DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Machine learning model for stress prediction
No full textModern advancements in wearable sensor technology have enabled the ease of non-
intrusive monitoring of physiological biomarkers like Electrodermal Activity (EDA)
and Heart Rate (HR) that could be directly linked to elevated stress levels. While
past research has been done in this domain, specifically in using machine learning
to detect stress levels, the endeavour of developing a machine learning model capa-
ble of generalising well on new, unseen data still remains. Stress responses can be
reflected by biological and psychological components that vary across unique individ-
uals, exacerbating the complications of developing a generic stress detection model.
One of the most significant challenges is the lack of large, publicly available datasets
labeled for stress prediction that can be used to develop robust machine learning mod-
els. In this project, we present a stacking ensemble model combining XGBoost and
Artificial Neural Networks (ANN) to predict stress levels using wearable device data,
and then evaluate their generalisation ability for predictions on new, unseen data. By
demonstrating the effectiveness of stacking ensembles in improving stress prediction
and highlighting the need for addressing class imbalance to enhance model sensitivity,
this research contributes to the development of robust stress monitoring systems for
health applications and provides insights for the further study of stress detection using
physiological bio-signals recorded using wearable technologies.Bachelor's degre
Convergence analysis of probabilistic schemes for nonlinear partial differential equations
No full textThis project studies the Deep Second-order Backward Stochastic Equations (2BSDE) solver for fully nonlinear partial differential equations (PDE). Several improvements to the Deep BSDE solver proposed in the recent years are discussed. Results produced by [8] are served as the benchmark for this project.
Two new ideas are proposed to improve the Deep 2BSDE solver - via modifying the objective function and incorporating an optimistic initialization of the neural networks. Numerical experiments are performed with PDEs in 1- and 5-dimensional settings to demonstrate the performance improvement of the proposed ideas. Finally, future work that can be done to extend the scope of this project is discussed.Bachelor's degre
High light yield perovskite scintillators
No full textPerovskite scintillators have garnered significant research interest due to their exceptional
photoluminescence properties, tunable bandgaps, and high light yields. Recent advancements
have demonstrated that perovskite nanocrystals can achieve photoluminescence quantum
yields (PLQY) of up to 200% through the quantum-cutting process, making them promising
candidates for next-generation scintillation applications.
This thesis explores the synthesis, characterization, and scintillation performance of two
distinct perovskite materials: Yb-doped CsPbCl_3 and CsPbBr_3 nanocrystals. The study
focuses on optimizing their synthesis methodologies, evaluating their optical properties, and
investigating their response to ultraviolet (UV) and X-ray excitation. The emission of
visible-range photons for CsPbBr_3 and NIR-range photons for Y b − doped CsPbCl_3 perovskite
materials makes them highly suitable for applications such as X-ray imaging, medical
diagnostics, and security screening.
Through a comprehensive analysis of their fluorescence and scintillation behaviour, this work
aims to assess their potential as efficient X-ray scintillators. The findings contribute to the
ongoing development of perovskite-based scintillators and offer insights into their viability for
practical radiation detection applications.Bachelor's degre
试论明星Billkin对泰华想象的影响 = The rise of Billkin: the shifting imaginations of Thai-Chinese
No full text同化与双文化是泰国华人社区历史中的关键词。从民族主义时代的浪潮，到泰 国(之前称为暹罗)建国前后，再到美国时代(自冷战以来)，泰国华裔在媒体、话 语和文学始终遵循 William Skinner 提出的易于分类的同化模型论。然而，在近年来的 流行文化再现中，如电视剧和电影，泰国华裔想象的再现发生了变化。本文通过流行 歌手兼演员 "Billkin" Putthipong Assaratanakul 的两部作品进行探讨。通过分析他的两部 影视作品《I Told Sunset About You》(2020)与《How to Make Millions before Grandma Dies》(2024)，本文尝试挑战当前对泰华想象的理解与认知。文章将分析 Billkin 的 作品如何提出了新的方式来想象“泰国华裔”。本文从他的第一部作品《I Told Sunset About You》论述以旅游业为功能，并通过与中国的联系实现资本化的想象。而在 《How to Make Millions before Grandma Dies》中，本文试论泰华形象的转变。泰华想 象不再仅限于本土，而是更加以东南亚为中心，整合了东南亚地区华人社群的经验， 进一步统一了东南亚华人社群。通过理解 Billkin 作品中泰国华裔想象的变化，本文旨 在探讨泰国在未来全球化华人社群中的位置。
Assimilation and biculturalism are keywords in the history of Thai-Chinese communities. From the waves of the Nationalism Era, before and after the formation of Thailand (previously known as Siam) to the American Era (since the Cold War), media, discourse, literature of the Thai-Chinese remained under the easily categorised Assimilation Model proposed by William Skinner. However, in recent representations in popular culture like television dramas and films, there has been a shift in the imagination of the Thai-Chinese. This paper dives into the two works of pop star and actor "Billkin" Putthipong Assaratanakul. Through two filmic texts, I Told Sunset About You (2020) and How to Make Millions before Grandma Dies (2024), this paper attempts to challenge current understandings and perceptions of the Thai-Chinese imagination. It will analyse how the works of Billkin suggest new ways to imagine “Thai-Chinese”. The paper argues that the imagination seen in his first major project, I Told Sunset About You, serves tourism functions and building connections with China for capitalistic gain. However, in How to Make Millions before Grandma Dies, this paper observes a shifting imagination of the Thai-Chinese. The Thai-Chinese imagination shifts to be more Southeast Asian-centric, consolidating experiences across ethnic Chinese communities across the region, by extension, unifying Southeast-Asian Chinese communities. By understanding the shifting imaginations of the Thai-Chinese in Billkin’s works, this paper participates in locating Thailand’s role in a future that involves a globalised Chinese community.Bachelor's degre
Evaluation of backdoor attacks on deep neural networks
No full textDeep Neural Networks (DNNs) are increasingly deployed for use in critical applications in organizations everywhere, but yet the time and cost complexity of a state-of-the-art model can mean outsourcing parts of the DNN model training, either through third-party datasets or models, opening themselves up to backdoor attacks such as data poisoning.
This study evaluates the effectiveness of BadNet attacks across multiple DNN models (e.g., WideResNet50, MobileNetV3-Large) and datasets (CIFAR10, TinyImageNet) under real-world constraints: black-box access, non-targeted attacks, and dataset-only manipulation. We analyse key attack parameters—poisoning ratio and trigger size—and their impact on a backdoored model’s clean accuracy (ACC) and attack success rate (ASR). Our findings reveal that complex datasets (e.g., TinyImageNet) are more vulnerable, requiring lower poisoning ratios (0.5–2%) for high ASR, while simpler datasets (e.g., CIFAR10) demand higher ratios. Smaller and less complex models (e.g., MobileNetV3-Large) are more susceptible, achieving 100% ASR with minimal poisoning.
We further assess three defences—Anti-Backdoor Learning (ABL), Channel Lipschitz Pruning (CLP), and Neural Attention Distillation (NAD)—and demonstrate that NAD is most effective for complex datasets, and CLP performs better on simpler models. However, no single defence universally mitigates all attack configurations. Our results highlight the need for risk-avoidant model selection, dataset verification, and selecting appropriate defences to protect against backdoor threats.Bachelor's degre
Functional lateralisation in arithmetic: investigating the role of handedness and problem size using functional near-infrared spectroscopy (fNIRS)
No full textArithmetic processing has been found to be left-lateralised in the parietal regions, specifically in the intraparietal sulcus (IPS). An explanation for this hemispheric asymmetry is handedness, yet most studies only include right-handers while left-handers remain understudied. Existing literature demonstrated mixed evidence on whether handedness alone influences IPS lateralisation during arithmetic processing. As problem size was found to modulate fronto-parietal lateralisation, our functional near-infrared spectroscopy study (fNIRS) examined whether handedness influences the fronto-parietal lateralisation differently in small versus large exact arithmetic calculation. A total of 42 participants (13 left-handers and 29 right-handers) were included in the fNIRS analysis. Regions of interest such as the IPS, angular gyrus, supramarginal gyrus, middle frontal gyrus, and inferior frontal gyrus were analysed. Using a Bayesian approach, we found moderate evidence that handedness does not influence fronto-parietal lateralisation during arithmetic processing and problem size does not modulate this result. Since handedness is unable to fully explain lateralisation differences during arithmetic processing, we postulate that strategy use for problem solving and finger counting might play a role. Moreover, problem size may not be the best proxy for strategy use and the effects of problem size and strategy use on lateralisation should be distinguished. As adults typically engage in finger counting during large, but not small problems, future studies can explore how finger counting habits vary with problem size and its influence on lateralisation.Bachelor's degre
Development of keratin composite film via the dip-coating technique for reduced synthetic cross-linker dependency
No full textConventional plastics have been integral to our daily lives since the 1900s. The reliance
on this non-biodegradable material stems from its affordability, durability, and vast
availability in the market. However, major environmental concerns pertaining to these
persistent synthetic materials are highlighted as critical issues that lead to land and ocean
pollution. This has led to a need for convenient and biodegradable-focused material to
tackle these rising issues. Keratin has been a favourable bio-based material that possesses
outstanding material properties that can be fabricated via a series of chemical reduction
reactions. Even though keratin protein is extensively studied and well-received as a
packaging material substitute, there lies the challenge of the poor biodegradability of
synthetic cross-linkers in the fabrication of upcycled keratin films. A carefully studied
human hair-derived keratin (HHDK) dip-coated composite film was fabricated to analyse
its overall hydrophobicity, hydrolytic stability, and dissolution as a coating material. This
composite film incorporates the feasibility of using ɑ-cellulose and keratin protein to
functionalise as a surface coating. The theory of dip-coating is implemented to potentially
improve the mechanical properties of the film, whilst eliminating the use of synthetic
cross-linkers. A revamped Shindai method is adopted in this study to extract KIF from
human hair and the latter will be used for mould casting and dip coating. This composite
film incorporates the feasibility of using ɑ-cellulose and keratin protein to functionalise
as a surface coating. The keratin-cellulose composite film, comprising a varying number
of dip-coating cycles, and showcased outstanding performance in its tensile strength of >
59.7 MPa ± 5.6 MPa at a relatively low concentration of 20 mg/mL keratin solution. The
7x dip-coated sample featured the most favourable moisture stability in containing the
Direct Blue 71 staining dye for more than 10 days. With a higher concentration of keratin
solution dip-coat, there remains a potential for optimising the number of dip-coating
cycles to allow proper deposition of keratin solute for better hydrolytic stability of the
film, thereby maintaining integrity under content-induced stress.Bachelor's degre
Establishing an in-house lentiviral vector platform and associated analytics
No full textLentiviruses are used in gene therapy to deliver corrective genes to patients with
genetic diseases. Evonik is interested in creating cell culture additives which improve
key performance parameters of lentivirus production. Product development for this
market would require a lentiviral production workflow, thus, this project aims to establish
and validate a lentiviral production platform for product testing. We determined
HEK293T to be the most suitable cell line for both production and testing of viruses due
to a high production titre and transduction efficiency. Successful lentiviral production and
transduction were validated with fluorescent microscopy, and functional viral titre and
multiplicity of infection was determined with flow cytometry. Process optimisation
experiments demonstrated 48 hours post-transfection to be optimal for viral harvesting,
fetal bovine serum free media at transduction increased viral production titre by 28%,
however plasmid ratio experiments were inconclusive. Lastly, to validate this platform for
product development, we attempted to discover a novel inactivation agent. Evonik
Surfactant at 0.05% and polysorbate 80 at 0.1% and 0.5% showed a decrease in
functional viral titre of 87%, 9% and 50% respectively. We demonstrated a functional
lentivirus production and testing platform with optimised parameters and validated its
utility in a viral inactivation assay.Bachelor's degre
Structure-tailored superlattice Bi7Ti4NbO21: coupling octahedral tilting and rotation induced high ferroelectric polarization for efficient piezo-photocatalytic CO2 reduction
No full textIntergrowth ferroelectric semiconductors with excellent spontaneous polarization field are highly promising piezo-photocatalytic candidate materials. In addition, developing structural design and revealing polarization enhancement in-depth mechanism are top priorities. Herein, we introduce the intergrowth ferroelectrics Bi7Ti4NbO21 thin-layer nanosheets for piezo-photocatalytic CO2 reduction. Density functional theory (DFT) calculations indicate that interlayer lattice mismatch leads to increased tilting and rotation angle of Ti/NbO6 octahedra on perovskite-like layers, serving as the main reason for increased polarization. Furthermore, the tilting and rotation angle of the interlayer octahedron further increase under stress, suggesting a stronger driving force generated to facilitate charge carrier separation efficiency. Meanwhile, Bi7Ti4NbO21 nanosheets provide abundant active sites to effectively adsorb CO2 and acquire sensitive stress response, thereby presenting synergistically advanced piezo-photocatalytic CO2 reduction activity with a high CO generation rate of 426.97 μmol g−1 h−1. Our work offers new perspectives and directions for initiating and investigating the mechanisms of high-performance intergrowth piezo-photocatalysts.Published versionThe authors gratefully acknowledge financial support from the Natural Science Foundation of Jiangsu Province (BK20220596). Innovative science and technology platform project of cooperation between Yangzhou City and Yangzhou University, China (No. YZ202026305). Natural Science Foundation of China (21922202, 21673202 and 22272147), the Priority Academic Program Development of Jiangsu Higher Education Institutions
Object detection for car cabin monitoring
No full textObject detection in vehicle cabin environments is becoming increasingly essential to improve vehicle safety, enhance user experience, and facilitate the integration of autonomous driving systems. This thesis addresses the challenges associated with detecting objects in the confined and dynamic space of vehicle interiors. By leveraging deep learning techniques, particularly state-of-the-art algorithms from the YOLO series, this work aims to enhance the accuracy and speed of object detection within the vehicle cabin, thereby supporting the development of more advanced in-cabin monitoring systems and improving overall driving safety.Master's degre