DR-NTU (Digital Repository of NTU)
Not a member yet
116018 research outputs found
Sort by
Intelligent shipping: integrating autonomous maneuvering and maritime knowledge in the Singapore-Rotterdam Corridor
No full textDesigning safe and reliable routes is the core of intelligent shipping. However, existing methods for industrial use are inadequate, primarily due to the lack of considering company preferences and ship maneuvering characteristics. To address these challenges, here we introduce a methodological framework that integrates maritime knowledge and autonomous maneuvering model. Based on historical maritime big data, the framework offers customized routes for companies with specific routing preferences. The autonomous maneuvering model then evaluates the safety and reliability of the routes by considering ship motion characteristics and ocean hydrodynamics. We validate its effectiveness on the world's longest Green and Digital Shipping Corridor between Singapore and Rotterdam. Results demonstrate that our model can provide customized route design for companies and enhance safety for shipping. The framework could serve as a fundamental structure to build a fully digitalized platform for route customization and evaluation for global shipping, optimizing operational decision-making and safety assurance.Ministry of Education (MOE)Published versionThis work is funded by Singapore MOE AcRF Tier 1 Grant (Grant number: RG75/23), Singapore Energy Consortium (SEC) Core Project (Grant number: SEC-Core2024-34), National Natural Science Foundation of China (Grant number: 72101046), and China Scholarship Council (Grant number: 202306320340)
Deep learning methods with less supervision
No full textThis study examines weakly supervised semantic segmentation in video datasets, eval-
uating the effectiveness of various prompting strategies on the initial video frame.
The prompting approaches explored include segmentation masks from initial frames,
sparse point-based prompting, augmented point-based techniques, text-based object
grounding, and a combination of the methods. The findings indicate that the un-
derlying models effectively propagate key image features across frames, mitigating
challenges such as motion blur, perspective shifts, and occlusion. Augmenting point-
based prompts enhances segmentation accuracy by reducing semantic ambiguity, while
text-based prompting with a grounded object detection model offers a low-annotation
alternative for object segmentation and tracking. The integration of Grounding DINO
and SAM2 for text based prompting also shows strong results but is heavily reliant on
the precision of image-level semantic labels. Overall, the results validate the efficacy of
prompt-based segmentation in weakly supervised settings and underscore its potential
for generating accurate semantic masks in video analysis.Bachelor's degre
Online lifelong learning app with personalized recommendation
No full textThis Final Year Project report explores the existing online learning space, where there is a plethora of online learning platforms for users to learn from.
Through analysis of the existing platforms and the features they provider, this study aims to implement, extend existing features, and build novel features from the understanding of these learning platforms through development work of a lifelong learning platform, which our group has named Journey.Bachelor's degre
Neuro-imaging data analysis
No full textBrain network analysis using graph neural networks (GNNs) has become a powerful method for modelling neuroimaging data such as fMRI. However, class imbalance remains a major challenge in real-world datasets, where some diagnostic groups are significantly underrepresented. This issue is often overlooked in existing GNN-based models, limiting their robustness and clinical relevance.
In this project, we introduce an oversampling strategy specifically tailored for brain connectivity graphs. Our method integrates biologically informed augmentation steps, including interpolation, symmetry enforcement, noise removal, Laplacian smoothing and edge symmetry correction. Experiments on two benchmark datasets, ADNI and PPMI, demonstrate strong improvements across key evaluation metrics. Confusion matrix analysis shows more balanced class-wise performance, while saliency mapping highlights neurologically meaningful regions (ROI) in line with existing literature. The proposed approach offers a practical and interpretable solution to address class imbalance, with a strong potential to generalize to other neuroimaging datasets and clinical applications.Bachelor's degre
Towards sustainable aquafeeds: safe and consistent microbial protein grown on food-processing wastewater
No full textMicrobial community-based single cell protein (SCP) holds promise as a sustainable source of protein in livestock feed; yet its feed-safety and consistency in composition and production when using variable real-world wastewater has not been investigated. Here, the effect of heterogeneity in soybean-processing wastewater on SCP quality was tested using four replicate sequencing batch reactors over 92 days. The microbial community-based SCP grown on soybean wastewater demonstrated high consistency, with replicates showing similar patterns of biomass growth and protein accumulation. The dry microbial biomass exhibited a protein content of 39.8 ± 5.8 %, and the yield was 17.7 ± 1.7 g dry weight/g soluble total Kjeldahl nitrogen (sTKN). Azospirillum, a nitrogen-fixing bacterium, was the prevalent SCP-producing genus in all replicates at a relative abundance of 40.6 ± 5.1 %. The organism was not detected in wastewater, where Lactococcus and Weissella dominated. SCP contained essential amino acids to supplement conventional animal diets and was deemed safe for fish due to the very low abundance of fish-pathogen-like sequences (< 0.009 %) via metabarcoding. This study demonstrates the consistency of microbial community-based SCP derived from food-processing wastewater and addresses feed safety through pathogen screening, highlighting its potential to substitute protein in traditional animal feed and contribute to sustainable aquaculture practices.Ministry of Education (MOE)National Research Foundation (NRF)Published versionThis research was supported by the Singapore National Research Foundation (NRF) and Ministry of Education under the Research Centre of Excellence Program (EDUN C33-62-036-V4), and the NRF Competitive Research Programme (NRF-CRP21-2018-0006) "Recovery and microbial synthesis of high-value aquaculture feed additives from food processing wastewater"
Traffic control at smart road intersections
No full textThis project extends an existing single-intersection mutex algorithm by connecting independent intersection controllers to manage traffic flow across a network of intersections. While prior work has established centralized mutex control for isolated intersections, the expansion to a network introduces new challenges, particularly in managing intermediary road segments and preventing deadlock conditions. With each intersection operating as an independent decision-making unit, the study designs a traffic control system that maintains centralized control at individual intersections while employing distributed independent controllers at the network level. Through implementation and testing, we demonstrate that while the system does not achieve complete deadlock freedom, it maintains reliable operation under typical traffic conditions, with potential deadlocks emerging only during extreme congestion scenarios or rare circular-wait conditions. This presents the system a feasible alternative solution for traffic network management. As transportation systems increasingly integrate autonomous vehicles and wireless communications, this work contributes to the growing field of intersection management algorithms by providing insights into the challenges and potential solutions for scaling single-intersection control mechanisms to network-wide applications.Bachelor's degre
Radio-frequency (RF) sensing for deep awareness of human physical status
No full textThis project explores the use of WiFi-based Radio-Frequency (RF) sensing for human activity recognition (HAR), enabling deep awareness of human physical status. Traditional HAR methods rely on wearable sensors or vision-based systems, which can be intrusive, impractical, or constrained to idealized environments. This study leverages Channel State Information (CSI) extracted from WiFi signals to recognize human activities involving multiple individuals using standard household WiFi routers and smartphones.
A comprehensive dataset was collected using commercially available WiFi devices, capturing various activities in dynamic conditions. Several deep learning models were implemented and compared, including Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Attention-Based GRU (ABGRU), CNN-GRU, Siamese GRU (SGRU), and Transformer-based networks. The results demonstrate that the top-performing models achieve over 92\% accuracy in activity classification.
This research underscores the feasibility of WiFi-based HAR for applications such as smart homes, elderly monitoring, and security systems. Future work will focus on expanding the dataset, improving model robustness in noisy environments, and enhancing real-world deployment feasibility.Bachelor's degre
Adhesive micro-liquid for efficient removal of bacterial biofilm infection
No full textBacteria are common infectious pathogens that can cause invasive and potentially life-threatening infections. Ionic liquids have emerged as a novel class of alternatives to antibiotics, however their inherent hydrophobicity and immiscible in water exhibits poor adhesion to bacteria and diminishes its utilization and bioavailability for infection control. Herein, an adhesive metal phenolic encapsulated ionic liquid choline and geranate (CAGE@MPN) microcapsules is designed to address the aforementioned challenges and remove bacterial biofilm infections. The CAGE@MPN microcapsules are prepared through self-assembly of quercetin and ferrous ions on the interface of CAGE and water via metal-phenolic coordination. The MPN interface can stabilize the micro liquid and effectively adhere to bacterial surfaces. The microcapsules can disrupt bacterial cell walls to facilitate the release of cellular contents and destruct the biofilm, thereby exerting a pronounced bactericidal effect. The in vivo bactericidal effect of CAGE@MPN microcapsules is demonstrated in a murine model of Staphylococcus aureus (S. aureus) skin infection. The proposed adhesive micro-liquid system offers a promising strategy for noninvasive and efficient removal of bacterial biofilm infection.Published versionThis work was supported by the National Natural Science Foundation of China (82470981, 82101042, 52100191, 82320108004, 82301081), National Key Research and Development Program of China (2023YFC250630), the Taishan Scholars Program of Shandong Province (tsqn201909180, tsqn202312344), Shandong Province Key Research and Development Program (2024CXPT090, 2021ZDSYS18 and 2022CXGC020511), Shandong Province Major Scientific and Technical Innovation Project (No.2021SFGC0502), the Independent Training and Innovation Team in Jinan (No. 202228055), Natural Science Foundation of Shandong Province (CN) (ZR2022QB103), Open Foundation of State Key Laboratory of Oral Diseases (SKLOD2024OF07) and Qilu Young Scholar Foundation of Shandong University, Young Scholars Program of Shandong University
Deep learning-based side-channel analysis: exploiting vulnerabilities and explaining neural network decisions
No full textSide-channel analysis first dates back to the 1960s and has gained popularity ever since
1996 when Paul Kocher published a paper introducing the concept of timing attacks [1].
Subsequently, various physical leakages such as power consumption [2] and electromagnetic emanation [3] were exploited to recover secret information. These attacks pose a
significant threat to cryptographic systems because physical leakages may exist even if
the underlying algorithm/primitive is proven mathematically to be secure. Furthermore,
such attacks can be executed even without very expensive equipment or set-ups. As the
usage of Internet-of-Things (IoT) devices in the last decade keeps growing, side-channel
analysis has become a critical consideration in the design and implementation of secure
cryptographic systems.
In order to protect the devices against side-channel attacks, side-channel countermeasures
like hiding and masking have been proposed and implemented. However, the advent of
deep learning has significantly impacted the field of side-channel analysis. Especially
when countermeasures, which could protect against classical side-channel attacks, can be
easily circumvented through the use of neural networks [4]. In this thesis, the aim is
twofold. Our first objective is to improve side-channel analysis further through the use of
deep learning, while the second goal is to understand what successful neural networks are
learning so that the developers and the evaluators can know which area of the primitive
requires a change and defend against deep learning-based side-channel analysis.
We tackle the first objective in various ways. This is first done in Chapter 3 by looking
into the use of a generative neural network known as Denoising Diffusion Probabilistic Models (DDPM) to generate artificial traces automatically for classical attacks like
Template Attacks and Correlation Power Analysis. We show that DDPM can help generate artificial traces automatically that capture the underlying characteristic and improve
the performance of classical attacks. Next, we investigate the hyperparameters of neural
networks within the realm of deep learning-based profiling side-channel attacks. In Chapter 5, two loss functions known as Soft Nearest Neighbour and Center loss are explored
in profiling side-channel analysis. These loss functions use the intermediate features of the neural networks to improve the inter-class or/and intra-class distance. We show that
these loss functions help in their performances. Furthermore, we explore the use of multifidelity optimization techniques called Bayesian Optimization HyberBand(BOHB) to
allocate resources wisely when finding good hyperparameters for key recovery in Chapter 5. We show the effectiveness of BOHB by being the first to recover the secret key of
the CTF2018 dataset when using the identity leakage model.
The second goal of this thesis is to understand what neural networks are learning when
they successfully recover the secret key within the profiling side-channel setting. This is
investigated in Chapter 6 and Chapter 7. We proposed the use of the interpretable neural networks known as Truth Table Convolutional Deep Neural Networks (TTDCNNs)
in Chapter 6. These networks can convert their weights into SAT equations for inter-
pretation. Methodologies were proposed to analyze the SAT equations for side-channel
analysis, allowing evaluators to peek into what the network is learning. In Chapter 7,
an occlusion technique known as Key Guessing Occlusion (KGO) is proposed. This tech-
nique obtains the minimum set of sample points that a neural network requires for key
recovery. KGO is a model-agnostic algorithm. In other words, it can be used for any
DNNs. Evaluators are offered a different method to understand what the neural networks
have learned by providing the areas where the networks exploit the leakages. This would
allow evaluators to identify and address these issues within the underlying primitive. This
thesis advances the field of deep learning-based side-channel analysis by providing evaluators with techniques to evaluate their devices against deep learning-based side-channel
attacks. All proposed methodologies are empirically validated using real measurements.
Furthermore, they are also tested across a variety of platforms.Doctor of Philosoph
Explainable AI to interpret fusion of neuroimaging and multi-omics datasets
No full textParkinson’s Disease (PD) is among the most common neurodegenerative disorders;
however, early-stage diagnosis is challenging as their symptoms are subtle and com
monly overlap with those of other movement disorders. Although structural MRI
(sMRI)hasprovidedmolecularinformationtoidentify neuroanatomical changes inPD,
sMRI detects the visible structural atrophy that generally occurs in advanced stages of
the disease, making early diagnosis a significant challenge. In contrast, genetic factors
including single nucleotide polymorphisms (SNPs) associated with increased suscep
tibility to PD could allow for earlier prediction of risk. However, neurodegeneration
related to diseases cannot be directly measured with SNP data because the spatial
resolution is not sufficient. Thus, the complementary biomarker nature of the two
modalities indicates the ability to achieve higher diagnostic accuracy, earlier detection
and a more complete view of PD pathology by merging sMRI-based structural features
with SNP genetic information.
We introduced a multimodal deep learning framework with sMRI and SNP genetic
features to improve PD classification. An sMRI subnet based on DenseNet121 CNN
architecture was used to extract the structural features from the extracted sMRI scans,
and a dedicated SNP subnet was designed to accept genetic variations. A cross-modal
attention mechanism was employed to fuse these modalities to gain the advantage from
the complementary structural and genetic markers.
Individual and combined contributions of the identified brain regions and genetic vari
ants to classification decisions were explained using Explainable AI (XAI) techniques,
like Grad-CAM and Integrated Gradients, to facilitate interactions between the two modalities.Bachelor's degre