DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Deposition of carbon nanotube on carrier substrate for improved electromagnetic shielding effectiveness
No full textThis study explores the influence of carbon nanotube deposition on various carrier substrates
such as glass fiber, polyethylene terephthalate, carbon cloth, and with a particular focus on
carbon fiber to evaluate the enhancements in electromagnetic interference shielding
effectiveness. Utilizing surface morphology via scanning electron microscopy and vector
network analysis, we assess the extent to which carbon nanotubes contribute to improved
shielding performance. A standardized set of preparation methods and processing steps were
defined to ensure consistency, while challenges encountered such as dispersion issues and
concentration limitations were analyzed and addressed. Special attention is given to carbon
fiber due to its inherently high conductivity, providing a benchmark for evaluating the
additional impact of CNTs. The findings reveal both the potential and limitations of CNT
integration, offering insight into future directions for optimizing CNT-based composite
materials for EMI shielding applications.Bachelor's degre
Anistropic thermomechanical properties via UV cross-linking through polariser
No full textThis project explores how material properties can be influenced through light-based processing techniques. Samples are exposed to varying light conditions using optical components that modulate intensity and direction. Mechanical testing is conducted using precision instruments capable of analysing material behaviour under stress. The results aim to reveal how adjustments in light exposure parameters affect the performance characteristics of the processed materials.Bachelor's degre
Label-efficient visual recognition with transfer learning
No full textVisual recognition, encompassing tasks such as image classification, object detection, and semantic segmentation, has been a fundamental challenge in computer vision research. It also serves as the foundation for numerous applications, including autonomous driving, remote sensing, and robotics. The emergence of deep learning has significantly advanced this field, enabling remarkable achievements through the use of end-to-end trainable deep neural networks (DNNs). However, this remarkable progress comes at the cost of requiring vast amounts of annotated training data, the collection of which is often extremely labor-intensive and time-consuming. This constraint presents a significant barrier to the scalability and adaptability of various deep learning approaches. Transfer learning, which aims to transfer the off-the-shelf learned knowledge from one or multiple related source domains to target domains, is a promising way to mitigate this constraint.
In this thesis, we explore label-efficient visual recognition with various transfer learning techniques, where the objective is to learn effective visual representations for target domains without any annotations by transferring the off-the-shelf knowledge learned in different but related source domains. Specifically, we investigate three representative types of transfer learning problems, including unsupervised domain adaptation, black-box unsupervised domain adaptation and test-time prompt tuning.
As one of the most typical transfer learning problems, unsupervised domain adaptation aims to adapt a visual recognition model that is trained on labeled source-domain data to perform well on a related but unlabeled target-domain data. The primary challenge is to mitigate the distribution differences between the source and target domains, which are also referred to as domain gap or domain discrepancy, such that the model can generalize well on the target domain. In this thesis, we propose three innovative unsupervised domain adaptation techniques designed to efficiently transfer knowledge from labeled source domains to unlabeled target domains across a range of visual recognition tasks. Specifically, we explore diverse approaches to align domains from multiple perspectives (e.g., data or feature space alignment), thereby reducing the domain gap and enabling to perform well on unlabeled target domain.
Black-box unsupervised domain adaptation aims to learn visual recognition model for target domain with the initial predictions of target data obtained by black-box source-domain model. In another word, during the adaptation process, only the model predictions of target data are available while either source data or source models are inaccessible. Compared to conventional unsupervised domain adaptation, black-box unsupervised domain adaptation offers notable advantages in terms of data privacy and flexibility, allowing adaptation to different target models regardless of the architecture or details of the source-trained black-box model. In this thesis, we analyze the challenges in black-box domain adaptation and design a novel approach for effectively exploiting the initial predictions of target data, leading to better adapted model on unlabeled target data.
We further explore label-efficient visual recognition in the recent era of vision-language models. Despite the impressive zero-shot generalization capabilities of vision-language models, they always face unneglectable gaps while applied to target datasets, i.e., a target dataset typically features unique image styles and text formats tailored to its specific tasks. To circumvent this limitation, we explore an effective transfer learning approach for vision-language models, i.e., test-time prompt tuning, where the prompts are optimized as learnable vectors during the test stage without requiring any annotations for target data. Specifically, we examine the limitation of existing test-time prompt tuning methods, and propose a novel approach that could effectively exploit the learned knowledge in vision-language models and learn prompts for target data on-the-fly.
Extensive experiments conducted over various visual recognition benchmarks indicate that our proposed methods achieve superior performance over target datasets, enabling to largely mitigate the dependence of labeled data across various visual recognition tasks.Doctor of Philosoph
Wireless communications using NOMA
No full textIn this dissertation, we first review the concept and application scenarios of non orthogonal multiple access (NOMA) schemes in wireless networks and study the basic principles of downlink and uplink channels. Due to different optimization methods for the two channels, the research was conducted separately. The focus of this dissertation is on the spectral efficiency of NOMA networks relative to orthogonal multiple access (OMA) networks. Matlab calculations are used to simulate the bit error rate performance (ratio of the number of received bits in error to the total number of bits transmitted during a specific test period) of additive white Gaussian noise (AWGN)for NOMA-2psk, NOMA-4psk, and NOMA-8psk (using the digital phase modulation and NOMA techniques) compared to orthogonal frequency-division multiple access (OFDMA) in two user scenarios. This dissertation also discusses power distribution optimization in multiple user scenarios, as well as simulation of NOMA bit error rate (BER) in Rayleigh fading channels.
This dissertation comprehensively explores NOMA and OFDMA technologies, including their theoretical foundations, fusion mechanisms, practical cases, and their application prospects in future communication systems. By comparing and analyzing the basic principles, key technologies, and fusion theory of NOMA and OFDMA, this dissertation proposes an optimized fusion architecture and resource allocation strategy and evaluates its performance through experiments and simulations. The dissertation also discusses the application examples of NOMA and OFDMA fusion technology in 4G/5G networks and IoT communication and looks forward to the challenges and future development opportunities it faces. Finally, this dissertation summarizes the research results and provides insights and suggestions for the industry, emphasizing the importance of NOMA and OFDMA fusion technology in promoting the development of future wireless communication services and standards.Master's degre
Customized storyboard creation for consistent visual narratives
No full textAutomating storyboard generation using artificial intelligence (AI) has gained a lot of
attention due to advancements in text to image generative models. Even with these
advancements, present-day models do not consistently depict similar visuals across
multiple images, especially in the case of the same character across different narrative
scenarios. The project seeks to improve this shortcoming via the innovations on the
personalization of latent diffusion models (LDMs). We specifically focus on three
categories of personalization methods, including Full Model Fine-Tuning, EmbeddingBased and Hybrid Approaches. We use CLIP-based similarity scores as well as visual
and user studies to assess and contrast the efficacy of the various personalization
techniques like DreamBooth, Textual Inversion, ELITE, TextBoost, and NeTI, Custom
Diffusion, AttnDreamBooth. Our results show various method’s trade-offs in identity
preservation and text-image alignment, providing actionable insights for creative
professionals in pursuit of personalized visuals with consistent and realistic narratives.
Eventually, all the research presented contributes to the automation of personalized
storyboards, fusing together AI-powered image generation and applications in creative
fields.Bachelor's degre
Energy efficient and scalable CGRA hardware computation accelerator
No full textNext-generation computing demands high data throughput and energy-efficient architectures to meet the explosive requirements of emerging applications. Coarse-Grained Reconfigurable Architectures (CGRAs), exemplified by the HyCUBE framework, offer flexibility and efficiency but face escalating power challenges at advanced process nodes. This study addresses these challenges by developing a comprehensive testing and energy optimization framework for HyCUBE’s processing element (PE tile) and its submodules (ALU, routing controller, SRAM).
In this project a hierarchical testing strategy was implemented, combining automated testbenches with coverage-driven verification to streamline validation cycles. Unified Power Format (UPF)-driven multi-voltage domain partitioning was integrated into the ASIC design flow, enabling voltage scaling for critical submodules. Simulations using a 12nm PDK and Cadence toolchain demonstrated that partitioning SRAM, ALU, and routing controller into 0.5 V domains, coupled with 0.8 V–0.6 V level shifters, reduced dynamic power consumption by 40% and improved energy efficiency (TOPS/W) by 2× compared to a unified-voltage baseline. Quadratic power-voltage models (R² > 0.95) were established for submodules, supporting data-driven optimization.
While achieving significant energy savings, the study identified future directions, including post-layout verification, substrate bias effect modeling, and integration of dynamic voltage-frequency scaling (DVFS) with reinforcement learning for adaptive power management. This work provides a systematic methodology for co-optimizing performance and energy efficiency in reconfigurable architectures, particularly for power-constrained edge computing applications.Bachelor's degre
Antenna array design and analysis
No full textAntennas are the basic building blocks of wireless communication systems, and
today’s society relies heavily on these systems to function. For communication
engineers, understanding the functionality of these components is essential. Many
antenna types have been invented, but dipole antennas are widely utilized due to their
versatility and simplicity. They serve many purposes like forming arrays using
multiple elements and modelling advanced structures like patch antennas.
This report delves into the principles of antenna analysis, focusing on analysis of
antenna radiation patterns. The impact of parameter modifications on these patterns
will also be explored. It also explores array factors that enable multiple elements to
replicate the radiation characteristics of larger antenna systems.
MATLAB is traditionally the program that is used to visualize the established formulas
from the literature. This paper attempts to visualize simple radiation pattern plots for
different antenna configurations in both 2D and 3D plots along with directional arrows
to simplify the directionality of antenna radiation. The code supports plotting of simple
dipoles with adjustable parameters, as well as linear arrays.Bachelor's degre
Exploring cell behavior on pollen microgels via immobilized pollen platforms
No full textThis study investigates the potential of pollen as a biomaterial for potential microcarrier
applications, focusing on its ability to support cell adhesion, proliferation, and
morphological changes. By processing sunflower pollen through defatting and microgel
formation and immobilizing it in polydimethylsiloxane (PDMS), the biocompatibility and
structural stability of the material were assessed. The experiments explored how pollen
concentration influences immobilization efficiency, particle spacing, and stability over the
culture period. Fluorescence microscopy imaging revealed that L929 fibroblast cells
readily adhered to the pollen surfaces and formed cellular bridges when the inter-pollen
distance was within a critical range of 50 μm. Over time, increased cell proliferation and
extracellular matrix deposition enabled bridging across larger gaps, with differences
observed between defatted and microgel pollen due to variations in surface chemistry and
topology.
Through stability and cell interaction studies, it was confirmed that pollen-based platforms
can support short-term cell adhesion and migration. The findings contribute to the
development of sustainable, plant-based microcarriers as an alternative to synthetic or
animal-derived materials for large-scale cell culture and bioprocessing. Future research
should focus on optimizing pollen surface properties, assessing its scalability in
bioreactors, and exploring its applications in regenerative medicine and immunotherapy.Bachelor's degre
Intelligent transportation algorithms for calculating qualitative road traffic parameters
No full textThis project focuses on the development of intelligent transportation algorithms to calculate the qualitative road traffic parameters using real-time video analysis. The system was trained and fine-tuned to detect and classify vehicles by leveraging the YOLOv8 model, and also determine the traffic conditions from video footages taken at overhead bridges in Singapore during daytime, night-time and rain. A few important parameters such as traffic density, vehicle count, and estimated speeds were extracted to classify the real-time traffic conditions as either “smooth” or “heavy”. After training the model, the model achieved a precision of 70.3% and a recall of 74.6%, with a mean average precision of 71.9% at 0.50 IoU threshold. These results demonstrate the potential of YOLO-based algorithms in enhancing traffic monitoring and classifications in Singapore.Bachelor's degre
Chime: enhancing family gathering experience
No full textFamily gatherings are meant to foster connection, yet many family gatherings are hindered by
digital devices attraction and annoying conversations. Young people often feel pressured by
traditional questions about academic performance, career progress, or personal life, leading to
stress or avoidance of family events. To address this issue, Chime is designed as an interactive
experience that enhances family bonding through a Blind Box containing traditional snacks and
an online tacit game.
The Blind Box introduces an element of surprise, making the gathering more engaging, while
the game encourages meaningful yet lighthearted interactions. By shifting the focus from
personal interrogations to shared enjoyment, Chime creates a relaxed atmosphere that appeals
to both older and younger generations. The project also integrates traditional snacks to bridge
generational gaps, ensuring that cultural heritage remains an essential part of the experience.
This report explores the problem of disengagement in family gatherings, and how a combination
of food and games can serve as a solution. Through research, design development, and user
testing, Chime aims to demonstrate how thoughtful design can transform family interactions into
more engaging, stress-free, and memorable experiences.Bachelor's degre