DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Data-driven modeling of magnetic materials
No full textMagnetic components account for more than 30% of both the cost and losses in nearly
all power converters. This project aims to enhance magnetic design by utilizing datadriven
methods to predict core losses in magnetic components.The primary focus is on
identifying the limitations of empirical equation-based modeling and proposing a datadriven
approach to improve the accuracy of core loss predictions using advancements
in artificial intelligence, machine learning, pattern recognition, and signal processing.
This data-driven modeling of magnetic materials will enable rapid and accurate prediction
of core losses, considering the complex behaviors and non-linear effects of
magnetic materials, which is essential for the future design of power converters.Master's degre
An N-phosphinoamidinato borasilenide: a vinyl-analogous anion containing a base-stabilised B=Si double bond
No full textBorasilenes, which feature a heterodinuclear Si=B double bond, show interesting reactivities, due to two proximal reactive sites-boron and silicon-each with distinct electronic properties. However, borasilenes remain relatively rare due to the challenge of stabilising them. To achieve a stable borasilene, both the boron and silicon centers must be supported by sterically hindered ligands, which can, however, interfere with their potential applications. In this work, we report the synthesis of an N-phosphinoamidinato potassium borasilenide (compound 3), which is a vinyl-analogous anion containing a base-stabilised B=Si- double bond. This B=Si- double bond is functional and exhibits new patterns of reactivity towards CuCl(PMe3), [IrCl(cod)]2, Me3SiOTf, and MeOTf, leading to the formation of a transition metal π-complex, boron-silicon-containing metallacycle, neutral borasilene and borylsilane, respectively.Agency for Science, Technology and Research (A*STAR)Ministry of Education (MOE)Published versionThis work is financially supported by the Ministry of Education Singapore, AcRF Tier 1 (RG72/21) and A*STAR MTC Individual Research Grants (M21K2c0117). M.-D. Su acknowledges the Ministry of Science and Technology of Taiwan for the financial support
High ability, hidden gaps: building quantitative reasoning amongst Singapore’s mathematically literate
No full textIn Singapore, where primary and secondary students routinely top standardized worldwide mathematics examinations, a paradox emerges: when reaching university, many struggle to apply their skills critically in real-world contexts. This commentary examines the challenges and strategies involved in teaching quantitative reasoning (QR) to mathematically literate students in a top-ranking Singaporean university. While these students arrive well-trained in computation and procedural problem-solving, they often lack confidence and flexibility in ambiguous, data-driven decision-making. This article argues that fostering QR education is crucial not only for Singapore but for education globally, as QR skills underpin evidence-based reasoning within and across disciplines. Such an approach would involve embracing the novelty of QR, cultivating confidence through inquiry-based learning, building skills through authentic problem-solving, and fostering a collaborative environment where communication – perhaps over and above computation – is a core competency.Published versio
Benchmarking lightweight deep learning models on real-time semantic image segmentation
No full textReal-time semantic image segmentation is critical for applications in autonomous systems, robotics, and edge computing, where computational efficiency and accuracy must be balanced. This study systematically evaluates various lightweight deep learning architectures for semantic segmentation using a comprehensive benchmarking framework on a standardized environment with the Nvidia V100 GPU. We assess models across seven key metrics: power consumption (W), floating-point operations per second (FLOPs), model parameters, frames per second (FPS), mean Pixel Accuracy (mPA), mean Intersection over Union (mIoU), and our novel Image Segmentation Model Benchmark (ISMB) Score. The ISMB Score serves as a standardized metric that integrates accuracy and efficiency, providing a robust measure for selecting optimal lightweight segmentation models. Our results offer valuable insights into the trade-offs between computational efficiency and segmentation performance, guiding the development of models suitable for real-time deployment in resource-constrained environments.Bachelor's degre
Detecting ransomware using deep learning and hardware performance counters
No full textRansomware has emerged as a significant cybersecurity threat, inflicting severe consequences
including data loss, operational paralysis, reputational damage, and financial devastation. With
global ransomware damages projected to reach $265 billion annually by 2031, organisations and
individuals alike face increasingly sophisticated attacks that can cripple critical infrastructure and
compromise sensitive information.
Traditional detection approaches using static and dynamic analysis face limitations including
ineffectiveness against zero-day attacks, high system overheads, and evasion techniques
employed by modern ransomware. This project proposes an innovative approach for ransomware
detection utilising Hardware Performance Counters (HPCs) and deep learning techniques.
This project first analyses the limitations of existing ransomware detection methods to establish
the case for hardware-based detection. Multiple deep learning architectures including
Convolutional Neural Networks (CNN), hybrid CNN-RNN, Long Short-Term Memory (LSTM)
networks, and Transformers are then developed and evaluated to process temporal and sequential
HPC data. Furthermore, Neural Architecture Search (NAS) is applied to optimise these
architectures, significantly enhancing detection accuracy while reducing model complexity.
Extensive evaluations reveal that the NAS-optimised models achieve up to 99.14% accuracy,
outperforming state-of-the-art frameworks including HiPeR (98.68%) and DeepWare (98.6%).
The NAS-LSTM model emerges as the most efficient solution, achieving superior performance
with only 16,769 parameters.
This approach enables effective ransomware detection during the critical pre-encryption phase
while maintaining minimal system overhead. The findings demonstrate that microarchitectural
events captured by HPCs when analysed through optimised deep learning models, provide an
effective method for ransomware detection that overcomes many limitations of traditional
approaches.Bachelor's degre
Evaluating the carbon footprint of code implementation
No full textWith the growing popularity of Artificial Intelligence (AI) and its integration into our daily lives, the environmental impact of code implementation is on the rise. Large Language Models (LLMs) in particular, consume massive amounts of resources throughout their training and deployment phases. This project focuses on the fine-tuning process of LLMs, namely these three models— Meta’s LLaMA-2 (7B), Mistral (7B), and Google’s Gemma (2B, 7B) across different computational configurations, presenting a comparative emissions analysis to discover methods of achieving more environmentally friendly LLMs. A global collaborative initiative was created to encourage transparency in emissions data, which is an important gap that needed to be addressed. The results of this study hope to present ways to achieve more energy-efficient methods of developing LLMs, leading to more sustainable AI development for the future.Bachelor's degre
Highly photoreactive semiconducting polymers with cascade intramolecular singlet oxygen and energy transfer for cancer-specific afterglow theranostics
No full textAfterglow luminescence provides ultrasensitive optical detection by minimizing tissue autofluorescence and increasing the signal-to-noise ratio. However, due to the lack of suitable unimolecular afterglow scaffolds, current afterglow agents are nanocomposites containing multiple components with limited afterglow performance and have rarely been applied for cancer theranostics. Herein, we report the synthesis of a series of oxathiine-containing donor–acceptor block semiconducting polymers (PDCDs) and the observation of their high photoreactivity and strong near-infrared (NIR) afterglow luminescence. We reveal that PDCDs absorb NIR light to undergo a photodynamic process to generate singlet oxygen (1O2), which intramolecularly transfers to and efficiently reacts with the oxathiine block to form the afterglow oxathiine intermediates due to the low Gibbs free energy changes required for this photoreaction. Following intramolecular afterglow energy transfer from the oxathiine donor block to the acceptor block, NIR afterglow emission is produced from PDCDs. Owing to the efficient cascade intramolecular photochemical process, PDCD-based nanoparticles achieve a higher brightness and longer NIR emission compared to most reported afterglow agents, even after ultrashort photoirradiation for only 3 s. Furthermore, the cascade photochemical process within PDCD can be inhibited after bioconjugation with a quencher-linked peptide. This allows the construction of a cancer-activatable afterglow theranostic probe (CATP) that only switches on the afterglow signal and photodynamic function in the presence of a cancer-overexpressed enzyme. Thereby, CATP represents the first afterglow phototheranostic probe that permits cancer-specific detection and photodynamic cancer therapy under preclinical settings. In summary, this study provides a molecular guideline to develop afterglow probes from photoreactive polymers.Ministry of Education (MOE)National Research Foundation (NRF)Submitted/Accepted versionK.P. thanks the Singapore National Research Foundation (NRF) (NRF-NRFI07-2021-0005) and the Singapore Ministry of Education, Academic Research Fund Tier 2 (MOE-T2EP30220-0010 and MOE-T2EP30221-0004) for the financial support
Art buddy 2.0
No full textIn light painting photography, various light sources can be used. Traditional sources of lights like
candles, matchsticks and fireworks provide basic illumination effects while common handheld
tools like flashlights and glow sticks enable the creation of simple light trails and patterns. More
advanced sources such as LED lights significantly expand the possibilities by providing precise
control over brightness, color and dynamic effects. In particular, programmable LED lights allow
photographers to craft complex visual elements such as shapes, patterns and images, elevating
the potential of light painting photography to a whole new level. Therefore, this FYP project aims to develop dedicated light painting devices to create light trails
to aid photographers in creating beautiful patterns or render images which could not be achieved
by painting using a single light source. The first part of the project involves the use of the
Raspberry Pi Pico with a DC motor to create circular LED patterns while the second part of the
project involves the use of the ESP32 to create a handheld LED device.Bachelor's degre
Who to blame: algorithmic awareness and users’ perception of AI gender bias
No full textIn an increasingly algorithm-driven digital environment, the evaluation of algorithms’ performance pays more attention to the subjective perception of users. Previous research highlights that AI systems, from job application algorithms to language models, often raise users' perceived gender bias, partly due to societal biases encoded within AI training data and users' own bias. But in users' mind, to what extent is the AI gender bias due to AI itself, and to what extent do humans play a role in it? Algorithmic awareness, which is a progressive process where users basically recognize, critically understand, and rhetorically interact with the algorithms, is expected to decrease the perception and attribution of bias in AI. Through a 3 (algorithmic awareness: basic vs. critical vs. rhetorical) × 2 (AI output: high gender bias vs. low gender bias) between-subject experiment, we explore how participants attribute the biased outcomes and to what extent they engage the machine heuristic, which reflects the belief in algorithmic objectivity. We further examine the mediating role of the machine heuristic between algorithmic awareness and perceived bias, and whether this heuristic is moderated by the actual presence of gender bias in the content. In addition, the study extends to investigate how algorithmic awareness and bias perceptions jointly influence participants’ attitudes toward the AI system and their future behavioral intentions. The findings may offer insights into how deep interaction with AI may have complex effects on bias perception and suggest pathways for enhancing digital literacy in human-AI interactions.Submitted/Accepted versio
Trade Fragmentation or reinforcement? Assessing the impact of geopolitical risk on inter- and intra-bloc trade flows
No full textGlobal trade patterns are increasingly shaped by the influence of geopolitical risk (GPR), which has become a persistent feature of the international economic landscape. While conventional wisdom posits that heightened GPR dampens trade flows, recent events suggest that temporary trade adjustments—such as trade diversion, supply chain reconfiguration, and preemptive stockpiling—may partially offset immediate disruptions. This paper explores the impact of GPR on international trade, drawing on quantitative analysis of panel data to examine how trade flows respond to rising geopolitical tensions. Particular attention is given to the distinction between intra-bloc and inter-bloc trade patterns. Case studies from the U.S.-China trade war illustrate how firms and states dynamically adapt supply chains to mitigate exposure to geopolitical shocks. The paper employs interaction terms to correct for unobserved heterogeneity and isolate the true effect of GPR on trade flows. The findings reveal that while short-term adjustments create an initial buffering effect—reflected in positive trade responses—these are temporary, and the underlying trend aligns with a net decline in trade volumes under sustained high GPR. Understanding these dynamics offers valuable insights for policymakers, businesses, and scholars concerned with the resilience of global supply chains and the future of international economic integration.Bachelor's degre