DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Fully automated segmentation of subcortical structures in CT head scans using deep learning
No full textWithin the subcortex of the human brain, there are many subcortical anatomies that hold key roles in many fundamental physiological functions. These anatomies exhibit volumetric and morphological changes during the development of neurodegenerative disorders. Thus, it would be useful to perform subcortical segmentation to study their potential as biomarkers for neurological conditions. However, their small sizes make manual segmentation challenging and time-consuming. While numerous automated subcortical segmentation studies have been conducted in MR modality, there is a lack of for CT modality. Given that CT is more accessible, affordable and faster, we aim to develop tools and frameworks for automated subcortical segmentation in CT modality.
To address the lack of publicly available CT subcortical segmentation datasets, cross-domain label propagation from MR to CT was performed through our proposed automated framework of ensembling established MR segmentation models. The proposed framework was validated against manually-annotated datasets to ensure its robustness. The usefulness of our generated CT subcortical segmentation dataset was also validated using transfer learning. Subsequently, SOTA deep learning models, including Swin UNETR and nnU-Net, were trained to perform CT subcortical segmentation and their performance was evaluated.
Our proposed novel automated subcortical label generation framework, ASLGF, achieved increased robustness compared to the MR models standalone. Using our proposed framework, we generated a CT subcortical segmentation dataset that will be the first publicly available CT subcortical segmentation dataset and will enable researchers to develop models by training on it. Our trained deep learning models also achieved commendable average dice scores of 0.741 to 0.901. The code and dataset is made available on the Biomedical Computing Group’s Github:
https://github.com/SCSE-Biomedical-Computing-Group/FYP_Augustine
Through this paper, we provided key contributions along various points of the automated segmentation process. Through the various toolkits and insights developed from this study, we hope to encourage and democratise more research in CT subcortical segmentation and in the broader scheme of things, to understand neurodegenerative disorders better.Bachelor's degre
Words that win: an exploration of the debating community in Singapore
No full textDebate has long been a cornerstone of intellectual discourse and a way for individuals to foster the art of critical thinking, the ability to communicate persuasively and appreciate multiple points of views.
As a debater myself, this topic was of close personal resonance to me as the art transformed me into a confident and articulate individual from someone who was often shy and afraid to express herself.
In Singapore, the significant evolution of the debate scene is reflective of the nation’s growth as a hub of education, innovation and exploration of new concepts.
From prestigious national level competitions like the Singapore Secondary School Debating Championships to the renowned World Schools’ Debating Championship there are several platforms for young talents to display their passion for the linguistic art.
Debate formats have also diversified, from traditional British Parliamentary and Asian parliamentary styles to World Schools debate formats gaining traction, as Singapore expands to be inclusive of various forms, learning from its international counterparts.
The art promotes teamwork, clarity, accessibility, sensitivity to current affairs alongside curiosity, tenacity and the grit to succeed.
This piece illuminates the vibrant and ever-evolving landscape of debate in Singapore that continues to shape critical thinkers and empathetic leaders.
As the community navigates challenges, the commitment of those dedicated to the art spurs continuous growth in intellectual argumentation.Bachelor's degre
AI-driven image search for mobile devices
No full textNowadays, with mobile phone cameras becoming increasingly common due to the convenience
and improved quality of mobile phone cameras, the taking of photos on mobile phones has
become much more frequent. Mobile phone photo collections thus grow in size and are
becoming more difficult to maintain.
Manual tagging or browsing is tedious and error-prone when handling large volumes of images.
It is for this reason that the requirement for automating image retrieval has become more critical,
and it can be spurred by artificial intelligence (AI). AI techniques, particularly in computer
vision (CV) and natural language processing (NLP), can facilitate the use of automatic
classification, tagging, and searching of images by analyzing the image contents.
Nevertheless, the use of AI in photo galleries at present is minimal, with not all mobile brands
having these features. This project will create an image search system with AI that can be used in
most mobile devices. The system will give enhanced image retrieval using AI for content-based
searching and comparison, which can be used easily and effectively by users of diverse phone
brands.Bachelor's degre
Speak without leaks: a modular pipeline for data-level privacy-preserving utilization of large language models
No full textThe widespread adoption of Large Language Models (LLMs) across domains has raised significant concerns about data privacy, particularly when fine-tuning these models on domain-specific or user-generated content that may contain sensitive information. This project addresses the challenge of preserving privacy during LLM fine-tuning by proposing a modular, data-centric pipeline that applies privacy-preserving transformations to training data or prompt before model utilization. Unlike techniques that require changes to model architecture or training algorithms, the proposed pipeline operates independently of the underlying model, making it suitable for black-box scenarios where model internals are inaccessible.
The pipeline integrates a suite of privacy-preserving methods — including classical anonymization, format-preserving encryption (FPE), and local differential privacy (LDP) — to sanitize sensitive content at different levels. The Implementation covers key phases such as entity identification, data sanitization, preprocessing, and model fine-tuning. Experiments conducted on benchmark text classification tasks demonstrate the trade-offs between privacy protection and model utility, with evaluation metrics highlighting the impact of different sanitization strategies.
This work contributes a practical and extensible framework for privacy-aware LLM deployment, offering insights into how organizations can responsibly fine-tune language models on sensitive data or query a third party black-box model with sensitive prompt without compromising compliance or exposing confidential information.Bachelor's degre
Dynamic tuning of single-photon emission in monolayer WSe2 via localized strain engineering
No full textTwo-dimensional (2D) materials have emerged as promising candidates for next-generation integrated single-photon emitters (SPEs). However, significant variability in the emission energies presents a major challenge in producing identical single photons from different 2D SPEs, which may become crucial for practical quantum applications. Although various approaches to dynamically tuning the emission energies of 2D SPEs have been developed to address the issue, the practical solution to matching multiple individual 2D SPEs is still scarce. In this work, we demonstrate precise emission energy tuning of individual SPEs in a WSe2 monolayer. Our approach utilizes localized strain fields near individual SPEs, which we control by adjusting the volume of a stressor layer through laser annealing. This technique allows continuous emission energy tuning of up to 15 meV while maintaining the qualities of SPEs. Additionally, we showcase the precise spectral alignment of three distinct SPEs in a single WSe2 monolayer to the same wavelength.Agency for Science, Technology and Research (A*STAR)Ministry of Education (MOE)National Research Foundation (NRF)Submitted/Accepted versionThis research is supported by the National Research Foundation, Singapore, and A*STAR under its Quantum Engineering Programme (NRF2022-QEP2-02-P13) and Ministry of Education Singapore (Grant MOE-T2EP50221-0002). S.W.L. acknowledges the support by the Ministry of Education, Singapore, under AcRF Tier 1 (reference RT8/23)
Dynamic spatial spillover effects of financial agglomeration on CO2 emissions: the case of China
No full textResearch on the impact of financial agglomeration on CO2 emissions is a vital avenue for advancing CO2 emissions reduction and aligning with the Sustainable Development Goals set forth by the United Nations. However, few studies have investigated the dynamic spatial spillover effect of financial agglomeration on CO2 emissions, and the concepts of “time inertia” and “spatial spillover effect” about CO2 emissions have received limited attention. To address this gap, we used provincial panel data from 30 provinces in China from 2006 to 2019. We employed the entropy method to measure the level of financial accumulation, constructed a dynamic spatial Durbin model, and analyzed the “time inertia” and “spatial spillover effects” of CO2 dioxide emissions in Chinese provinces. Additionally, we explored the dynamic spatial spillover effects of financial agglomeration on CO2 emissions. We examined the transmission mechanism of the impact of financial agglomeration on CO2 emissions through energy consumption and technology market development. The results show that CO2 emissions exhibit significant “time inertia” and “space spillover” effects and considerable regional differences. Financial agglomeration levels positively impact reducing CO2 emissions in the short and long term. Through two intermediary paths of energy consumption and technology market development, financial agglomeration can indirectly reduce CO2 emissions. Our research effectively provides information and serves as a decision-making reference for policy planning related to CO2 emissions reduction, especially in countries with high CO2 emissions, particularly developing nations.Published versionProvincial Science and Technology Program of Guizhou Province “Research on methods and applications of ecological asset valuation in the context of big data - a case study of Guizhou Province (No.20201Y288)”
Atomically thin high-entropy oxides via naked metal ion self-assembly for proton exchange membrane electrolysis
No full textDesigning efficient Ruthenium-based catalysts as practical anodes is of critical importance in proton exchange membrane water electrolysis. Here, we develop a self-assembly technique to synthesize 1 nm-thick rutile-structured high-entropy oxides (RuIrFeCoCrO2) from naked metal ions assembly and oxidation at air-molten salt interface. The RuIrFeCoCrO2 requires an overpotential of 185 mV at 10 m A cm-2 and maintains the high activity for over 1000 h in an acidic electrolyte via the adsorption evolution mechanism. We discuss the role of each element in the RuIrFeCoCrO2 and find that the Cr, Co, and Ir sites contribute to the catalytic activity, while the Cr atoms weaken the Ru-O bond covalency and improves the catalyst stability. The assembled proton exchange membrane electrolyzer operates stably for more than 600 h at a large current of 1 A cm-2. The naked ion assembly demonstrated in this work may provide an effective pathway for the controlled synthesis of a diversity of high-entropy materials.Ministry of Education (MOE)Published versionWe acknowledge the financial support from the Singapore Ministry of Education by Tier 1 (RG80/22, RG125/21, H.J.F), the Singapore Ministry of Education by Tier 2 (MOE-T2EP50121-0006, H.J.F.), the University of Electronic Science and Technology of China for startup funding (A1098531023601467, B.Z.)
WeatherSG - a real-time Singapore weather visualization
No full textThis project focuses on developing a proof-of-concept web application that leverages geospatial visualisation techniques to present real-time weather patterns across Singapore. By utilising real-time data from Singapore's government data services, the system visualises key weather attributes such as rainfall, air temperature, wind speed, direction, and humidity. The application aims to provide an intuitive and interactive experience, allowing users to observe and analyse weather trends dynamically. Additionally, animations are incorporated to enhance the visualisation of weather pattern changes over selected regions. Through this approach, the project seeks to offer an effective and user-friendly platform for real-time weather monitoring and analysis.Bachelor's degre
LLM hallucination study
No full textLarge Language Models (LLMs) exhibit impressive generative capabilities but often produce
hallucinations—outputs that are factually incorrect, misleading, or entirely fabricated. These
hallucinations pose significant challenges in high-stakes applications such as medical diagnosis,
legal reasoning, financial analysis, and scientific research, where factual accuracy is critical. As
LLMs become increasingly integrated into real-world systems, mitigating hallucinations is
essential to ensuring reliable, trustworthy, and ethically sound AI deployments. Without effective
strategies to reduce hallucinations, AI-generated content risks contributing to misinformation,
undermining user trust, and limiting the adoption of LLMs in professional domains.
My report investigates techniques to reduce hallucinations through systematic experimentation
on Meta’s LLaMa model, a state-of-the-art open-source LLM. Specifically, I explored the impact
of key generative parameters, including temperature scaling, top-k sampling, and retrieval-
augmented generation, on factuality and coherence. These parameters play a crucial role in
balancing response creativity and accuracy, directly influencing the probability of hallucinated
content. By carefully tuning these hyperparameters and integrating external knowledge retrieval,
I aimed to assess how different configurations affect the reliability of LLaMa’s generated
responses.
I systematically evaluated the effectiveness of these mitigation techniques using factuality
scoring and response coherence analysis. Factuality was assessed by measuring the alignment of
generated responses with authoritative sources, while coherence analysis examined the logical
consistency and contextual appropriateness of outputs. Results from these experiments provide
quantitative insights into the trade-offs between factual reliability, creativity, and response
variability, offering practical guidelines for optimizing LLMs across different use cases.Bachelor's degre
Security on GenAI, LLM, & Chatbot
No full textThe rapid adoption of chatbots across industries since 2021 has introduced significant security
challenges. This emphasizes the requiring of a comprehensive examination of vulnerabilities in
chatbots. This project aims to identify and address the critical security risks in chatbot
development and focusing on five key vulnerabilities such as Denial of Service (DoS), Insecure
Plugin Design, Prompt Injection, Insecure Output Handling, and Model Theft. Using tools such
as HuggingFace, GPT, and Mistral 7B. I conducted systematic testing to evaluate the impact of
these vulnerabilities and proposed practical solutions to mitigate them. The key solutions
include rate limiting for DoS attacks, command whitelisting and role-based access control for
insecure plugin design, input validation for prompt injection, HTML escaping and Content
Security Policies (CSP) for insecure output handling and obfuscation techniques for model theft.
To enhance developer awareness and ease the development process for new comers, a secure
chatbot development handbook was created and evaluated through workshops with
participants. The pre and post workshop surveys demonstrated a significant improvement in
participants’ understanding of chatbot security, with confidence levels rising from an average of
2.4/5 to 3.55/5. This project has successfully achieved its objectives. However, there are
limitations such as the scope of testing and reliance on specific models that highlight areas for
future work.Bachelor's degre