ cycle-accurate control ability. Moreover, Kathryn provides hardware-aggregation features, helping designers manage complex hardware component structures. As a result, in a simple RISC-V design, the model written in Kathryn shows a 2.6x reduction in lines of code compared to those of Verilog. [1] Moreover, in an Out-of-Order design, the model on Kathryn …