DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Synthesis and printing of MXene and MXene composites for energy storage applications
No full textAs the world moves towards electrification, the demand for durable and highperformance energy storage devices have been increasing over the years. In 2011, twodimensional (2D) transitional metal carbides, nitrides, and carbonitrides, also known
as MXenes, were discovered. These materials were known for their attractive
properties such as high electrical conductivity, good mechanical properties, chemical
properties, and magnetic properties. There have been many compositions synthesised
since its discovery. These materials can be used for various applications like
electromagnetic shielding, sensors, biomedical, and energy storage applications.
This project aims to offer an alternative to the existing energy storage systems through
the study of MXene, where MXenes and its composite inks are synthesised and
characterised. Multiple characterisation techniques will be employed to determine the
morphology and electrical conductivity of the printed MXene. Additionally, to
evaluate the energy storage performance of MXene, supercapacitors will be fabricated
and tested for their electrochemical performances such as rate capability, long-term
cycling stability, and specific capacitance.Bachelor's degre
Full mesh VPN architecture for modern computing paradigms
No full textThis project presents a novel approach to site-to-site connectivity using a full mesh VPN architecture over the public internet. Traditional hub-and-spoke architectures with dedicated leased lines face scalability and resilience challenges in today’s distributed computing environment.
To address these challenges, we propose a full mesh VPN architecture over the public internet, utilising lightweight WireGuard tunnels which allows for direct connections between all sites without significant performance overhead. Our solution addresses the inherent unpredictability of internet routing through three key innovations: (1) a dual-stack transport strategy that dynamically switches between IPv4 and IPv6 tunnels based on real-time performance metrics, (2) comprehensive network measurement services that continuously monitor latency, packet loss, and jitter across all potential paths, and (3) advanced traffic engineering algorithms that select optimal routes based on graph modelling and empirical data.
This reference implementation offers organisations a practical and scalable approach to site-to-site connectivity that adapts to changing traffic patterns and the distributed nature of modern computing.Bachelor's degre
Advisor-advisee relationship and the organizational culture of doctoral programs on doctoral students’ mental health and academic performance: a scoping review protocol
No full textDoctoral students’ mental health is increasingly recognized as a critical issue in academia, with advisor-advisee relationships playing a key role in both well-being and academic performance.
The organizational culture of doctoral programs may also influence these outcomes, but existing literature has yet to fully addressed the interplay between these factors. This scoping review aims to identify elements within the advisor-advisee relationship and supervision process that are associated with doctoral students’ mental health and academic performance. It also seeks to examine how the organizational culture of doctoral programs relates to these dynamics. The review will include both empirical studies and literature reviews focusing on these relationships. The
following databases will be searched: Medline (Ovid), Embase (Ovid), Cochrane Library, Web of Science, ERIC, PubMed, CINAHL (EBSCOhost), and PsycInfo (APA). Studies will be screened by
two independent researchers, with duplicates removed. There will be no restrictions on publication date or language. Data extraction will be conducted using a standardized spreadsheet, and
findings will be synthesized using thematic analysis, with results presented in both narrative form and summary tables.Published versionThis work was carried out with the support of the Coordination for the Improvement of Higher Education Personnel — Brazil (CAPES) — Financing Code 88887.910184/2023–00. This work was produced with the support of INCD/CNCA and it was funded by FCT I.P. under the project Advanced Computing Project 2024.10172.CPCA.A1 and DOI https://doi.org/10.54499/2024.10172.CPCA.A1, platform Cirrus
Stated SLn-skein Theory
No full textThe stated SLn-skein module Sn(M, N , v) of a (marked) 3-manifold (M, N ) is
a quantization of the regular function ring of the SLn-representation variety of
π1(M, N ). It is a module over a commutative domain R with an invertible parameter v. This framework extends the Kauffman bracket skein theory (the SL2
case) to a more general SLn setting, and it is closely related to the quantum higher
Teichm¨uller space, quantum cluster algebras, and quantum moduli algebras.
In chapter 2, we show that the classical limit of the stated skein module of a marked
3-manifold M corresponds to the ring of functions on the SLn-representation variety
of π1(M). Chapter 3 introduces a C-linear map, called the Frobenius map, from
Sn(M, N , 1) to Sn(M, N , v), where R = C and v is a root of unity. This map is
pivotal in the representation theory of the stated SLn-skein algebra, as its image
lies within the center of the stated SLn-skein algebra. In chapter 4, we explore the
implications of the stated SLn-TQFT theory.
Chapters 5 and 6 concentrate on the SL2 case of stated skein theory. We prove
that for compact marked 3-manifolds, the stated skein module at roots of unity is
finitely generated over the image of the Frobenius map. Additionally, we define the
representation-reduced stated skein module by scaling the action of the Frobenius
image on the stated skein module. For closed 3-manifolds over C, we establish that
the dimension of the representation-reduced stated skein module is 1.
In chapter 7, we examine the Bonahon-Wong-Yang volume conjecture. This conjecture is formulated based on intertwiners between two isomorphic irreducible
representations of the skein algebra. We compute these intertwiners for the closed
torus and the once-punctured torus, showing that the trace of these intertwiners
approaches zero in the limit for Seifert manifolds.Doctor of Philosoph
An initial assessment of volcanic meteo-tsunami hazard in the South China Sea: what we learned and how to move forward
No full textVolcanic meteo-tsunamis (VMTs), though rare, can pose significant threats to people, as exemplified by the 2022 Hunga Tonga–Hunga Ha’apai (HT-HH) eruption in the SW Pacific. While various studies have explored such phenomena, none have investigated analogous scenarios in regions with potential occurrence of large undersea eruptions. We focus on areas along the South China Sea (SCS), which is a region among the most densely populated on Earth and historically prone to volcanic activity. We simulated VMTs from one intra-basin volcano (KW-23612) and three extra-basin volcanoes (Banua Wuhu, Kikai, and Fukutoku-Oka-no-Ba), to assess which countries around the SCS could be more exposed to such phenomena. Our results generally indicate that the SCS can be considered a low-hazard region from VMTs, and that the worst-case scenarios are produced by eruptions/tsunamis from within the SCS basin itself, with offshore waves up to 10 and 20 cm offshore Hong Kong and Manila respectively. Countries bordering the shallower Sunda Shelf (Malaysia, Thailand, Cambodia, and southern Vietnam), instead, receive much smaller waves (<2 cm). Despite the limitations, this study sets the basis to quantitatively assess hazard from volcanic meteo-tsunamis at key locations in the SCS.Ministry of Education (MOE)National Research Foundation (NRF)National Environmental Agency (NEA)Published versionThis project was supported by the National Research Foundation, Singapore, and National Environment Agency, Singapore, under the National Sea Level Programme Funding Initiative (Award No. USS-IF-2020-2). AV and ADS were also supported by the Singapore Ministry of Education Academic Research Fund (Award No. MOE2019-T3-1-004)
When victims freeze: investigating the influence of victim freezing and victim-perpetrator relationship on victim blame in sexual assault
No full textThis study investigated how the trauma response of freezing during rape, compared to active
resistance, influences blame attribution toward victims in Singapore. It also examined the role
of the victim-perpetrator relationship (stranger vs. date-acquaintance) and whether rape myth
acceptance (RMA) and conservative sexual attitudes (CSA) moderate these effects. Using a 2
x 2 factorial experimental design, 101 adult Singaporean participants (M age = 22.74) were
randomly assigned to read one of four rape vignettes varying in victim behaviour and
relational context. Participants then completed measures of victim blame, RMA, and CSA.
Contrary to expectations, the Mann-Whitney U test indicated no significant difference in
victim blame between those who froze and those who resisted, = 1,097, = .11, =
−.14, suggesting potential generational shifts in trauma awareness. However, victims
assaulted by a date-acquaintance were attributed significantly more blame than those
assaulted by a stranger, consistent with “real rape” stereotypes, = 302, < .001, =
−.76. No significant moderation effects were found for RMA or CSA, likely due to low
variability in these constructs within the sample. Exploratory analyses revealed that male
participants endorsed significantly higher RMA scores than female participants, =
845, = .01, = .31, underscoring the importance of gender-sensitive intervention
strategies. Overall, the findings suggest that while trauma-informed education may reduce
blame for passive victim responses, entrenched beliefs about acquaintance rape persist. This
study contributes to the literature on victim-blaming in non-Western contexts and highlights
the need for more ecologically valid designs and diverse samples to further understand how
contextual and ideological factors influence perceptions of sexual violence.Bachelor's degre
Optimization of differential low noise amplifier with inductive source degeneration
No full textThis dissertation presents the design, simulation, and analysis of a differential low noise amplifier with inductive source degeneration intended for high-frequency applications. The low noise amplifier is designed using TSMC 0.18 μm technology and simulated in Cadence Spectre RF tool to validate its performance.
The work begins by providing an overview of the historical background and development trends of LNAs, clarifying the research motivation, design challenges, and objectives addressed in this study. A detailed literature review is conducted to consolidate fundamental concepts crucial to LNA design, including noise figure, linearity, scattering parameters, and stability metrics. Additionally, the use and practical applications of the Smith chart are introduced, followed by an in-depth discussion of inductive source degeneration techniques and their impact on amplifier performance. The design methodology is systematically presented, starting with the selection of a capacitively cross-coupled differential topology aimed at optimizing gain and noise performance. Strategies for input and output impedance matching are elaborated to ensure efficient signal transfer and system integration. Design considerations for MOS transistors, including parameter selection and biasing techniques, are also addressed.
Finally, a source degeneration network using inductors is implemented to enhance linearity and improve input matching. Ultimately, simulation results, including the scattering parameters (S11, S12, S21, S22) and transient response, are analyzed to verify the effectiveness of the proposed design. The differential low noise amplifier realizes a gain of 17.96 dB and NFₘᵢₙ of 0.82 dB. The results confirm that the LNA achieves the intended performance objectives, demonstrating its suitability for practical high-frequency applications.Master's degre
Deblurring of satellite imagery using deep learning-based image deconvolution
No full textSatellite imagery in today’s world has offered valuable data for understanding and
analyzing the Earth at surface level. It continues to be a critical tool for researchers,
policymakers, and various other industry people. In the recent decade, satellite imagery
has expanded in capabilities, with the use of Artificial Intelligence and Machine Learning
technology for remote sensing tasks such as image classification and semantic
segmentation. However, the accuracy of insights drawn from these techniques fall short
when the obtained raw satellite images are subjected to extreme optical blur. This project
leverages on computer vision and machine learning to implement a deblurring model that
effectively aims to tackle extreme blur conditions. Specifically, a Deep Convolutional
Generative Adversarial Network (DCGAN) was developed to recover satellite images with
both linear, and optically nonlinear blurs for the purpose of comparing its ability to
generalize both within and beyond the blur conditions it encountered during training. The
model was trained on paired sharp and blurry satellite image data selected from Google
Earth. Notably, the deblurred images from Gaussian-blurred images achieved a Peak
Signal-to-Noise Ratio (PSNR) of 19.01dB, a Structural Similarity Index Measure (SSIM)
of 0.6069 and a Learned Perceptual Image Patch Similarity (LPIPS) of 0.3254. To further
improve performance, traditional Richardson-Lucy deconvolution was introduced as a
preprocessing step for the DCGAN. The new results had revealed that while the deblurred
images showed reasonable improvement in objective fidelity, they still lacked clarity in
perceptual realism, as judged by human-aligned metrics like LPIPS. These findings
indicate the promising potential of hybrid deep learning approaches for deblurring to
support greater accuracy in downstream state-of-the-art image processing applications.Bachelor's degre
Learning cross-modal visuomotor policies for autonomous drone navigation
No full textDeveloping effective vision-based navigation algorithms adapting to various scenarios is a significant challenge for autonomous drone systems, with vast potential in diverse real-world applications. This paper proposes a novel visuomotor policy learning framework for monocular autonomous navigation, combining cross-modal contrastive learning with deep reinforcement learning (DRL) to train a visuomotor policy. Our approach first leverages contrastive learning to extract consistent, task-focused visual representations from high-dimensional RGB images as depth images, and then directly maps these representations to action commands with DRL. This framework enables RGB images to capture structural and spatial information similar to depth images, which remains largely invariant under changes in lighting and texture, thereby maintaining robustness across various environments. We evaluate our approach through simulated and physical experiments, showing that our visuomotor policy outperforms baseline methods in both effectiveness and resilience to unseen visual disturbances. Our findings suggest that the key to enhancing transferability in monocular RGB-based navigation lies in achieving consistent, well-aligned visual representations across scenarios, which is an aspect often lacking in traditional end-to-end approaches.Economic Development Board (EDB)Submitted/Accepted versionThis work was supported by the Economic Development Board (EDB) through Space Technology Development Programme (STDP) Thematic Grant Call on Space Technologies under Award S23-020019-STDP
Goal-oriented temporal action segmentation and action anticipation
No full textTemporal action segmentation and action anticipation are crucial for enabling intelligent
systems to understand and interact with humans in dynamic, real-world environments.
Despite recent advancements, significant challenges persist, including
the lack of goal-oriented reasoning, limited interpretability, and biases introduced
by imbalanced training data. Existing approaches have primarily relied on deep
learning architectures such as temporal convolutional networks and transformerbased
models, which often treat actions as isolated events rather than components
of structured, goal-oriented activities. Additionally, current methods typically operate
as black-box systems, offering limited insight into their decision-making processes
and struggling with long-range dependencies in complex action sequences.
diverse
video datasets, ranging from controlled instructional recordings such as 50Salads,
Breakfast, and EGTEA Gaze+ to unstructured “in-the-wild” footage including
Epic-Kitchens-100 and Ego4D. Results demonstrate significant improvements
in standard metrics over state-of-the-art models for both temporal action segmentation
and action anticipation tasks. Furthermore, the interpretability of the
proposed frameworks provides actionable insights into the behavior and decisionmaking
processes of the models.
This thesis addresses these challenges through three key contributions. First, a
grammar-based framework for temporal action segmentation that constructs hierarchical
task grammars from object-centric action transitions and refines neural
model predictions through a shortest path formulation, combining neural confidence
scores with grammar-derived task priors to improve action sequence coherence.
Second, a multivariate Markov chain model for dense action anticipation
that provides interpretable predictions through explicit goal modeling. The
approach employs an object-centered goal representation where objects serve as
variables, actions as object states, and goals as inter-object interactions within
a multivariate Markov chain framework. This formulation captures the probabilistic
influence between objects and their interactions, enabling long-term action
selection while maintaining full interpretability through transparent mathematical
structures. Third, a novel sequence augmentation strategy (ActSeq) that generates
diverse training sequences through grammar-guided candidate generation and
random sequence modifications. The approach introduces an Action Grammar Induction
algorithm to extract context-free grammars capturing action dependencies,
a Cross-Tree Earley Parser to generate valid next-action candidates across multiple
goals, and random modification techniques to simulate real-world input noise,
collectively addressing action ordering bias and enhancing model robustness.Doctor of Philosoph