Report Title : Research Title Listing Rpt

No. Department College Supervisor Research Title Research Description URL

Excitons in 2d materials, especially twisted system serves as a sensitive probe for various condensed matter physics such as magnetism and correlated states.
1 EEE CoE Gao Weibo 2D material optics experiment*
Here you can join the 2D optics team to learn quantum photonics techniques and prepare for the next PhD stages.
2D materials are most fascinating material platforms in recent years. Especially, recently fractional quantum anomalous hall effect, magnetism and superconducting are all observed in the twisted
material system.
2 EEE CoE Gao Weibo 2D material transport experiment*
Here you can join the 2D material transport team to learn and conduct state-of-the-art experiments, to prepare for the PhD stages.
The advancement of generative AI technologies has opened new frontiers in medical imaging, promising unprecedented capabilities in data synthesis and interpretation. The project aims to enhance
3D Medical Image Analysis with Multi-modal the analysis of 3D medical images, such as Computerized Tomography (CT) and Magnetic Resonance Imaging (MRI), using Multi-modal Large Language Models (MLLMs), addressing the gap left by
3 LKCMedicine LKCMedicine Yeo Si Yong [Link]
Large Language Models previous research that predominantly focused on 2D images. Multimodal AI, incorporating capabilities such as computer vision and Natural Language Processing (NLP), is proving to be a crucial tool in
medical diagnostics and treatment. This research seeks to leverage these advances to improve the robustness and accuracy of 3D image analysis in the clinical setting.
In this project, we explore the use of computer vision to determine 3D human pose data. Our work focuses on the application of this tool to unpack student learning trajectories in open-ended learning
4 NSSE NSSE Edwin Chng 3D Pose Reconstruction for Education Research environments (i.e., with minimal teacher supervision). If you have a background in programming (preferred but not necessary), a keen interest in human motion analysis, and a drive to explore new [Link]
technologies, we encourage you to apply.
The response of a tree depends on its shape and each tree has a unique shape. To perform a more meaningful analysis of a tree response due to external loading, the 3D model of the tree needs to be
5 CEE CoE Leong Eng Choon 3D models of trees the basic input. This project investigates the optimum process of capturing the images of a tree under varying light condition and producing 3D model of the tree with minimum errors. Accuracy of the
3D model of the tree compared to the prototype tree is also evaluated.
Origami-inspired design provides a powerful approach to creating structures with programmable geometry and tailored behavior. This project investigates the use of 3D printing to fabricate origami-
inspired elements and to study their mechanical performance. The research focuses on translating origami patterns into printable geometries and evaluating their structural behavior under mechanical
3D-printed Origami-Inspired Structures for Civil loading.
6 CEE CoE Zhou Wen
Engineering Applications
During the visit, the student will participate in origami-inspired element design, parametric modeling, and preparation of 3D-printed specimens and gain hands-on experience in digital fabrication and
modeling.
Object detection is a fundamental task in computer vision, critical for applications such as autonomous driving, surveillance, and image retrieval. Deep learning has revolutionized object detection,
offering various approaches that significantly outperform traditional methods. This project aims to conduct a comprehensive comparative study of different deep learning techniques for object
A Comparative Study of Different Deep Learning detection, evaluating their performance, efficiency, and suitability for various applications. The student should (1) carry out a literature survey on the topic, (2) gain hands-on experiences using either
7 EEE CoE Wang Lipo [Link]
Approaches to Object Detection an existing program downloaded from the Internet or a new program coded by the student for an existing approach, and (3) then summerize and compare the advantages and disadvantages of
different deep learning approaches to object detection.

This project will identify new ways to solve NP-hard problems by algorithmically leveraging machine learning predictions, building on recent breakthroughs in learning-augmented algorithm design.
Drawing inspiration from complementary approaches, this research will synthesize the structural insights from various prediction models. We will explore ways to combine the pairwise ordering
predictions used by Bampis et al. (2025) for permutation problems with the degree-based partitioning strategies and edge-based predictions proposed by Aamand et al. (2025). Furthermore, we will
incorporate the theoretical underpinnings for constraint satisfaction problems with advice, as studied by Ghoshal et al. (2025), to develop a novel and more general algorithmic framework. The
objective is to create a new class of robust algorithms that strategically integrate imperfect heuristic guidance with traditional computational methods to find practical, efficient solutions for classically
intractable problems.

Themistoklis A New Approach to Solving NP-Hard Problems [1]: Bampis, Evripidis, Bruno Escoffier, Dimitris Fotakis, Panagiotis Patsilinakos, and Michalis Xefteris. "Polynomial time learning-augmented algorithms for np-hard permutation problems." arXiv
8 CCDS CCDS
Gouleakis with Imperfect Predictions preprint arXiv:2502.00841 (2025).

[2]: Aamand, Anders, Justin Y. Chen, Siddharth Gollapudi, Sandeep Silwal, and Hao Wu. "Improved Approximations for Hard Graph Problems using Predictions." arXiv preprint arXiv:2505.23967 (2025).

[3]: Ghoshal, Suprovat, Konstantin Markarychev, and Yury Markarychev. "Constraint satisfaction problems with advice." In Proceedings of the 2025 Annual ACM-SIAM Symposium on Discrete
Algorithms (SODA), pp. 1202-1221. Society for Industrial and Applied Mathematics, 2025.

Numerous highly-valuable Large Language Models (LLMs) have gained significant attention and are applied across diverse fields. To protect the Intellectual Property (IP) of the original owners over such
A Robust LLM Watermarking Technique against LLMs, red-green list-based watermarks have been extensively studied. However, most of such watermarks fail upon knowledge distillation attack, which utilizes input samples to query the target model
9 CCDS CCDS Dong Wei
Knowledge Distillation and obtains the corresponding outputs, thus training a substitute LLM using such input-output pairs. A fundamental task in robust LLM watermark against knowledge task is ensuring watermark
persistence through iterative distillation processes while maintaining output semantic coherence. Addressing this challenge is critical for enabling practical IP protection of LLMs.

In this 3-month project, we focus on the AI & Science paradigm, integrating advanced artificial intelligence technology with scientific research to generate innovative insights. In Science4AI, we draw
upon classic neuroscience and psychology theories, offering new approaches to designing and evaluating AI algorithms from the perspective of human-like modeling. In AI4Science, we leverage the
AI & Science: Exploring Cutting-Edge Research
10 SPMS CoS Cheong Kang Hao latest AI technologies, including large language models (LLMs) and multimodal LLMs, to support research in education, social sciences, medicine, and cognitive science. We offer a wide range of
with LLMs and Human-Like Modeling*
research directions, including visual object tracking, visual language tracking, AI agent design, medical image processing, and misinformation detection in LLMs. We aim to help you develop systematic
scientific research thinking and comprehensively improve your research capabilities.
The design and development of expressive visualizations often require professional knowledge about data visualization design and principles, which is often a tedious and time-consuming process. This
AI Agents for Automated Generation of
11 CCDS CCDS Wang Yong project aims to develop a novel approach using AI agents to build a feedback loop and further automatically generate desirable and customized visualizations, with little or even no human effort in this
Customized Visualization
process.

Studies on AI and the changing nature of work have explored users' emotions, attitudes, uses, and experiences. Yet, there is no cogent explanation as to how these aspects relate and interact and result
in various acceptance or rejections of recommendations from an AI agent. Research currently offers a technological deterministic perspective focusing on AI?s affordances that trigger workers?
cognitions and behaviours. Additionally, there is limited research on how workers' emotions (e.g., fear or opportunity), or attitudes (e.g., trust) towards AI impact behaviours.

This research project aims to contribute to research and practice by drawing expertise from three disciplines, i.e., information technology, communications, and medicine, towards four aims:
[Link] qualitative research to investigate users? emotions, attitudes, and experiences arising from the use of healthcare AI agents to develop a theory of acceptance or rejection of AI
recommendations.

[Link] a mixed-method approach to validate and generalize the theory developed above considering the role played by emotions, attitudes, and experiences towards accepting or rejecting an AI
12 ITOM Div CoB (NBS) Damien Joseph AI and the Changing Nature of Work
collaborator?s recommendation across contexts.

[Link] longitudinal methods to investigate changes in emotions, attitudes, experiences, and responses as healthcare professionals become familiar with collaborating with an AI agent.

[Link] extend the scope of the proposal to examine the role of acceptance/rejection of AI recommendations on downstream tangible outcomes experienced by healthcare professionals at a local
hospital.

The above four objectives build on established frameworks of theory construction and theory generalization.
 This research examines how government funding can catalyze AI-driven innovations to address climate change. We hypothesize that government funding and procurement policies play a decisive role
in stimulating the development of technologies that mitigate climate risks. Firms engaged as government contractors are particularly well-positioned to achieve breakthroughs in climate-related
innovations.
AI for Climate Innovation: Insights from
13 BF Div CoB (NBS) Zhimin Chen
Government Contracts
Methodologically, we construct a novel dataset of patents with government interest, linking them to funding sources, procurement contracts, and subsequent innovation outcomes. A key innovation of
this project is the use of large language models (LLMs) to extract and structure contract data?such as funding amounts, terms, and technology focus?from unstructured government documents. These
data, combined with machine learning techniques, allow us to rigorously evaluate the impact of government funding on technological progress in climate solutions.
This project is in collaboration with UC Berkeley (Prof. Joel Ager for electrocatalysis and Prof. Gerd Ceder for thermocatalysis) with automated labs. The work will be based on NTU, focused on high-
14 MSE CoE Kedar Hippalgaonkar AI for Heterogeneous Catalyst Development [Link]
throughput synthesis techniques with machine-learning guidance.
In this project, we explore the application of artificial intelligence (AI) technologies to enhance special needs education. Our work focuses on developing AI-driven tools and interventions that cater to
15 NSSE NSSE Edwin Chng AI for SPED the unique learning requirements of students with intellectual and physical disabilities, aiming to create inclusive, personalized educational experiences. We are seeking motivated and curious [Link]
undergraduates with a strong work ethic. If you are eager to tackle this challenge and make a tangible difference, we encourage you to apply.

With the continuous advancement of power electronics, magnetic components account for over 30% of both the cost and energy loss in nearly all power converters. The performance of magnetic
components has become a critical bottleneck in the development of high-performance power electronic systems. Although finite element simulations have effectively improved the design efficiency of
magnetic components, progress in magnetic component modeling (particularly under transient conditions) remains insufficient, especially in capturing the BH behavior. This project aims to leverage the
AI-Based Forecasting of Transient
latest advancements in artificial intelligence (AI) to develop an AI-based predictive model capable of forecasting the BH response under transient conditions. The student will work alongside a
16 EEE CoE Yang Yun Electromagnetic Responses in Soft Magnetic
postgraduate researcher to develop and optimize the AI model, with a focus on improving prediction accuracy and reducing model complexity. Based on an existing dataset, the project will explore
Materials
various neural network architectures and optimization strategies, ultimately identifying the most effective model. The final model will be validated using the database, demonstrating the advantages of
the AI approach in predicting the transient response of magnetic materials.

Understanding and predicting microbial community structures in environmental systems is critical for managing ecosystem health, biogeochemical cycles, and pollution remediation. This project aims
to develop an artificial intelligence (AI)-based framework to model and forecast microbial community composition and functional potential across diverse environmental contexts, such as soil,
freshwater, and wastewater treatment systems.
Leveraging high-throughput sequencing data and environmental metadata (e.g., pH, temperature, nutrient levels) available in the lab, the project will employ machine learning techniques?including
AI-Driven Prediction of Environmental Microbial random forests, neural networks, and ensemble models?to identify key environmental drivers and predict microbial taxa abundance and diversity. The integration of time-series data and spatial
17 CEE CoE Yin Xiaole
Community Dynamics mapping will enable dynamic modeling of microbial shifts in response to environmental perturbations.
A key innovation lies in the use of explainable AI (XAI) to interpret model outputs, providing insights into microbial-environment interactions and enhancing transparency for ecological decision-making.
The project will also explore transfer learning to adapt models across different ecosystems, improving generalizability and reducing data dependency.
Ultimately, this research will contribute to predictive microbial ecology, offering tools for environmental monitoring, early warning systems, and sustainable ecosystem management.

Machine learning and large language models are starting to act not only as tools, but as active ?research partners? that can search for new patterns, proofs, and algorithms. A striking example is
DeepMind?s AlphaEvolve, which can automatically write and refine code to solve hard mathematical and computational tasks. This project explores how such AI methods can be used for automated
discovery in algorithmic game theory and computational social choice.

The main idea is simple but powerful: we build AI systems that search, propose, and test new mathematical objects and algorithms. Examples include discovering voting rules with desirable fairness or
efficiency guarantees, or constructing hard instances and counterexamples for outstanding open problems in fair division and game theory. In many cases, these AI-generated discoveries can teach us
AI-Driven Scientific Discovery in Game Theory
18 MAS Div CoS Bei Xiaohui something new about the structure of the problem itself.
and Social Choice
Students in this project will (1) learn how modern LLM-based agents explore hypothesis spaces, (2) design pipelines that mix symbolic reasoning, simulation, and optimization, and (3) apply these tools
to core questions in game theory. The work sits at the intersection of theory and AI, so there will be space both for formal reasoning and for experimentation.

This project will suit students who enjoy thinking mathematically but are also curious about how AI can change the way research is done. A successful outcome would not only produce new theoretical
insights, but also contribute practical tools to support future automated discovery in economics and theoretical computer science.

The lack of artificial intelligence (AI) skills or talent has been cited as the number one challenge to using AI in scientific research and medicine. Large language models (LLMs) like ChatGPT have shown
remarkable performance for tasks based on human language and are a promising approach to address this challenge. They could offer an intuitive interface for users to tap into sophisticated deep
Accessible Deep Learning for Science through
19 CCDS CCDS Chan Guo Wei Alvin learning technologies. This project will focus on evaluating the current state of LLMs in scientific research and developing technologies to enhance their capabilities through science-related deep [Link]
Large Language Models
learning models. The work here will contribute towards making deep learning more accessible to scientific research and medicine, accelerating progress in these critical fields.

This project delves into the application of meta-learning techniques in the domain of financial forecasting, aiming to enhance the adaptability and accuracy of predictive models in dynamic market
environments. Traditional financial forecasting models often struggle to generalize well across diverse market conditions, requiring frequent retraining and manual adjustments. Meta-learning offers a
promising avenue to address these challenges by enabling models to learn from past experiences and adapt quickly to changing market dynamics. In this project, we will investigate one or two meta-
Adaptive Financial Forecasting with Meta-
20 EEE CoE Wang Lipo learning approaches tailored to financial forecasting tasks, such as stock price prediction, market trend analysis, and risk assessment. Through empirical experimentation using historical financial data, [Link]
Learning
we will explore how meta-learning algorithms can improve the robustness and efficiency of forecasting models, especially in scenarios with limited training data or evolving market conditions.

This project aims to develop a spatial wireless charging system specifically designed for underwater devices. Traditional planar wireless charging systems often suffer from limitations such as prolonged
docking times and unstable communication, which hinder operational efficiency. To overcome these challenges, the project will introduce innovations in coil structure, circuit architecture, and control
Advanced 3D Wireless Power Transfer System
21 EEE CoE Yang Yun strategies, while leveraging existing hardware platforms. The student will collaborate with research staff and postgraduate researchers, with a primary focus on developing AI-driven control algorithms
for Intelligent Underwater Devices
to enhance charging speed, accuracy, and system efficiency. This work is expected to contribute to the next generation of underwater wireless charging technologies, supporting Singapore?s leadership
in marine electrification and smart maritime solutions.
This project focuses on the innovative application of neural networks to electric motor control systems, aiming to significantly enhance their performance, adaptability, and robustness. Traditional
methods of motor control often struggle with nonlinear behaviors and variations in parameters, which can lead to suboptimal performance in various operating conditions. By leveraging neural
networks, we can implement adaptive and intelligent control strategies that are capable of adjusting to changing circumstances in real time. This approach is particularly beneficial for achieving precise
speed regulation, efficient torque control, and improved fault tolerance, thus ensuring smoother operation and greater reliability.
Advanced Electric Motor Control Systems with
22 EEE CoE Christopher H. T. Lee [Link]
Neural Network Architecture The project encompasses several key phases, including the design, analysis, simulation, and evaluation of the neural network-based motor control system. Extensive simulations will be conducted to
assess the effectiveness of the proposed control strategies under varying conditions. Should the initial results demonstrate exceptional performance metrics, the next step will be the development of
an experimental prototype for real-world testing and concept validation. This hands-on approach aims to bridge the gap between theoretical advancements in control algorithms and practical
applications, ultimately contributing to the future of motor control technology in various industries. Through this project, we aspire to set a new standard for motor control systems that can respond
intelligently to their environments.
Artificial Intelligence Data Centers (AIDCs) require highly efficient, reliable, and scalable power conversion systems to support increasingly dense computational workloads and stringent availability
requirements. Advanced multilevel converter technologies are key enablers for next-generation AIDC power supplies, offering superior efficiency, improved voltage regulation, reduced device stress,
and enhanced fault tolerance at high power levels.

Advanced Multilevel Converters for Ultra- This project focuses on the analysis and design of advanced multilevel converter topologies tailored for AIDC applications. Key technical aspects include modulation and control strategies, capacitor
23 EEE CoE Tang Yi Efficient Artificial Intelligence Data Center (AIDC) voltage balancing methods, and fault management mechanisms to ensure stable and efficient operation under dynamic load profiles and high-power conditions. Simulation-based modeling will be
Power Supplies employed to assess converter performance in terms of efficiency, power quality, transient response, and robustness.

Through this project, students will gain in-depth exposure to state-of-the-art power electronics technologies for AI data centers, with emphasis on system-level design considerations, reliability and
scalability challenges, and energy efficiency. The project is well suited for students with strong interest in power electronics, high-performance computing infrastructure, and sustainable power
solutions for next-generation data centers.
 'Our skeletons play a crucial role in regulating key physiological processes such as mineral homeostasis, energy metabolism, and blood cell production; however, how bones control these multiple
functions at the cellular level remains active area of research. The presence of multiple blood vessel (BV) subtypes and the distinct microenvironments they support contributes to the multifaceted
24 LKCMedicine LKCMedicine Saravana Ramasamy Ageing of skeletal microenvironments functions of skeletons. In this study, we investigate age-related changes in skeletal microenvironments and their relation to altered skeletal functions. We will explore why ageing results in reduced
repair and regenerative capacity of bone microenvironments. The student will learn and employ established advanced imaging techniques, flow cytometry and gene expression studies to analyze young
and aged mouse bones
Project Overview
--This project involves building an agent-based financial market where trading agents learn strategies through reinforcement learning (RL). The aim is to study how individual trading behavior
aggregates into market-level outcomes and how different strategies affect price dynamics, liquidity, and stability under counterfactual scenarios.

Key Objectives
--Implement an agent-based market simulator with interacting trading agents.
--Train agents using reinforcement learning to optimize trading objectives.
--Analyze emergent market behavior arising from strategic agent interactions.
--Explore counterfactual scenarios such as market manipulation or regulatory constraints.

Possible Tasks
--Design market environments (e.g., order book or simplified price formation).
Agent-Based Financial Markets with
25 SPMS CoS Patrick Pun Chi Seng --Implement and train RL agents with different objectives.
Reinforcement Learning Agents
--Run simulation experiments under varying market conditions.
--Analyze outcomes such as volatility, liquidity, and price impact.

Expected Outcomes
--A functioning agent-based financial market simulator.
--Experimental results demonstrating emergent market phenomena.
--A written analysis connecting micro-level strategies to macro-level effects.

Prerequisites
--Python programming
--Basic knowledge of reinforcement learning or willingness to learn
--Interest in computational finance and market microstructure
This research project explores the use of ammonia as a hydrogen carrier and its catalytic decomposition to produce hydrogen. The focus will be on developing and optimizing catalysts that can
26 MSE CoE Jason Xu Zhichuan Ammonia catalysis for hydrogen generation efficiently break down ammonia into nitrogen and hydrogen under mild conditions. Students will engage in catalyst synthesis, characterization, and performance evaluation. This project is perfect for [Link]
those interested in catalysis, green chemistry, and alternative energy sources, offering a comprehensive introduction to cutting-edge research in sustainable hydrogen production.

There are various control-theoretic notions and objectives of interest for a given control system (Lyapunov stability, asymptotic stability, exponential stability, etc.). Differential dynamic logic has
proven to be a useful formalism for specifying, organizing, and analyzing such properties and systems [A,B,C]. For example, it enables proof-theoretic analysis of such systems, combining classical
reasoning for discrete programs (loop invariants) with reasoning over continuous dynamics (continuous invariants) [C]. This project is about understanding how new forms of dynamics and their
stability properties can be fruitfully specified and analyzed using extensions of dynamic logic, for example, using temporal, adversarial, or stochastic extensions.
An Axiomatic Approach to Formally Verified
27 CCDS CCDS Tan Yong Kiam The interested student should have a strong background in formal methods with a focus on deductive verification; some relevant mathematics background courses, such as analysis, differential [Link]
Control
equations, or control theory, would be very useful.

[A] Deductive Stability Proofs for Ordinary Differential Equations, Tan and Platzer, TACAS 2021.
[B] Switched Systems as Hybrid Programs, Tan and Platzer, ADHS 2021.
[C] Verifying Switched System Stability With Logic, Tan et al., HSCC 2022.
This is a very chemistry but interesting topic. We are developing aqueous solution based electrolyte to construct aqueous Zn batteries to enable long cycles. This is more like cocktail strategy to ensure
the electrolyte elements can fulfil multiple functions and stablize both the cathode and anode sides of the battery. Student will learn basic properties and fabrication of the electrodes and batteries
28 PO PO Fan Hongjin Aqueous batteries: materials and devices [Link] (publication No 254, 218)
under guidance by senior member in the group.
Please refer to our website and publications: [Link]
AI is rapidly shaping how companies support their managers and manage their employees. Many companies employ AI technologies to support managers and employees (e.g., decision support system
Artificial Intelligence in Supporting Employees
29 ACC Div CoB (NBS) Yin Huaxiang using machine learning techniques) and to manage employees (e.g., machine learning techniques to set targets for employees, evaluate employee productivity, monitor employee behavior, determine [Link]
and Managing Employees
employee bonus, etc.). This project seeks to understand the impact of AI technologies in support manager and employee in decision-making and the impact of AI technologies in motivating employees.
As micro- and nanoplastics (MNPs) increasingly infiltrate marine ecosystems, concerns about their impact on seafood safety are growing, particularly in high-consumption regions like Singapore. This
two-month project will investigate the biological effects of MNP accumulation in edible seafood species commonly found in local markets. Whole tissues from selected fish and shellfish will be analyzed
Assessment of Micro- and Nanoplastic Toxicity in
30 CEE CoE Xiaole Charlotte Yin for histopathological changes?such as inflammation, cellular necrosis, or structural disruption?in organs known to accumulate contaminants (e.g., gills, liver, and digestive tract). These pathological
Edible Seafood Species
observations will be correlated with measured MNP burdens to evaluate potential dose-dependent toxicity. The study aims to provide early evidence of MNP-induced tissue damage in seafood,
informing food safety considerations and public health guidance.
Plant-based meat serves as an alternative to conventional meat, where it composes of proteins, fats, vitamins, and minerals. However, there are certain food safety risks in the ingredients and
processing of plant-based meat. A common ingredient of plant-based meat is legume (beans and nuts), which contains anti-nutritional factors such as phytates, tannins, trypsin inhibitors and
Assessment of anti-nutrients content in legume
31 CCEB CoS Chew Kit Wayne oligosaccharides. Several types of legume also contain allergens, especially in the family of seed and nuts. While this source has good health benefits, the anti-nutrients and allergen content has been [Link]
feedstock sources*
associated with negative physiological effects that can affect the digestive system and nutrient intake within the human body. This project proposes the assessment of anti-nutrient content in raw
plant-based meat feedstock to evaluate the presence of these anti-nutrients. The project will enhance the understanding of the food safety and risk concerns in plant-based meat.
Highly compact active nanophotonic devices integrated with silicon are important for future optical interconnects. One approach toward fabrication of nano-materials based optoelectronic devices
toward integrated photonic future.
Since the first graphene was isolated by researchers at University of Manchester in 2004, more two-dimensional (2D) materials has been intensively discovered. Transition metal dichalcogenides
(TMDs) are an emerging class of materials with direct bandgap semiconductor properties that make them highly attractive for fundamental studies of novel physical phenomena and for applications
Atomically-thin two-dimensional (2D) materials ranging from nanoelectronics and nanophotonics at the nanoscale. Recent research advancement of 2D materials is to assemble heterostructures having higher technological potentials and variable
32 EEE CoE Chae Sanghoon [Link]
based optoelectronics nano-scale device applications.
In this project, we will develop new class of optoelectronic devices that assembles each graphene and other 2D materials into heterostructures. Student will learn the basic skill to make nano-scale
building blocks and characterization/analysis techniques. This will be good chance to learn how to deal with nano-materials and optimize the device structures for high performance optoelectronic
applications.

Data visualization plays a crucial role in effectively communicating complex information, enabling extracting meaningful patterns and trends from large datasets. However, there are situations where
AudibleVis: A Spatial Audio Approach for
33 CCDS CCDS Wang Yong data visualization lacks accessibility, such as for blind and low vision (BLV) people who cannot fully engage with traditional visual representations like graphs or charts. This project aims to develop a
Accessible Data Visualization
spatial-audio-based approach to help BLV people easily access and interpret data visualizations.
In this work, the student will examine with a PhD student existing approaches as well as to propose new approaches for Audio deep fake detection and classification.

In this work, the focus is to examine explainability methods for state-of-the-art transformer-based audio deepfake detectors. Additionally, we will also consider deep fakes as out of class, and assume
34 CCDS CCDS Chng Eng Siong Audio DeepFake detection and classification
real data as in-class. I.e, we will examine methods such as one class classification, and assume outliers are any new types of deep fake data.

Metal organic frameworks and zeolites are a class of porous materials that find applications in catalysis, separations and carbon capture. Their applicability is limited by defects present inside them.
This project will focus on imaging these defects using TEM, simulating TEM images and using machine learning to segment images and identify unit cell defects within the MOF and zeolite crystal
structures.
Automated identification of angstrom scale
35 MSE CoE Prashant Kumar [Link]
defects in porous materials
Reference-
Kumar, P., Kim, D.W., Rangnekar, N. et al. One-dimensional intergrowths in two-dimensional zeolite nanosheets and their effect on ultra-selective transport. Nat. Mater. 19, 443?449 (2020).
[Link]
 This 3-month project focuses on automating the design of heuristics for complex search and optimization problems using Large Language Models (LLMs). Traditionally, crafting and configuring
heuristics requires expert knowledge and is time-consuming. By leveraging LLMs, we aim to automatically generate and optimize algorithms, making them more adaptable and practical for various
applications. Key objectives include:
Automating Heuristic Design with Large
36 SPMS CoS Cheong Kang Hao 1. Enhancing the feasibility of LLM-generated algorithms for real-world problems.
Language Models*
2. Improving LLMs' ability to generate coherent long-form content for complex problem-solving.
3. Developing techniques to optimize and refine these algorithms automatically.

Students will work on integrating LLMs with optimization techniques, contributing to innovative AI tools that simplify and advance heuristic design across multiple domains.
Our research group has long focused on developing functional polycyclic aromatic molecules for various applications. In this research, we aim to synthesize novel azacorannulene-based ligands and
Azacorannulene-Based Ligands for Transition
37 CCEB CoS Ito Shingo their transition metal complexes. The obtained metal complexes will be applied as a catalyst or photosensitizer for organic transformations. The student working on this project will be able to learn
Metal Complexes
basic skills and various analytical techniques on organic synthesis.
The rise of large language models (LLMs) has significantly impacted various sectors, from research to programming, enhancing productivity and facilitating complex tasks. However, this reliance on
LLMs also exposes users to a range of potential threats, including phishing, misinformation, and social engineering attacks. This study investigates the vulnerability of users to LLM-driven attacks,
exploring how different factors such as technical proficiency, trust in AI, and task complexity influence their susceptibility. We present an interactive framework designed to simulate and evaluate a
Benchmarking User Vulnerability to LLM-Driven
variety of LLM-driven attacks, focusing on phishing, misleading content, and malicious code injection. Through a controlled user study involving diverse participants, we benchmark their responses to
38 CCDS CCDS Dong Wei Attacks: A User Study on Perceptions,
these attacks, assessing the effectiveness of different mitigation strategies. The study also examines the role of real-time feedback mechanisms, user education, and interface design in enhancing user
Susceptibility, and Mitigation Strategies
resistance to deceptive LLM outputs. Our findings shed light on how user behaviors, demographics, and environmental factors impact the likelihood of falling victim to LLM-driven threats, providing
valuable insights for the development of more robust security measures and user awareness programs. This research contributes to the growing body of work on AI safety and user protection in the
context of emerging AI technologies.

As artificial intelligence (AI) has become increasingly integrated into learning and training environments, voice-based AI agents are emerging as powerful alternatives to traditional text-based
Beyond Text: Theorizing and testing the impact
interfaces. However, Information Systems (IS) literature has not fully examined how the acoustic characteristics of AI-generated voices influence human learning. This study addresses this gap by
39 ITOM Div CoB (NBS) Ben Choi of AI voice production and transmission on AI-
theorizing and empirically testing how two key auditory attributes, voice synthesis (generic vs. personalized) and voice transmission (radiation vs. conduction), affect learning
Guided learning
outcomes in AI-guided education.
In this project, we shall establish and characterize bio-mimicking scaffolds with topographical and sustained biochemical signaling ability to direct cell and tissue response. Methods include analyses
40 CCEB CoS Chew Sing Yian Bio-functional scaffolds for nerve regeneration* [Link]
cell/tissue-implant interactions through in vivo behavioral analyses and tissue morphometric evaluations and quantifications.
At the Rehabilitation Research Institute of Singapore (RRIS), we study biomechanics of lower and upper limb for subjects, typically elderly, engaging in activities of daily living.

41 MAE CoE Domenico Campolo Biomechanics: from labs to homes Equipped with state-of-the-art technologies ([Link] we seek to understand which portable technology is suitable to run similar (possibly simplified) analysis outside of [Link]
our structured lab, e.g. at home.
We seek students who are well-versed in coding, who can help us test portable technologies versus standard equipment available in our labs.
Data definition and data manipulation language platforms of aDBMS are often human-centric. For instance, declarative queries are often formulated by humans or their results are directly consumed
by humans. Consequently, superior and effective design of these platforms needs to be informed by theories from cognitive psychology, a branch of psychology that studies how individuals think and
Sourav Saha Bridging Cognitive Pyschology and Data
42 CCDS CCDS process information. In this research, we explore the emerging paradigm of psychology theory-informed design of human-centric components of a DBMS. [Link]
Bhowmick Management
In this paradigm, the design of the human-centric components is guided by theories from cognitive psychology in addition to theories from computer science. We invent novel techniques and
frameworks that embrace this paradigm in the arena of database querying.

Social computing platforms typically deal with data that are either related to humans or generated by humans. Consequently, effective design of these platforms needs to be cognizant of social
Sourav Saha
43 CCDS CCDS Bridging Social Psychology and Social Computing psychology theories. In this research, we explore the paradigm of psychology theory-informed design of social computing platforms where the design is guided by theories [Link]
Bhowmick
from social psychology in addition to theories from computer science. Specifically, we invent novel techniques and frameworks that embrace this paradigm in the arena of social influence.

Microporous polymers are promising membrane materials for gas separation due to their high permeability and molecular sieving capability. This project introduces students to polymer membrane
science through guided synthesis, membrane fabrication and performance evaluation. Students will explore basic structure-property-performance relationships by carrying out simple polymer synthesis
Building Smart Polymer Membranes for Efficient
44 CEE CoE Ye Chunchun and characterisation, followed by fabrication of defect-free membranes using solution casting. Membrane morphology, mechanical stability and gas permeability will be evaluated using standard [Link]
Gas Separation
characterisation and gas permeation techniques. The project is highly structured and achievable within two months, providing hands-on training in polymer chemistry, membrane fabrication and gas
separation measurements.
How can we know if a person or animal is in a state of emotion? Building on recent advancements in computer vision from the Nair lab at LKCMedicine, this project will utilize new vision transformer
Building a deep transformer-based system for
models to predict behavioral states of emotion across species, from mice to humans. The student will work with the lab to train new multimodal large language models which can segment and create
high-throughput behavioral analysis of emotion
45 LKCMedicine LKCMedicine Aditya Nair interpretable representations of diverse emotion states, including aggression (see see Nair et all., 2023 Cell and Nair et al., 2025 Science). The student will also explore how behavioral representations [Link]
states in models of neurodevelopmental
of emotions are altered in animal models of neurodevelopmental disorders. This research has the potential to advance high throughput screening of drug targets and create a platform for behavioral
disorders
phenotyping towards translational applications.
The Hippo/YAP signaling pathway controls cell growth and survival and is often deregulated in cancers. YAP is a transcriptional coactivator that translocates into the nucleus to promote the expression
of pro-proliferative genes via the transcription factor TEAD. The canonical Hippo core kinases MST/LATS phosphorylate YAP and thereby prevent nuclear translocation. The activity of MST/LATS is
controlled by several intracellular and extracellular cues including cell-cell junction formation, cell polarity, the actin cytoskeleton, and extracellular matrix stiffness. How YAP activity is regulated is an
important question in cancer cell biology. We have identified a family of scaffolding proteins, called Homers, that regulate YAP via a novel non-canonical mechanism. The goal of this project is to
understand fundamental aspects of how Homers function.
Characterisation of a novel mechanism to
46 SBS CoS Alexander Ludwig The candidate will use cell culture models, immunofluorescence and confocal microscopy to characterise the cellular localisation of YAP in response to Homer downregulation or overexpression. YAP [Link]
regulate YAP signaling*
activity in such cells will be measured via qPCR and TEAD reporter assays. In addition, immunoprecipitations will be performed to dissect how Homers interact with and regulate YAP. The candidate will
become familiar with techniques in molecular biology (cloning), mammalian cell culture, plasmid and siRNA transfections, immunoprecipitations, qPCR and TEAD reporter assays, SDS-PAGE and
Western blotting, as well as confocal microscopy and quantitative image analysis.

Please visit the Ludwig lab homepage for more information: [Link]

A fundamental question in developmental biology is how cells interpret intrinsic and extrinsic cues to make fate decisions. While the roles of transcription factors and niche-derived signals are well-
characterized, recent work has identified the mitotic spindle as a key mechanical regulator of cell fate, coupling cytokinesis with transcriptional reprogramming in adult systems. However, it remains
Characterizing spindle dynamics in pluripotent
47 LKCMedicine LKCMedicine Yi Liu unknown whether the spindle plays a similar role in early embryogenesis, where pluripotent stem cells first generate heterogeneity. To test whether the mitotic spindle drives cell fate decisions in early [Link]
stem cells
embryogenesis, this project aims to characterize and interfere with spindle behavior during the division of mammalian pluripotent stem cells (PSCs) under unbiased differentiation conditions. The
student will use advanced high-resolution live-cell microscopy and optogenetic strategies to observe and disrupt the spindle machinery.

The Polyphonic Futures Lab at NTU's School of Art, Design, and Media is seeking a design student to conduct research under the supervision of Asst/Prof Veronica Ranner for the pilot project
"Mattering and Designing with Transience." This project examines the intersection of posthumanist practices, materiality, and transience in design, with the aim of addressing ecological and social
Charting relations between design and tropical
48 ADM HASS Veronica Ranner challenges. The role involves active participation in critical and speculative design, thought experiments, and material studies, focusing on human impacts and ethical dilemmas within the bio-digital [Link]
biomaterial transience
convergence. This includes working with tropical biomaterials and insects. The successful candidate will document and catalogue transient temporal qualities and contribute to the early stages of
design projects that push the boundaries of biodesign and critical speculative design research at Nanyang Technological University.

At the Institute for Digital Analytics and Science (IDMxS), we are developing diagnostic tools that can detect molecules and pathogens at very low abundance. So doing can enable early disease
detection and facilitate better healthcare outcomes. Central to our approach is to use a digital paradigm, where instead of making bulk measurements of a patient?s sample, we measure individual
molecules thereby enabling precise concentrations to be found. Many species of molecules exhibit a property called chirality in which there is an asymmetry in the molecular configuration. The
imbalance between left and right handed configurations (or enantiomers) in a sample is of significant importance in a wide range of fields since enantiomers often exhibit drastically different biological
49 EEE CoE Matthew Foreman Chiral digital assays activities. For example, precise enantiomer composition of drugs can be vital to avoid severe adverse health effects (e.g. thalidomide), whilst a change in the enantiomeric ratio of some amino acids [Link]
can also signal the development of certain diseases (e.g. Alzheimer?s disease). This project aims to extend the concept of digital assays, to enable quantitative measurements of enantiomer imbalance
in sample. The student will develop the quantitative mathematical tools behind the technique and study theoretically how chirality parameters can be extracted from digital measurements. This project
is a great opportunity for a student who is interested in mathematics and statistics. The student will gain experience with a variety of statistical techniques, and they will have the opportunity to
contribute to the development and optimisation of new diagnostic tools that could have a significant impact on public health.
 Many machine learning and data science libraries are written in Python, for example, scikit-learn and PyTorch. The APIs in these libraries often have non-trivial usage constraints, such as option A can
only be used when option B is also enabled, or value C must be positive when option D is disabled. Users of these libraries rely heavily on the documentation to be able to use them correctly. But there
Code-Documentation Inconsistency Checking for
50 CCDS CCDS Li Yi are often inconsistencies between the library code and the corresponding documentation, which makes the usages of the APIs troublesome. In this project, we would like to automatically discover
Python Software Libraries
these inconsistencies at scale by leveraging the power of Large Language Models (LLMs) and advanced program analysis. Through this project, we will try to contribute to the documentation quality of
these software projects and enhance their user experience.
Artificial intelligence (AI) is reshaping education by enabling real-time feedback in language learning. This study investigates how the representation format of AI pronunciation feedback and learners?
Cognitive Fit in Human-AI Interaction: Adaptive cognitive orientations jointly influence pronunciation improvement. Drawing on Cognitive Fit Theory, we theorize that performance improves when instructional representations align with learners?
51 ITOM Div CoB (NBS) Ben Choi
Feedback Design for Pronunciation Learning cognitive processes. We distinguish between symbolic feedback, which provides phoneme-level textual and auditory cues, and demonstrative feedback, which offers perceptually rich visualizations of
the articulatory movements. We further distinguish between assimilative and accommodative orientations, which reflect learners? tendencies to integrate or restructure linguistic schemas.

We are interested in understanding how local behaviorial rules at the individual level affect organization patterns and functions at the population scale for biological systems such as cells. In this work,
we will investigate the collective behavior of cell monolayers using a stochastic interacting particle model. Our cells move in a random manner and can adapt their motion in response to their
52 PAP Div CoS Yong Ee Hou Collective motion of cells in a monolayer [Link]
neighbors. Moreover, the polarity and the velocity of a cell do not align due to intercellular forces. We will run simulations to understand interactions and patterns.

This project outlines a computational workflow for the optimization of antibodies and nanobodies, aiming to improve their therapeutic properties such as binding affinity, specificity, and stability. The
approach combines advanced in-silico methods, including molecular docking and molecular dynamics (MD) simulations with GROMACS and AMBER software, to model and refine protein-ligand
complexes. By performing free energy calculations (MM-PBSA/GBSA, FEP), the project will identify and quantify key interaction sites, or "hot spots," within the complementarity-determining regions
Computational Affinity Maturation of Antibodies
(CDRs). This data will then guide the introduction of targeted rational mutations, effectively designing improved antibody and nanobody variants in a streamlined, iterative process.
53 SBS CoS Julien Lescar and Nanobodies Using Molecular Simulations [Link]
The strategy is further enhanced by integrating AI and deep learning bioinformatics pipelines, which will be trained on experimental datasets to predict the impact of mutations on binding. This creates
and AI-Driven Bioinformatics
a powerful feedback loop: simulations validate AI-predicted variants, and the most promising candidates are subsequently tested using in vitro validation methods such as SPR, BLI, and ELISA. This dual-
pronged, integrated approach is designed to create a robust and efficient platform for developing high-affinity biologics, significantly reducing the reliance on traditional, time-consuming, and expensive
screening techniques.

In certain scenarios, Convolutional Neural Networks (CNNs) trained within specific parameter ranges necessitate substantial re-training when transitioning to different parameter ranges. This process
of re-training can prove to be highly time-consuming. Compounding this challenge is the potential unavailability of the requisite data. In light of these obstacles, this project endeavors to devise
Constructing Convolutional Neural Networks
54 EEE CoE Wang Lipo strategies for constructing CNNs capable of adapting to diverse parameter ranges without the need for extensive re-training. This approach hinges on a deeper comprehension of the decision-making [Link]
without Re-training
mechanisms inherent in CNNs. Through this endeavor, the student aims to develop methods that enhance the adaptability and robustness of CNN models across varying parameter domains.

With the development in mass spectrometry (MS) methodology and instrumentation, LC-MS analysis sensitivity has been drastically improved. At state-of-art, analyte loss due to non-specific
interactions with reaction containers become evident and a bottleneck in many high-sensitivity methods. This issue is currently addressed by coating the container with low-absorbent materials, such
Containerless/Surfaceless Targeted Proteomics as Protein LoBind tubes from Eppendorf.
55 SBS CoS Xueming Dong
via Acoustic Levitation and LC-MS* Here we aim to address this issue from an alternative direction. Acoustic levitation introduce pressure differences in the air and thus can be used to float liquid without the need of container,
essentially removing the context for non-specific interaction. Apart from bringing the current state-of-art for LC-MS analysis sensitivity to next level. Acoustic levitation also significantly reduces the
amount of waste associated with disposable containers, thus offering an additional advantage in terms of sustainability.
In this project, we seek to develop an innovative approach for teacher training with the use of large language models (LLMs) to create detailed student personas. These student personas will ultimately
Creating Student Personas with AI for
56 NSSE NSSE Edwin Chng be used in simulated microteaching environments to help educators refine their teaching techniques and better understand diverse student needs. If you possess a keen interest in AI, education, or [Link]
Microteaching Simulations
both, and hold a basic understanding of machine learning concepts and prompt engineering, we invite you to apply.
This project investigates the intricate mechanisms by which immune cells activate at the nanoscale. We will utilize state-of-the-art nanofabrication and advanced microscopy techniques to observe the
Decoding the Activation of Immune Cells at
57 CCEB CoS Zhao Wenting geometrical regulation of the key components in immune activation signaling at the molecular level. By deciphering these nanoscale events, we aim to develop a new strategy for targeted therapies
Nanoscale
that can modulate the immune response and potentially revolutionize treatment strategies for cancer immune therapy.
The two key research objectives:
1. Develop state of the art deep learning models to classify avian calls
2. Adaptation of above models to unseen urban and wildlife noise by using transfer learning.
Deep Bioacoustic Classification for Avian Calls [Link] [Link]
58 CCDS CCDS Chng Eng Siong
under the presence of urban and wildlife noise Transformer Models (e.g. google's BirdNET), are used in state of the art bio-acoustic classification. The project will explore transfer learning methods to adapt existing transformer-based classification research-group/BirdSet
models to unseen urban and wildlife noise.

Such a module will play a crucial role in reducing the manual labour required in ecological monitoring, species identification, and conservation efforts in a real-world setting.
This project aims to use deep learning to predict the charge mobility of organic semiconductors, a key parameter in the efficiency of organic electronic devices. Traditional prediction methods rely on
computationally expensive quantum mechanical simulations.
The project involves collecting a dataset of molecular structures of organic semiconductors and their charge mobility values. Deep learning models, such as artificial neural networks, will be trained on
Deep Learning based Charge Mobility Prediction this dataset to predict charge mobility based on molecular structures.
59 MSE CoE Li Shuzhou
for Organic Semiconductors Developing a deep learning model for charge mobility prediction in organic semiconductors has several important implications. It can significantly reduce the computational cost and time needed for
predictions, accelerating the discovery of new organic electronic materials. Additionally, it can provide insights into the structure-property relationships of organic semiconductors and aid in designing
more efficient organic electronic devices. This project contributes to advancing deep learning applications in materials science and could have an impact on the field of organic electronics.

The student project focuses on developing a deep neural network for disease classification from endoscopic images. Endoscopy is a crucial medical procedure used for diagnosing and monitoring
various diseases, including gastrointestinal disorders, respiratory conditions, and more. Automating disease classification from endoscopic data using deep learning can enhance the speed and accuracy
of diagnosis and treatment planning. The project will involve training a neural network to classify different diseases based on endoscopic images. Objective: The primary objective of this project is to
Deep Neural Network for Disease Classification
60 LKCMedicine LKCMedicine Yeo Si Yong design, implement, and evaluate a deep neural network model which can accurately classifying diseases from endoscopic data. The model should facilitate medical professionals by providing an
from Endoscopic Imaging
automated diagnostic tool that can facilitate in the early detection and monitoring of diseases through endoscopic procedures. This project gives students valuable experience in deep learning, medical
data analysis, and medical technology. It has the potential to make a significant impact on the medical field by providing a tool that can help medical professionals in diagnosing diseases and improve
efficiency of doctors.

The student project involves the development of a deep neural network for medical image segmentation, a crucial task in the field of medical imaging. Medical image segmentation is essential for
various medical tasks, including disease diagnosis, treatment planning, and monitoring of disease. The project will focus on using deep learning techniques to automatically delineate and classify
61 LKCMedicine LKCMedicine Yeo Si Yong Deep Neural Network for Medical Segmentation regions of interest within medical images, such as MR scans, CT, or microscopic dataset. Objective: The objective of this project is to design, implement, and evaluate a deep neural network model for
accurate and efficient medical image segmentation. The tool can be used for the segmentation of structures of interest from medical data which may facilitate healthcare professionals in the diagnostic
processes. Students will learn how to implement Deep Neural Network for medical data segmentation, improve on Python skills and may publish the project.

As autonomous shipping advances, Unmanned Surface Vessels (USVs) face significant risks when operating in congested environments like busy ports and narrow channels. In these settings, vessels
must react to surrounding traffic under uncertainty. Traditional rule-based navigation often relies on predefined thresholds, resulting in rigid and erratic maneuvers that compromise not only safety but
also energy efficiency. How can a vessel learn to navigate safely on its own while minimizing fuel or battery consumption?

This project answers this question by applying Deep Reinforcement Learning (DRL). Instead of hard-coding safety rules, we will train an AI agent to intelligently steer and adjust speed by observing
Deep Reinforcement Learning for Safe and
62 CEE CoE Kim Youngrong relative distances, headings, and velocities of nearby vessels. Using advanced algorithms like Proximal Policy Optimization (PPO) to learn a continuous control policy, the project focuses on optimizing [Link]
Energy-Efficient Autonomous Ship Navigation
the trade-off between collision avoidance and energy conservation in a single-vessel scenario.

Students will engage in training a DRL-based navigation policy within a maritime simulation environment, evaluating the AI's performance using comprehensive metrics such as collision rates, near-miss
frequency, path smoothness, and energy consumption. By the end of the project, participants will gain practical experience in autonomous systems, validating the capability of AI to enable safer and
greener autonomous shipping in complex maritime traffic.
A graph neural network (GNN) is a class of machine learning algorithms designed to handle data with an underlying graph structure. Graph representation learning has many applications, including in
63 EEE CoE Tay Wee Peng Deep graph neural networks [Link]
sensor networks, social networks, and transportation networks. This project investigates deep GNNs in non-traditional sensor network applications.
Deep learning neural networks for free energy
64 CCEB CoS Ni Ran We will develop deep learning neural networks to evaluate the free energy of molecular systems to understand the stability of molecular structures.
evaluations*
 Traditional computing architecture consists of processors and memories. However, this architecture is not energy-efficient for artificial intelligence and machine learning because of excessive data
transfer between them for computing. The concept of computing-in-memory (CIM) tackle this issue by minimizing the data transfer between processors and memories. However, most computing-in-
Design and Analysis of Digital Computing-in- memory macros use analog signals to represent digital computation results. Analog signals are susceptible to noises and environmental variations. Besides, analog-signal-based computing-in-memory
65 EEE CoE Kim Tae Hyoung Memory for Next Generation Smart IoT macros are not scalable. [Link]
Applications In this project, we will develop synthesizable computing-in-memory macros. The developed computing-in-memory macros will be compatible to the normal digital circuits design flow. This will facilitate
many circuits design companies to develop energy-efficient smart devices without significant design time overhead caused by analog signal processing. We will also co-optimize the computing-in-
memory architectures and circuits so that the overall energy efficiency can be further enhanced.
This project focuses on the design and fabrication of high-quality factor (High-Q) microring resonators, which are essential components in photonic integrated circuits (PICs) for applications such as
optical filtering, wavelength division multiplexing (WDM), sensing, and quantum photonics. Microring resonators are highly compact and efficient structures that enable precise control and
manipulation of light, making them critical for advancing modern photonic technologies.
Design and Fabrication of High-Quality Factor
66 EEE CoE Chae Sanghoon The project is divided into two main phases. In the design phase, advanced simulation tools such as Lumerical FDTD and COMSOL will be utilized to model and optimize the microring resonator's [Link]
(High-Q) Microring Resonators
geometry, material properties, and coupling mechanisms. Key design parameters, including ring radius, waveguide width, gap spacing, and refractive index contrast, will be carefully analyzed to achieve
high-Q factors, low insertion loss, and optimal resonance characteristics. The simulations will also assess critical performance metrics such as free spectral range (FSR), extinction ratio, and thermal
stability.
Contaminated wound dressings are a major biological hazard and source of medical waste. Furthermore, this issue is compounded with the increasing number of chronic non-healing wounds. Thus, a
new approach is required. This interdisciplinary project aims to completely rethink the development of advanced wound dressings through the utilisation of biodegradable biomaterials to enable
Design and development of biodegradable 3D decontamination and industrial composting. Using integrated advanced manufacturing processes, electrospinning and 3D printing, a multi-layer wound dressing comprising a barrier layer, adhesive
Paulo Jorge Da Silva
67 MAE CoE bioprinted and electrospun secretome releasing region, and therapeutic hydrogel delivery system will be developed. A 3D printed and electrospun barrier layer provides a protective environment for the wound, whilst a mussel-inspired adhesive
Bartolo
wound dressings coating will be used to reversible adhere to the skin. Finally, a bioprinted hydrogel containing a novel tissue regenerative secretome formulation will be incorporated to aid in wound healing. This
sustainable approach to wound dressing design; material selection and development; and manufacturing process integration offers an opportunity to solve a significant waste issue and clinical
problem.
This project aims to create a passive silicon (Si) photonic waveguide, a crucial component in photonic integrated circuits (PICs) that enables the manipulation and guidance of light signals for various
applications including optical interconnects, sensors, and optical communication systems.
The project is divided into two phases. The first phase will focus on designing the Si photonic waveguide using software tools like COMSOL and Lumerical FDTD. The design process will consider the
waveguide dimensions, material properties, and operating wavelength, and simulation software will optimize the design parameters and assess the waveguide's performance, such as dispersion, mode
Design and fabrication of passive Si photonic
68 EEE CoE Chae Sanghoon confinement, and propagation loss. [Link]
waveguide
In the second phase, the Si photonic waveguide will be fabricated using standard CMOS processing, starting with Si deposition on a silicon dioxide substrate, followed by waveguide patterning using a
lithographic mask, and then waveguide etching using reactive ion etching (RIE) to achieve the desired shape and dimensions. Finally, the waveguide's performance will be evaluated using various
characterization techniques like optical microscopy, scanning electron microscopy (SEM), and optical spectroscopy. The ultimate goal is to produce a high-performing Si photonic waveguide that can be
utilized in PICs and contribute to the development of new and improved photonic technologies.

Polyurethanes are widely used in applications such as footwear soles, seals, and damping components. In these applications, mechanical failure is typically governed by the growth of cracks rather than
by bulk strength alone. Once a crack forms, its growth under mechanical loading ultimately determines material durability, reliability, and service life.

Despite their widespread use, the fundamental design principles that control crack growth resistance in polyurethane elastomers remain poorly understood. Polyurethanes possess complex network
architectures with coexisting soft and hard domains that enable flexibility and toughness. While this hierarchical structure can dissipate energy during deformation, its effectiveness is often diminished
Designing Polyurethane Elastomers with High under repeated mechanical loading, leading to progressive damage and changes in mechanical properties. As a result, improvements in bulk mechanical properties do not necessarily translate into
69 MSE CoE Matthew Tan
Crack Resistance improved resistance to crack growth during service.

This project aims to establish general design principles for crack-resistant polyurethane networks by linking polymer network architecture to crack growth behavior. The insights gained will guide the
development of more durable polyurethane materials for applications subjected to repeated mechanical loading and demanding operating conditions. By improving crack resistance and extending
service lifetimes, these durable polyurethanes can reduce premature failure and help mitigate material waste.

Augmented Reality (AR) has emerged as a promising tool for enhancing science learning by providing immersive and interactive experiences. Prior studies highlight AR?s capacity to overcome physical
constraints of laboratories, support safe experimentation, and foster deeper understanding of invisible, abstract, and spatial concepts through enhanced visualization. Despite these affordances,
70 NSSE NSSE Jina Chang Designing and Implementing AR-Based Inquiry current research has primarily focused on developing AR-based inquiry activities around limited topics, with insufficient attention to pedagogical frameworks that guide teachers in integrating AR [Link]
across diverse areas of science curricula. To address this gap, this project seeks to design and apply a pedagogical framework for AR-based inquiry at the primary school level. By aligning AR?s unique
affordances with the diverse epistemic focus of inquiry types, the project aims to provide teachers with practical guidance for implementing AR in science classrooms.

Diabetes is a result of poor blood glucose concentration regulation and hyperglycaemia. Transplanted reconstituted pancreatic islet beta cell microtissues have emerged as a potential treatment option
for T1D and severe T2D whereby transplanted islet beta cells will secrete insulin in response to detected blood glucose to control hyperglycaemia. Most transplanted cell strategies contain only a single
cell type ? the insulin secreting beta cell. However, native islets contain a variety of cells directly involved in blood glucose homeostasis and literature has demonstrated potential benefits of
Designing therapeutic microtissues for treatment
71 CCEB CoS Dang Thuy Tram incorporating additional cells into such microtissues to recapitulate the native islet architecture and function for application in transplantation and subsequent blood glucose regulation of diabetic
of insulin-dependent diabetes*
sufferers. This project involves the modulation of microtissue composition containing different cell types and investigating the resultant quality and functional impacts for treatment of poorly regulated
blood glucose. The student working on this project will gain experience in experimental planning, cell culture and viability characterization as well as exposure to entrepreneurship in biotech towards
the formation of a start-up company.

Micro- and nanoplastics (MNPs) span a wide size range?from micrometers down to tens of nanometers?posing analytical challenges due to their heterogeneity and low detection limits at smaller
Detection and Size-Resolved Characterization of scales. This two-month project will focus on optimizing and applying a size-fractionated approach to detect and characterize MNPs across multiple size classes (<1 ?m, 1?10 ?m, and >10 ?m) in marine
72 CEE CoE Xiaole Charlotte Yin Micro- and Nanoplastics in Environmental environmental samples collected from Singapore?s coastal waters. The workflow will include sample digestion, sequential filtration or centrifugation for size separation, staining with Nile Red, and
Samples quantification using fluorescence microscopy and/or flow imaging. Polymer identification will be performed on representative particles using micro-Fourier transform infrared spectroscopy (?-FTIR).
The outcome will provide a preliminary size-resolved profile of MNP contamination, supporting future monitoring and source-tracking efforts.

Biofouling on ship hulls acts as a hidden penalty that significantly increases hydrodynamic drag and fuel consumption. However, current inspection methods rely heavily on manual review of
underwater footage by human divers or ROV operators. This process is not only time-consuming and costly but also prone to subjectivity and error, which makes it difficult to obtain consistent data on
hull conditions.

This project focuses on overcoming these limitations by developing a robust AI-based biofouling detection tool. Leveraging advanced Computer Vision and Deep Learning techniques, we aim to build an
Developing Intelligent Underwater Vision
73 CEE CoE Kim Youngrong automated system capable of analyzing underwater imagery to detect and classify various fouling types with high precision. The research addresses technical challenges such as image enhancement to [Link]
Systems for Hull Biofouling Detection
handle turbid water conditions and the application of semantic segmentation models to calculate exact fouling coverage on the hull surface.

Students will engage directly in the AI development pipeline, from preprocessing underwater datasets to training and fine-tuning state-of-the-art deep learning models. By optimizing algorithms for
accuracy and speed, participants will contribute to creating a scalable diagnostic tool that digitizes hull maintenance, which provides the essential technical foundation for data-driven and green
shipping operations.

This project focuses on developing a Boundary Element Method (BEM) code that accounts for poroelastic effects, crucial for understanding the complex interactions between fluid flow and fault
Developing a Boundary Element Method Code mechanics. By integrating solid-fluid interactions into a physics-based computational framework, the code will model the coupled system of poromechanics, fault friction, and fully dynamic earthquake
74 ASE CoS Luca Dal Zilio [Link]
for Poroelastic Effects in Earthquake Mechanics* rupture. This innovative approach will provide deeper insights into the processes that trigger earthquakes and influence fault slip evolution, both in natural settings and as a result of human activities.
The project aims to advance our ability to predict seismic hazards and contribute to safer geo-energy exploration practices.

This project involves the development, integration, and optimization of key sub-components for an advanced atomic gravimeter. The student will be tasked with building critical parts of the gravimeter,
Rainer Helmut Development and Integration of Sub- such as laser systems, vacuum chambers, and control electronics, under the guidance of the research team. They will also be responsible for integrating these components into the existing system and
75 SPMS CoS
Dumke Components for an Atomic Gravimeter* ensuring seamless operation. The project includes rigorous testing of each sub-component and the assembled gravimeter to validate its performance. This project aims to provide the student with
practical skills in precision engineering, system integration, and experimental physics, contributing significantly to the overall improvement and reliability of the atomic gravimeter.
 We have discovered that mature Staphylococcus aureus bacterial biofilms secrete a toxin that rapidly kills squamous cell carcinoma (SCC) cells at very low concentrations but has no negative effects on
human keratinocytes of fibroblasts. We have identified the primary toxin to be alpha hemolysin (Hla). SCC specific toxicity is exerted primarily through Hla-ADAM10 interaction, Hla oligomerisation
and pore formation. The high target specificity and potential to cause cell death in a controlled manner highlight Hla as a good candidate for an alternative SCC therapeutic.
We Aim to test the Hla toxin on a mouse model of SCC and optimise the delivery in different hydrogels for progressive stages of the condition. Our hypothesis is that Hla will kill SCC in vivo without
Development and delivery of a novel squamous
David Lawrence damaging the surrounding normal cells. Skin cancer is a serious problem worldwide, especially for people with pale skin in the Fitzpatrick 1 and 2 class living in sunny environments like Singapore. SCC
76 LKCMedicine LKCMedicine cell carcinoma therapeutic based on a bacterial
Becker typically develop on the sun exposed neck and face areas and account for about 20% of skin cancers, which is about 2.2 million people a year. Whilst basal cell carcinoma is more prevalent SCC is
biofilm secreted toxin
aggressive and more likely to metastasize and kill which makes it more dangerous. Excision of SCC at early stages is normally effective but sometimes they form near the eyes and it is hard to take a
large enough margin for safety. We will inject Hla in hydrogel into early stage papilloma and place the hydrogel into the open craters of SCC. We will explore hydrogels with short lifetimes of around a
day and others that can release drug for a week or more in order to optimise the treatment regime. Development of a novel SCC therapeutic will be of benefit to the Singapore economy.

Hybrid solid-state batteries (HSSBs) require heating to achieve optimal performance and fast charging capability, as their ionic conductivity is significantly limited at low temperatures. This project aims
to develop a heating system capable of rapidly elevating the temperature of HSSBs to their ideal operating range during the initial charging and discharging phases, and of precisely maintaining that
Development of A Heating System for Hybrid
77 EEE CoE Yang Yun temperature during steady-state operation. The student will work alongside a research staff member and a postgraduate researcher to implement the heating system and explore advanced control
Solid-State Batteries
strategies for achieving uniform temperature distribution among the cells. The proposed approach will be validated through simulation or experimental testing to assess its effectiveness and
practicality.
Critical thinking is a highly relevant competency for students to thrive in and beyond school. To foster students' scientific reasoning, an aspect of critical thinking in the domain of science, this research
project aims to develop a set of design principles for constructing what we call Claim-Reason Test (CREATE) items. These items are intended to develop as well as evaluate students' conceptual
understanding and reasoning in science.

CREATE items are based on Assertion-Reason Questions (ARQs), a type of MCQ reportedly used in US medical exams in the 1950s and UK A-level exams in the 1960s. ARQs are two true/false
statements (X, Y) linked (or not) by causal reasoning. They test students' conceptual understanding as students have to determine the independent correctness of X and Y, and decide whether causal
Development of Claim-Reason Test (CREATE)
78 NSSE NSSE Ong Yann Shiou reasoning exists between the two (i.e., whether Y is the cause of X). Based on the limited literature around the use of ARQs, ARQs better assess higher-order thinking than traditional MCQs, but are
Items for Scientific Reasoning
difficult to write. From our understanding, no design principles currently exist to guide the writing of such items. Additionally, when students choose the wrong option, particularly when they
erroneously attribute the causal reasoning, there is no further information for assessors to diagnose the possible student thinking behind. To overcome the abovementioned limitations, we propose the
CREATE item, which adds to the existing ARQ MQC format an open-ended response for students to elaborate on their selected option. Our project aims to distill design principles to guide practitioners
in the construction of CREATE items, so that the use of this assessment type for fostering students? scientific reasoning may be promoted. Currently, we are exploring CREATE items for
primary/elementary and lower secondary/middle grade students, targeting the scientific reasoning style of causal reasoning. Other areas of exploration include constructing items for high school
students and/or other styles of reasoning in science.

Development of broad-spectrum anti-Dengue Dengue is a critical threat to public health system, prevalent in tropical and sub-tropical countries. However, development of anti-Dengue drugs remains lackluster. Our lab has been focusing on
79 SBS CoS Xiao Tianshu
inhibitor* development of antiviral inhibitors targeting viral entry step. Here, we will utilize a variety of biophysical and biochemical approaches to identify and characterize novel entry inhibitors against Dengue.

The ability to engineer the genome and transcriptome of living cells lends itself to many biomedical and biotechnological applications. In recent years, CRISPR-Cas has emerged as a powerful system for
Development of new programmable RNA editing genome and transcriptome engineering. Briefly, a Cas enzyme is recruited to a target site by a programmable guide RNA. In so doing, it can also bring along an effector domain to modulate the target
80 CCEB CoS Tan Meng How [Link]
tools* gene. Here, we are interested in developing new Cas13-based technologies to install A-to-I or C-to-U editing events in RNA transcripts via site-selective recruitment of a deaminase domain. The tools
developed may be used as a new therapeutic modality for well-defined genetic diseases and can also be utilized to study RNA editing in various biological contexts.
Deep neural networks take many "features" as input, e.g., the pixels of an image, and map them into a small number of meaningful variables, e.g., to classify the image. They, therefore, act as
dimensionality or projector operator.
Massimo Pica Dimensionality reduction in Deep Neural
81 PAP Div CoS
Ciamarra Networks*
This computational and theoretical research explores the inner workings of DNN to shed light on the mechanisms by which this dimensionality reduction process occurs. Overall, the research
contributes to developing an understanding of the inner workings of these AI systems.
"Graphene, is a single layer (monolayer) of carbon sheet, is the thinnest compound known, the lightest material known, the strongest compound discovered, the best conductor of heat at room
temperature, and the best conductor of electricity known. In nature, layers of graphene stacked on top of each other form graphite, with a layer spacing of 0.33 nm. In 2004, researchers at University
of Manchester first isolated graphene with the so-called ?Scotch-tape? method, then they got a 2010 Nobel price. It is still used today to obtain single atomic layer flakes.
Disassembling 2D van der Waals crystals into
82 EEE CoE Chae Sanghoon Two-dimensional materials from layered van der Waals (vdW) crystals hold great promise for electronic, optoelectronic, and quantum devices, but technological implementation will be hampered by [Link]
centimeter-sized monolayers
the lack of high-throughput techniques for exfoliating single-crystal monolayers with sufficient size and high quality. In this project, student will demonstrate a facile method to disassemble vdW single
crystals layer by layer into monolayers with near-unity yield and with dimensions limited only by bulk crystal sizes. The macroscopic monolayers are comparable in quality to microscopic monolayers
from conventional Scotch tape exfoliation."
Coastal infrastructure is critical to protect Singapore and other coastal countries. There is an unmet need to monitor the conditions of this type of structure, especially for dikes, during their long service
Distributed and continuous dike monitoring
83 CEE CoE Fu Yuguang life under climate change effects. This project aims to develop cost-effective and scalable smart sensing technologies based on fiber optic sensing and IoT technologies. We will explore the feasibility of
under climate changes
using FOS + IoT technologies for dike monitoring through a series of lab tests.
Graphene, is a single layer (monolayer) of carbon sheet, is the thinnest compound known, the lightest material known, the strongest compound discovered, the best conductor of heat at room
temperature, and the best conductor of electricity known. In nature, layers of graphene stacked on top of each other form graphite, with a layer spacing of 0.33 nm. In 2004, researchers at University
84 EEE CoE Chae Sanghoon Drawing patterns on Graphene of Manchester first isolated graphene with the so-called ?Scotch-tape? method, then they got a 2010 Nobel price. It is still used today to obtain single atomic layer flakes. [Link]
Atomic Force Microscopy (AFM) is widely used to characterize the surface of graphene with sub-nanometer resolution. In this project, we will use the AFM not only for the surface characterization, but
also for drawing the patterns on graphene. By coordinating humidity and electrical bias to AFM tip, we can generate nano-pattern on graphene.
This study focuses on the vibrational analysis of a high-speed rotating spherical shell that rotates about its symmetric axis, based on the set of motion governing equations developed with consideration
of both the Coriolis and centrifugal accelerations as well as the hoop tension arising in the rotating shell due to the angular velocity, especially on the rotating spherical shell with Coriolis and centrifugal
[Link]
Dynamics of High-Speed Rotating Truncated forces as well as the hoop tension. A thin rotating truncated circular spherical shell with variable thickness is taken as an example for the free vibrational analysis, after numerical validation of the
85 MAE CoE Li Hua abstract/135/3/031006/472400/Free-Vibration-of-a-High-Speed-Rotating-
Shell present formulation. Several case studies are conducted for investigation of the influence of the important parameters on the frequency characteristics of the rotating spherical shell with variable
Truncated?redirectedFrom=fulltext
thickness. The parameters studied include the circumferential wave number, the rotational angular velocity, Young's modulus of the shell materials, and the geometric ratio of the thickness to radius of
the spherical shell.
Recently, a great deal of progress [1, 2, 3, 4] has been made in the development of highly efficient approximation algorithms for certain computationally hard (specifically, #P-hard) probabilistic
inference, counting, and sampling problems that arise in computer science. These algorithms, known as fully polynomial-time randomized approximation schemes (FPRASes), come with rigorous
guarantees on the accuracy of their output and runtime. Such algorithms have applications in several areas of computer science, including machine learning (e.g., inference and sampling in certain
graphical models) and data management (e.g., counting and sampling answers of a database query).

However, key challenges remain. Much of the work in developing these algorithms has been purely theoretical in nature, and there is still work to do in seeing how these approaches can be made
amenable to practical implementation. This project aims to do just that: develop a software tool that implements the aforementioned ideas, and explore its implications for real-world problems.

The ideal student would have a strong background in mathematics and/or computer science. Some knowledge of (or willingness to learn) a fast, compiled programming language (e.g., C, C++, Rust, ...)
Efficient Approximation Algorithms with would also be helpful.
86 CCDS CCDS Timothy van Bremen Guarantees for Probabilistic Inference and
Sampling [1] Arenas, M., Croquevielle, L. A., Jayaram, R., & Riveros, C. (2021). #NFA Admits an FPRAS: Efficient Enumeration, Counting, and Uniform Generation for Logspace Classes. Journal of the ACM (JACM),
68(6), 1-40.

[2] Meel, K. S., Chakraborty, S., & Mathur, U. (2024). A faster FPRAS for #NFA. Proceedings of the ACM on Management of Data, 2(2), 1-22.

[3] Arenas, M., Croquevielle, L. A., Jayaram, R., & Riveros, C. (2021, June). When is approximate counting for conjunctive queries tractable?. In Proceedings of the 53rd Annual ACM SIGACT Symposium
on Theory of Computing (pp. 1015-1027).

[4] Meel, K. S., & de Colnet, A. (2024). #CFG and #DNNF admit FPRAS. arXiv preprint arXiv:2406.18224.
 Pentameric ligand-gated ion channels (pLGICs) are crucial for synaptic transmission and are implicated in neurological disorders. Understanding their gating mechanism and modulation by lipids is
essential for therapeutic development. Prokaryotic pLGICs like GLIC have been instrumental in advancing our understanding of pLGICs due to their structural and pharmacological similarities to their
Elucidate the gating mechanism of pentameric
mammalian counterparts. Our recent cryo-EM research on GLIC has revealed insights into symmetric channel gating. However, previous MD simulations hinted at asymmetric channel opening.
ligand-gated ion channels (pLGICs), 5-HT3AR by
87 SBS CoS Sandip Basak Therefore, a new cryo-EM data processing pipeline has been developed in our laboratory to isolate asymmetric protein particles, yielding additional insights into asymmetric channel opening.
cryo-EM using a new in-house data processing
Furthermore, we found that lipidic modulators which alter the rate of desensitization also influence asymmetric channel gating. This indicates the potential link between structural asymmetry and
pipeline.*
channel desensitization. This work will be extended in studying human 5-HT3AR, a eukaryotic channel with similar gating mechanisms. In this project we will express the human 5-HT3AR in insect cell
expression system and purify in detergent environment. We will explore the asymmetric activation process using cryogenic electron microscopy (cryo-EM).
Pentameric ligand-gated ion channels (pLGICs) are crucial for synaptic transmission and are implicated in neurological disorders. Understanding their gating mechanism and modulation by lipids is
essential for therapeutic development. Prokaryotic pLGICs like GLIC have been instrumental in advancing our understanding of pLGICs due to their structural and pharmacological similarities to their
Elucidate the gating mechanism of pentameric
mammalian counterparts. Our recent cryo-EM research on GLIC has revealed insights into symmetric channel gating. However, previous MD simulations hinted at asymmetric channel opening.
ligand-gated ion channels (pLGICs), GLIC by cryo-
88 SBS CoS Sandip Basak Therefore, a new cryo-EM data processing pipeline has been developed in our laboratory to isolate asymmetric protein particles, yielding additional insights into asymmetric channel opening.
EM using a new in-house data processing
Furthermore, we found that lipidic modulators which alter the rate of desensitization also influence asymmetric channel gating. This indicates the potential link between structural asymmetry and
pipeline.*
channel desensitization. In this project we will express, purify and reconstitute GLIC in lipidic nanodisc. We will study the effect of specific lipidic modulators on GLIC using cryogenic electron
microscopy (cryo-EM).
The microstructure of clay-rich soils and rocks governs their mechanical, hydraulic, and transport behavior. Particle size and size distribution strongly influence how clay particles aggregate, align, and
form pore networks. However, the mechanisms linking particle size variations to microstructural fabric and connectivity remain poorly understood due to experimental and computational limitations.
Elucidating the Role of Particle Size in Governing
This project aims to clarify how particle size controls the microstructural characteristics of clay assemblies using a newly developed coarse-grained (CG) modeling framework. The CG approach bridges
89 CEE CoE Zheng Xiaojin the Microstructural Characteristics of Clay [Link]
nanometer- to micrometer-scale structures by representing each clay platelet as an interacting particle with effective forces derived from molecular potentials. Systematic simulations will quantify
Assemblies via Coarse-Grained Modeling
how particle size and distribution affect packing density, orientation, pore size, and connectivity. Results will be validated against experimental data for smectite-rich materials. The findings will
advance quantitative understanding of structure?property relationships in clays, providing new insights into natural and engineered systems.
EmoVis is a visualization system designed to support the analysis of students? emotional states using classroom video data. By integrating automated emotion recognition with intuitive visual analytics,
Yuvaraj EmoVis: Visualization of student?s emotions EmoVis enables educators and researchers to examine temporal patterns of student emotions at individual and group levels. The system transforms complex video-based emotion data into
90 SoLEC SoLEC [Link]
Rajamanickam using classroom videos interpretable visual representations, facilitating insights into engagement, boredom, and affective dynamics during learning activities. EmoVis aims to enhance emotion-aware reflection and data-
informed decision-making in educational research and classroom practice.
Organic solvent nanofiltration (OSN) membranes enable energy-efficient separations in pharmaceutical and chemical manufacturing. This project introduces students to polymer membrane design for
liquid-phase separations in organic solvents. Students will fabricate solvent-resistant polymer membranes using simple solution-casting and controlled crosslinking methods. Membrane stability will be
Enabling Effective Organic Solvent Nanofiltration
91 CEE CoE Ye Chunchun evaluated through swelling tests in representative solvents. Nanofiltration performance will be assessed using polymer and small-molecule solutes, including polystyrene in heptane and 1,3,5- [Link]
with Polymer Membranes
triisopropylbenzene in toluene, to probe size-dependent solvent flux and solute rejection. The project is highly guided and achievable within two months, providing hands-on training in polymer
processing, membrane characterisation and organic solvent separation techniques.
Context: Large AI models, such as those used in natural language processing, computer vision, and other advanced machine learning applications, require substantial computational resources, often
necessitating large GPU clusters. These clusters, while powerful, come with significant financial costs.
Serverless computing, a cloud computing paradigm, has demonstrated its effectiveness across various scenarios through its pay-as-you-go model. By abstracting the underlying infrastructure and
allowing automatic scaling based on demand, serverless computing helps organizations save costs and improve resource utilization. The flexibility and cost-efficiency of serverless computing make it an
attractive option for many applications.
Problem:
Despite the benefits of serverless computing, integrating it with GPU-centric workloads, particularly those involving large AI models, remains a significant challenge. The primary issues stem from the
substantial amount of data that needs to be transferred between storage, memory, and processing units. This heavy data transfer poses challenges to both the hardware and software system design,
including but not limited to:
Latency: High data transfer latency can severely impact the performance of AI inference tasks, making it difficult to achieve real-time processing.
Bandwidth: The available bandwidth between serverless functions and GPU resources can become a bottleneck, limiting the scalability and efficiency of the system.
Resource Allocation: Efficiently allocating GPU resources in a serverless environment, where workloads are highly dynamic and unpredictable, is complex.
Cost Management: While serverless aims to reduce costs, the unpredictable nature of AI workloads and the high cost of GPU resources can lead to cost overruns if not managed carefully.
Enabling Large-Scale Serverless AI Inference Integration Complexity: Combining serverless architectures with GPU acceleration requires seamless integration of diverse technologies, which can be technically challenging and may require significant
92 CCDS CCDS Dmitrii Ustiugov [Link]
Serving development effort.
Proposed Research
To address these challenges, the proposed research will explore the following areas:
Optimized Data Transfer Mechanisms:
Develop efficient data transfer protocols that minimize latency and maximize throughput between serverless functions and GPU resources.
Investigate the use of advanced networking technologies, such as RDMA (Remote Direct Memory Access), to reduce data transfer times.
Dynamic Resource Allocation Strategies:
Design intelligent resource allocation algorithms that can dynamically provision GPU resources based on workload demand and priority.
Implement machine learning models to predict resource requirements and optimize the allocation in real-time.
Cost-Efficient Serverless GPU Models:
Create cost management frameworks that predict and control GPU usage costs in a serverless environment.
Explore billing models that align with the dynamic nature of AI workloads, ensuring cost-efficiency without compromising performance.
Seamless Integration Frameworks:
Develop middleware solutions that facilitate seamless integration of serverless computing platforms with GPU clusters.
Beta-carotene serves as a precursor of vitamin A and provides relevant health benefits. Beta-carotene is a fat-soluble compound and its bioavailability from natural sources is very poor. Hence, studies
Loo Say Chye Encapsulating beta-carotene for improved
93 MSE CoE have been focused on the development of specific core/shell micro- or nano-structures that encapsulate beta-carotene to improve its bioavailability. Here, the aim is to formulate delivery systems to
Joachim bioavailability
encapsulate beta-carotene with the aim of improving bioavailability and shelf-life. Industrially-scalable techniques will be explored to translate this into a commercially-viable product.

Developing energy-saving passive cooling technology is crucial for achieving COP26? goal of carbon neutrality and sustainability. Passive cooling has wide applications from power electronics, food
preservation, to building coating. From application prospectives, passive cooling can be categorized to indoor and outdoor cooling. For indoor applications, such as cooling of power electronics, design
of materials and devices based on evaporative cooling is crucial. For outdoor applications, such as cooling paint/construction materials, both evaporative cooling and radiative cooling are important.
For instance, passive radiative cooling in tropical climate (e.g., Singapore, 1.3477N 103.6816E) is challenging due to the high humidity (84% on average), abundant rainfall (~167 days annually), intense
94 MAE CoE Li Hong Energy-Free Cooling solar radiation (up to 1200 W/m2 at daytime with 40% higher UV radiation), and strong downward atmospheric radiation (~20% higher than that of mid-latitude area), which halve the radiative cooling [Link]
potential; and thus far, there is no report of outdoor daytime sub-ambient radiative cooling under strong solar radiation (power density >1000 W/m2) in tropical climate. Besides, heat insolation,
conduction and dissipation play critical roles in different applications. In this project, we will develop materials and devices for passive cooling of primarily (1) power electronic device/system (indoor
application), and (2) buildings (outdoor application). Fundamental understanding of passive cooling mechanisms will be obtained, followed by rational design of the materials/devices. We anticipate
new materials/devices that offer new solutions to the energy-saving cooling of power electronics and buildings will be designed, fabricated, characterized, and evaluated.

Engineering fibre optic devices for expanded Microscopic images in clinical and biological research are usually small and cover very limited, submillimetre fields of view. In many demanding applications such as imaging of large organs or plates of
95 CCEB CoS Liang Kaicheng
fields of view cells, coverage should extend to the centimetre scale. The student will work on optical and mechanical designs for miniaturized microscopes capable of wide coverage and high resolution imaging.

Could our gut-brain biology be driving our unhealthy snacking behaviour? Building on this inquiry, the Hwei-Ee Tan lab at NTU Lee Kong Chian School of Medicine aims to understand how signals in our
intestines influence the brain, which then mediates our mood and behaviours (such as that of food craving). Experimental mouse models have been useful in allowing us to interrogate how the brain
Engineering in neuroscience: innovating a device controls behaviour, but phenotyping animal behaviours reliably remains a significant challenge. In this project, student(s) will join us in engineering a 'open-source' device that can continuously monitor
96 LKCMedicine LKCMedicine Tan Hwei Ee [Link]
to phenotype laboratory mice the food and water consumption of mice in their cages, which will then allow us to better characterise their feeding patterns. Student(s) should be interested in or comfortable with prototyping with
Arduino or similar microcontrollers, as well as making (wiring, 3D printing, etc) and programming. This project offers engineers who enjoy making and tinkering an interdisciplinary platform to innovate
and create in traditional neuroscience and biomedical research spaces, and to potentially create 'disruptive' devices in this emerging arena in future.

This project aims to enhance the performance of quantum gate operations in superconducting quantum processors, with the goal of reducing error rates and increasing fidelity. The student will start by
reviewing current techniques for optimizing quantum gates, followed by utilizing simulation tools to model and discover possible enhancements. They will then implement these improved gate designs
Rainer Helmut Enhancing Quantum Gate Operations in
97 SPMS CoS on superconducting qubits in collaboration with the research team. Extensive testing and characterization will be performed to assess the performance of the optimized gates. The student will
Dumke Superconducting Processors*
thoroughly document their process and results, culminating in a comprehensive final report and presentation. This project offers practical experience in quantum computing, focusing on essential gate
optimization for the development of reliable quantum processors.
 Biological microscopy usually relies on toxic molecules for fluorescent labeling or achieving transparency in tissue. It was recently discovered that certain food dyes can be used to make tissue samples
Enhancing optical transparency and bioimaging and living animals partially transparent. The student will study the principles underlying this new phenomenon and develop experimental protocols to exploit this capability for enhancing advanced
98 CCEB CoS Liang Kaicheng
quality in tissue using food dyes imaging techniques in the laboratory.

Many marine organisms, such as corals, oysters, barnacles, and mussels, build carbonate mineral skeletons or shells. These biocalcifying organisms not only play essential roles in marine ecosystems
but also serve as valuable indicators of environmental conditions. The trace element chemistry of these hard shelly parts reflects the environment in which they grow and can be modified by changes in
both the physical and chemical conditions of their surroundings, particularly in an era of rapid global and local environmental change.

In this project, the student will have the opportunity to participate in coral growth experiments conducted at the Singapore National Marine Laboratory. The project will involve the analysis of
carbonate minerals formed during these experiments, using advanced techniques at ASE, NTU, including Scanning Electron Microscopy (SEM) and Electron Microprobe Analysis.
Environmental change impacts on shells and
SEM will allow the student to examine the microstructure of the carbonate minerals at high resolution, providing insights into the morphology and crystallography that can indicate how environmental
99 ASE CoS Simon Redfern corals - how environment affects [Link]
stressors, such as temperature and pH, influence coral growth. The Electron Microprobe will be used to perform precise quantitative analyses of trace elements within the carbonate structure,
biomineralisation
revealing the geochemical signatures that correspond to specific environmental conditions.

By combining these techniques, the project aims to enhance our understanding of how changes in the environment impact the growth and health of biocalcifying organisms. The findings could have
broader implications for predicting the resilience of marine ecosystems in the face of climate change.

Methane (CH4) is a potent greenhouse gas (GHG) with significantly greater heat-trapping capacity than carbon dioxide (CO2), contributing to a third of GHG-induced global warming. Landfills (including
dumpsites) represent a substantial source of anthropogenic CH4 emissions due to the release of landfill gas (LFG) consisting of methane CH4 and CO2. The LFG emissions arise from anaerobic
100 CEE CoE Fei Xunchang Estimating methane emissions from landfills degradation of organic waste constituting municipal solid waste (MSW). The global annual CH4 emissions from landfills are estimated to be on the order of several tens of Tg, accounting for ~20% of all [Link]
annual anthropogenic CH4 emissions. In certain megacities, the contributions of landfills to city-wide CH4 emissions even surpass 50%. However, current estimation of landfill CH4 emissions is
inaccurate and incorporates high uncertainty. In this FYP project, you may improve the current methane estimation method and apply the improved method to major landfills worldwide.

Recent work by Selvam, Zhang, and Van den Broeck [1] proposes the Mixture of All Trees (MoAT) as a compact, fixed-structure graphical model which admits tractable computation of normalized
likelihood, as well as efficient estimation of marginal probabilities. Essentially, MoAT defines a mixture of all the possible n^{n-2} trees that can be drawn as a graphical model on n variables.

This raises the natural question of whether this idea can be generalized to other tractable models beyond trees (e.g., graphs of low treewidth), and how this affects the performance of the model.
Moreover, there is the question of connections to other results in lifted inference [2], as well as the development of alternative inference algorithms for the MoAT model. This project aims to address
these questions.
Evaluating Mixtures of Tractable Graphical
101 CCDS CCDS Timothy van Bremen
Models in Machine Learning
The ideal student would have a strong background in mathematics and/or computer science, and interest in machine learning. Some knowledge of Python (or willingness to learn) would also be helpful.

[1] Selvam, N. R., Zhang, H., & Van den Broeck, G. (2023, April). Mixtures of all trees. In International Conference on Artificial Intelligence and Statistics (pp. 11043-11058). PMLR.

[2] Van Bremen, T., & Kuzelka, O. (2023). Lifted inference with tree axioms. Artificial Intelligence, 324, 103997.

Lots of research has been done in evaluating the performance of Large Language Models (LLMs) in the NLP field. But there is still limited research in assessing the performance of LLMs on visualization
Evaluating the performance of Multimodal Large
102 CCDS CCDS Wang Yong related tasks. This project aims to conduct a systematic evaluation on the performance of multimodal LLMs in understanding visualizations, including both standard charts and customized
Language Models in visualization understanding
visualizations.
Extreme weather events such as flooding and landslides increasingly disrupt health systems in regions affected by climate sensitive infectious diseases. Beyond their effects on disease transmission,
these events can substantially alter physical access to healthcare by damaging transport infrastructure, modifying travel routes, and increasing travel time to health facilities. Such disruptions may
delay treatment, redirect patients to alternative providers, or deter care seeking altogether, with important consequences for disease outcomes, health system strain, and interpretation of routine
surveillance data.

This project aims to quantify how extreme weather dynamically reshapes healthcare accessibility and how these changes influence treatment seeking behaviour during outbreaks. The central objective
Event Driven Disruptions to Healthcare
103 LKCMedicine LKCMedicine Michele Nguyen is to develop a hazard responsive modelling framework that captures short term variation in travel time to healthcare facilities and links these changes to when, where, and whether individuals seek
Accessibility: A Dynamic Modelling Framework
care. Unlike conventional accessibility models that rely on static representations of transport networks and landscape characteristics, the proposed approach explicitly accounts for event driven
disruptions caused by natural hazards.

The project will develop an automated geospatial pipeline that uses satellite derived and other spatial data to update travel friction surfaces during extreme weather events, enabling estimation of time
varying travel times to health facilities and changing facility catchments. These dynamic accessibility estimates will support analysis of treatment delays and facility choice, adjustment for accessibility
driven biases in surveillance data, and anticipatory planning for healthcare access bottlenecks during extreme events.
Diffusion models have emerged as state-of-the-art generative models, excelling in domains such as computer vision, natural language processing, and interdisciplinary applications like drug design,
material discovery, and more. While highly effective, these models are limited by their reliance on the distributions of the training data, restricting their ability to generate data beyond these
Expanding the Capabilities of Diffusion Models: A
104 CCDS CCDS Ong Yew Soon boundaries. To address this limitation, diffusion target generation integrates external oracles to guide model outputs toward desired targets. This research focuses on the diffusion target generation [Link]
Guidance-Based Approach
problem and shall propose novel guidance-based algorithm to steer diffusion models toward generating target data, so as to achieve high computational efficiency and broad applicability on real-world
problems.
This research emphasizes the experimental design, synthesis, and characterization of advanced materials for energy and power applications, with a particular focus on batteries and hydrogen
technologies. By developing novel electrode, electrolyte, and catalyst materials, the work seeks to address critical challenges such as energy density, cycle stability, safety, and efficiency. State-of-the-
art techniques?including nanomaterials engineering, in situ/operando characterization, and electrochemical testing?will be employed to:

1. Develop high-performance electrode and electrolyte materials for next-generation batteries.

2. Design and optimize catalysts and storage media for hydrogen production, storage, and utilization.

3. Investigate fundamental structure?property relationships that govern energy conversion and storage efficiency.

4. Bridge laboratory-scale discoveries to practical applications, advancing scalable and sustainable energy solutions.

Candidates are expected to have strong experimental expertise in materials synthesis, electrochemical testing, and advanced characterization techniques, with a demonstrated interest in batteries
Experimental Materials for Energy & Power in and/or hydrogen-related technologies.
105 NSSE NSSE Edison Ang Huixiang [Link]
Batteries & Hydrogen Technology
Ultimately, this research will accelerate the development of sustainable energy storage and conversion systems, supporting global efforts in renewable energy integration and decarbonization.

This research emphasizes the experimental design, synthesis, and characterization of advanced materials for energy and power applications, with a particular focus on batteries and hydrogen
technologies. By developing novel electrode, electrolyte, and catalyst materials, the work seeks to address critical challenges such as energy density, cycle stability, safety, and efficiency. State-of-the-
art techniques?including nanomaterials engineering, in situ/operando characterization, and electrochemical testing?will be employed to:

1. Develop high-performance electrode and electrolyte materials for next-generation batteries.

2. Design and optimize catalysts and storage media for hydrogen production, storage, and utilization.

3. Investigate fundamental structure?property relationships that govern energy conversion and storage efficiency.

4. Bridge laboratory-scale discoveries to practical applications, advancing scalable and sustainable energy solutions.
 This project focuses on developing and evaluating machine learning models for assessing the damage of structural elements based on measurable features. The student will collect and organize
Explainable machine learning for damage datasets containing experimental results of tested structural components, including their geometric characteristics, material properties, and observed damage features extracted from images. Using
Bryan German
106 CEE CoE assessment of structural elements using damage this data, they will train models to predict remaining mechanical properties such as residual stiffness, force capacity, and drift capacity. The project will explore both explainable (glass-box) and non-
Pantoja Rosero
features explainable (black-box) machine learning approaches, perform quantitative comparisons of their predictive performance, and analyze the interpretability and reliability of the results. Through this
project, the student will gain experience in feature engineering, supervised learning, and model explainability for structural damage evaluation.

Exploring Cryptography for Secret In this project, we will try to implement RSA key exchange to establish secret keys in public communication channel, followed by AES-128 for secret communication. This will demonstrate the possibility
107 MAS Div CoS Guo Jian
Communications to establish secret communication channel from public. Little to none background in cryptography is required, while programming skills in C++ or python is preferred.

Halogens are crucial trace elements in Earth's geochemical cycles. Despite their low concentrations, these elements play a significant role in the physical and chemical properties of mantle minerals,
influencing processes such as melting, mineral stability, and volatile cycling within the Earth's interior.

The student will have the opportunity to conduct high-temperature and high-pressure experiments to simulate mantle conditions and investigate how halogens are incorporated into and released from
Exploring Halogens in Earth's Deep Interior: mantle silicates. These experiments will involve using advanced analytical techniques to measure halogen concentrations and distribution within the minerals, providing insights into the mechanisms of
108 ASE CoS Simon Redfern
Experimental and Computational Studies halogen incorporation and their potential effects on mantle processes. In parallel, we may employ computational methods, including density functional theory (DFT) and molecular dynamics
simulations, to model the interaction of halogens with mantle silicates at the atomic level.

This project will contribute to our broader understanding of the role of halogens in the Earth's deep interior, with implications for mantle geochemistry, volcanic activity, and the global cycling of
volatiles. The findings from this research could enhance our knowledge of deep Earth processes and inform models of Earth's geochemical evolution.
RNAs in living cells are modified in over 150 different ways. A major challenge is to map the locations of these modifications in endogenous RNAs. Existing methods suffer from different shortcomings,
Exploring RNA modifications with nanopore including reliance on antibodies that are not always available, lack of single-nucleotide resolution, inability to map multiple modifications simultaneously, and tedious workflows. Nanopore sequencing
109 CCEB CoS Tan Meng How [Link]
sequencing* has recently emerged as an attractive solution as RNA can be sequenced directly and modified nucleotides will give a different signal output from unmodified nucleotides. In this project, the student
will explore the use of nanopore sequencing to study RNA modifications in living cells.
Ultraviolet microscopy has been quite successful for tissue imaging, but the standard wavelength ~270nm used is absorbed by DNA and may cause damage to living samples. The DNA absorption peak
Exploring tissue microscopy across deep is narrow, suggesting that wavelengths just tens of nanometres away could have significantly improved safety profiles while preserving or even enhancing the advantages of ultraviolet imaging. The
110 CCEB CoS Liang Kaicheng
ultraviolet wavelengths for translational imaging student will explore the effects of using UV wavelengths from 200 to 300 nm on microscopy image quality and sample viability, paving a translational pathway towards safe ultraviolet microscopy for
healthcare.
Microalgae, particularly the Spirulina family, are known for their high phycocyanin content, making them a valuable source for natural pigments in food product formulations. Existing extraction
methods for phycocyanin often rely on volatile organic solvents, which pose toxicity risks and environmental hazards due to problematic solvent disposal. To address these issues, there is a need for
Extraction of phycocyanin from Spirulina using
innovative bioprocessing technologies that are both environmentally friendly and efficient. This project aims to develop a green and sustainable extraction method using a deep eutectic solvent-based
111 CCEB CoS Chew Kit Wayne deep eutectic solvent-based liquid biphasic [Link]
liquid biphasic system (DES-LBS) for the rapid recovery of phycocyanin from Spirulina. The study will evaluate the performance of DES-LBS under various operating conditions to determine optimal
system*
recovery, extraction efficiency, and partition coefficient. By leveraging DES-LBS, this project seeks to minimize environmental impact while maximizing the extraction yield of phycocyanin from
Spirulina.
Current privacy visualization research and practices are predominantly user-centric, focusing on data collection, usage, and sharing while overlooking organizational interests and incentives. Moving
Facilitating Rapid and Rational Privacy Decisions: beyond conventional approaches, this research proposes a novel visualization-based and protection-focused privacy communication mechanism designed to support user rapid yet rational app
112 ITOM Div CoB (NBS) Ben Choi Exploring the Role of Data Protection Label adoption decisions. We introduce a data protection label that visually conveys an app?s protection practices, and grounded in the Elaboration Likelihood Model (ELM), examine how three key label
Combining Behavioral features, data protection arguments (central route), amount of information and authenticity verification (peripheral cues), individually and interactively influence user decision-making while mitigating
cognitive biases inducing from prior privacy invasions. A behavioral experiment (Study 1) and an EEG study (Study 2) will be conducted.

The rapid proliferation of Large Language Model (LLM)-based web agents, particularly those operating solely on the User Interface (UI) level, has significantly enhanced user interaction with the web,
offering unparalleled accessibility and automation. However, this transformative technology introduces novel security challenges, especially since the majority of current research focuses on the
increasingly prevalent UI-only paradigm. This paper investigates the dual role of ?noise? ? web elements imperceptible or irrelevant to humans but interpretable by UI-only agents ? in both protecting
and exploiting these agents. We explore two primary scenarios: ?Faith,? where noise is strategically employed to prevent unauthorized agent actions like automated data scraping, and ?Fate,? where
noise is leveraged to mislead agents into performing malicious actions, such as clicking hidden advertisements or downloading malware. We delve into shallow attack techniques that exploit these
113 CCDS CCDS Dong Wei Faith and Fate: The Role of Noise in Web Agents
vulnerabilities, demonstrating how subtle webpage modifications can manipulate the behavior of UI-only agents. Simultaneously, we propose deeper defense mechanisms that leverage an
understanding of agent perception to mitigate these risks, focusing on both action generation and action grounding processes. We further analyze a case study of stealthy traffic redirection, where
agents are subtly guided towards malicious content while the user remains unaware. This work aims to establish guiding principles for designing secure web agent interactions, emphasizing the need to
balance usability with robust security measures in the face of evolving web threats, particularly within the context of UI-only agents. Our findings underscore the critical need for a paradigm shift in
web security to address the unique challenges posed by the increasing prevalence of LLM-based, UI-only web agents, ensuring their responsible development and deployment.

This projects aims to address a critical bottleneck in modern hardware security: automating the physical verification of Integrated Circuits (ICs) without compromising proprietary Intellectual Property
(IP). Currently, recovering circuit layouts via Scanning Electron Microscopy (SEM) image segmentation is a standard assurance process to detect hardware Trojans, yet training high-performance Deep
Learning models for this task is hindered by data silos, where semiconductor companies cannot share sensitive chip designs for centralized training. This project aims to overcome this barrier by
Federated Learning for Privacy-Preserving AI-
114 EEE CoE Gwee Bah Hwee implementing Federated Learning (FL), a decentralized approach that allows multiple entities to collaboratively train a segmentation model without ever exchanging raw image data. Students will [Link]
based IC Assurance
design and validate a framework where models learn locally on private datasets, sharing only mathematical updates to aggregate a powerful global model. By participating, students will work at the
cutting-edge intersection of Computer Vision, Cybersecurity, and IC assurance. This is a unique opportunity to build ?Trustworthy AI? solutions that solve real-world industry conflicts between data
utility and privacy, equipping students with highly sought-after skills and significant potential for academic publication.
Few-shot learning has emerged as a promising machine learning approach in the fields of molecular, chemical, and medical research, where obtaining large, labeled datasets can be challenging. This
research project focuses on developing and applying few-shot learning techniques to these domains, aiming to improve model performance in tasks such as molecular, chemical, and medical image
analysis.
The core objective of the project is to create models that can learn from a limited number of examples, thereby reducing the need for extensive data collection and annotation. In the molecular and
chemical domains, this involves leveraging existing knowledge of molecular structures and properties to make accurate predictions with minimal data. For instance, few-shot learning models can be
Few-Shot Learning for Molecular, Chemical, and trained to predict the activity of new compounds by learning from a small set of known compounds with similar structures. This has significant implications for accelerating the drug discovery process,
115 EEE CoE Wang Lipo
Medical Domains where identifying promising candidates early can save both time and resources.
In the medical domain, the project explores the application of few-shot learning in tasks such as disease diagnosis from medical images, where labeled data is often scarce. By training models that
generalize well from a few labeled examples, the research aims to enhance diagnostic accuracy, especially in rare or underrepresented conditions.
Overall, the project seeks to bridge the gap between data scarcity and the need for high-performance models, offering a versatile approach that can be adapted across various applications within the
molecular, chemical, and medical fields.

Few-Shot Image Classification focuses on training models to classify images with very few examples per category. Traditional models need large datasets, which are often impractical to obtain. Few-
shot learning aims to overcome this limitation by enabling models to generalize from a small number of examples. This project provides a foundation for understanding and implementing few-shot
Few-Shot Medical Image Classification: a learning. There are papers with codes on Few-Shot Image Classification at [Link] , focusing on medical images. The student will download an
116 EEE CoE Wang Lipo [Link]
Comparative Study existing code, reproduce the experimental results reported in the paper. Then the student should summerize existing research on few-shot medical image classification, comparing their advantages and
disadvantages.
 Innovation has long been recognized as a driver of firm competitiveness, but how firms communicate their innovation to stakeholders is less well understood. While prior research has focused on
reports, patents, or press releases, today's innovation narratives are increasingly shaped on social media platforms such as X (Twitter). Social media enables firms to frame innovation in real time and
at scale, making it a rich yet underexplored site for research.

This project will investigate how firms discuss innovation, with particular attention to the language, framing, and diffusion of innovation narratives. It will address questions such as: How do firms
construct stories of technological breakthroughs versus organizational or social innovations? What differences emerge across industries, firm sizes, and geographies? And how do stakeholders engage
117 ACC Div CoB (NBS) Xiang Zheng Firms' Discussion of Innovation on Social Media with and amplify these narratives in digital spaces? [Link]

Methodologically, the project is ideal for students with experience in natural language processing and large language models who wish to apply these tools to substantive questions at the intersection
of business and society. Approaches may include topic modeling, discourse analysis using transformer-based models, or LLM-assisted coding of complex narratives.

For prospective applicants, this project offers the opportunity to bridge technical expertise with theoretical insight in innovation studies, corporate communication, and digital media. It is well suited to
students eager to advance both computational social science methods and our understanding of how firms craft and project their identities as innovators in the digital age.

Light emitting photonic fibers have found extensive use in industries such as textiles, smart clothes, healthcare, and consumer goods due to their lightweight, flexible, and cost-effective nature. Unlike
conventional silica fibers, polymer fibers are a new emerging class of tools for sensing owing to their strong stretchability and versatile applications. This project seeks to develop an advanced and
efficient manufacturing process that will significantly improve the quality, versatility, and accessibility of polymer light emitting fibers. In this project, students will learn how an optical fiber works and
how a fiber is made. The student will learn together with senior students to fabricate flexible optical fibers and involve in the integration process on a chip to form a simple biomedical sensor. Simple
Flexeible Photonic Fibers for Biomedical Sensing biosensing experiments for proteins or cancer related biomarkers will be tested on a flexible fiber chip. For students who are keener in software, the student can learn how to use COMSOL software to
118 EEE CoE Yu-Cheng, Chen [Link]
on Chips design their fiber with different material and find the best size and structure of fiber, and then make the polymer fiber for real applications.
Lab homepage: [Link]

Refs: [Link]
Refs: [Link]
WebAssembly is a widely used virtual machine, now supported by all major Web browsers. WebAssembly's official standards body maintains a large corpus of tests, used to ensure the conformance of
its major implementations, including those in Chrome, Firefox, and Safari. However, these tests are written in a bespoke, undocumented test format called "WAST". This project would aim to properly
Formalising the definition of the WebAssembly document and formally model this WebAssembly test format, including the behaviour of ambitious concurrent extensions to WAST which are currently in discussion. Since other parts of the
119 CCDS CCDS Conrad Watt
test format WebAssembly language are already formally specified, this will bring the WAST format up to the same standard. The project member will become familiar with the basics of the WebAssembly
semantics and its approach to formal specification, the existing state of the WebAssembly test suite and the WAST format, and will work with members of the WebAssembly standards community to
formally specify WAST in an idiomatic way.
This project is about formalizing an interesting piece of mathematics relevant
to theoretical computer science in the Isabelle/HOL proof assistant. The exact
topic will be discussed later, but examples include: advanced results for ODEs,
120 CCDS CCDS Tan Yong Kiam Formalized Mathematics queueing theory, randomized algorithms, and so on (see link). Students [Link]
interested in this project MUST have background working with an interactive
theorem prover, e.g., via a course or from self-study. Strong background in
mathematics would also be beneficial.
We are interested in exploring how educators and professionals think about disability, and how these views shape the ways they work with students. We want to explore mixed-methods approaches to
examine when they lean toward emphasising skill development and when they provide supports that enable participation, and how these orientations influence opportunities for self-determination.
From perspectives to practice: How educators'
121 PCHD PCHD Kan Dang Dang Delia We hope this work will guide professional development for teachers and professionals who work with people with disabilities, while also contributing to better educational experiences for students [Link]
views of disability shape student opportunities
with disabilities, ultimately opening up more opportunities for their growth and participation in school, community, and future pathways. We are looking for students who bring a passion for
supporting diverse learners and expanding their opportunities for self-determination and growth.
This project aims to develop a methodology for automatically generating LOD3 building models with varying object detail from image data by integrating photogrammetry and deep learning?based
segmentation. The student will refine and extend existing methods that combine 3D reconstruction with semantic segmentation to produce models containing different sets of exterior components.
Bryan German Generation of LOD3 building models from images The work includes post-processing photogrammetric outputs, fine-tuning segmentation models, and integrating visual grounding models that identify and locate objects in images based on textual
122 CEE CoE
Pantoja Rosero with computer vision descriptions. The goal is to generate 3D building models where the overall level of exterior detail varies, from basic versions including only windows and doors to more detailed ones incorporating for
instance balconies, chimneys, stairs, fences, and surrounding vegetation. Through this project, the student will gain experience in photogrammetry, semantic segmentation, and 3D modeling for digital
reconstruction and visualization.
Overview
Obtaining large, high-quality side-channel datasets is expensive, time-consuming, and hardware-dependent. This project explores the use of generative AI models to synthesize realistic power or EM
traces conditioned on cryptographic inputs and limited real measurements.
Research Objectives
[Link] conditional generative models for side-channel traces.
[Link] traces for unseen plaintexts, keys, or nonces.
[Link] critical leakage characteristics used in SCA attacks.
[Link] usefulness of synthetic traces for attack development and countermeasure testing.
Generative AI for High-Fidelity Side-Channel
123 EEE CoE Gwee Bah Hwee Methodology [Link]
Trace Simulation
Using a small set of real traces captured from a target device, the student will train generative models (e.g., diffusion models or GANs) conditioned on cryptographic inputs. The model will learn
temporal, noise, and leakage distributions. Generated traces will be evaluated using standard SCA metrics (CPA success rate, TVLA, ML-based attacks).
Expected Outcomes
?A generative trace simulator requiring minimal real measurements
?Quantitative comparison between real and synthetic traces
?Applications in data augmentation, benchmarking, and attack prototyping
?Publishable results.

This project aims to address the critical challenge of data scarcity in hardware security, specifically the lack of large-scale, annotated Scanning Electron Microscopy (SEM) datasets required to train
robust AI models for IC layout recovery. While Deep Learning excels at segmentation, acquiring real SEM data is destructive, slow, and expensive. To combat this, this research aims to establish a
Generative AI for IC Assurance: A Synthetic Data comprehensive Generative AI Benchmark Suite using state-of-the-art image synthesis models (e.g., GANs or Diffusion models) developed by our team. These models generate high-fidelity "fake" SEM
124 EEE CoE Gwee Bah Hwee Benchmarking Suite for IC SEM Image images from IC layout patterns or cross-layer data, providing an infinite source of training data. Students will consolidate these disparate generative models into a unified repository to systematically [Link]
Segmentation evaluate their effectiveness in training downstream segmentation networks. The core objective is to quantify the "Sim-to-Real" gap: determining if segmentation models trained purely on synthetic
data can accurately recover layouts from real physical chips. Students will gain highly marketable skills in Generative AI evaluation, Data-Centric AI, and Computer Vision, while contributing a
standardized benchmark that could accelerate future research in automated hardware assurance.

A combination of AI, high-throughput experiments (robotics) and high-performance simulations can be used to accelerated materials development. For inorganic crystalline materials, structure
determines property. Therefore, property-driven generative design, driven by machine learning, critically requires understanding of the structure of materials. A deep understanding of crystal structures
125 MSE CoE Kedar Hippalgaonkar Generative Design for Materials [Link]
and their symmetries is essential for accurate generative design and the student will work with a team of graduate students on this exciting new frontier.
 This project focuses on the computational design of next-generation optical biosensors capable of detecting single biological particles, such as individual viruses or protein aggregates. While
conventional sensors rely on monitoring discrete points, this research explores the use of complex photonic networks?disordered meshes of interconnected optical waveguides?where the presence of a
single particle is detected through its perturbation of the global light scattering signature. The central challenge you will address is determining the optimal network topology that maximizes the
sensitivity and likelihood of detecting a particle landing anywhere on the chip surface. You will employ generative design algorithms to mathematically evolve the structure of these photonic graphs,
moving beyond standard regular lattices to discover non-intuitive, informationally optimal architectures. By modelling the sensor as a graph, you will utilize metrics from algebraic graph theory,
126 EEE CoE Matthew Foreman Generative design of photonic network sensors [Link]
specifically the Fiedler value (algebraic connectivity), to quantify the "information percolation" of the network, which serves as a proxy for how well the sensor can transport local scattering
information to the detection ports. Working primarily in Python or MATLAB, you will develop an evolutionary optimization framework that iteratively rewires the network connectivity to maximize this
sensitivity metric. The ultimate goal is to produce a computational blueprint for a biosensor that leverages engineered topological disorder to achieve single-particle resolution with minimal hardware
complexity. This work sits at the intersection of nanophotonics, graph signal processing, and algorithmic design, offering a unique opportunity to apply abstract mathematical optimization to the
concrete physical problem of ultra-sensitive biodetection.
We investigate data-driven control methods that exploit partial geometric structure in complex dynamical systems. The focus is on extending min-max optimization techniques for robust predictive
control within a geometric framework, enabling stability and performance guarantees under uncertainty. The project will involve mathematical modelling, algorithm development, convergence
127 SPMS CoS Cyrus Mostajeran Geometric data-driven control algorithms
analysis, and potential applications such as human-robot interaction, coordinated quadcopter systems, and robotic platforms where first-principles modelling is not wholly tractable but provides
valuable prior information.
Graph signal processing techniques and Graphs are commonly used to represent entities and relationships. Graph signal processing (GSP) has emerged as a powerful tool for addressing these problems by leveraging the domain knowledge of
128 EEE CoE Tay Wee Peng [Link]
applications graphs. In this project, we investigate the development of GSP frameworks for vertices with mixed or non-traditional signals. We aim to apply our frameworks to sensor network data.

The trillions of bacteria living in our body encode over 46 million genes, suggesting tremendous functional capacity (by contrast, the human genome has less than a thousandth of that). Moreover,
emerging evidence suggests that the gut microbiota profoundly influences host physiology and behaviour. In this project, we will investigate the fundamental biological basis of gut-feelings, how
Gut-Brain Neurobiology: Microbiome, Brain and
129 LKCMedicine LKCMedicine Tan Hwei Ee microorganisms and chemicals in the gut signal to brain to modulate neural substrates that control physiology and behaviour. The Hwei-Ee Tan lab at NTU Lee Kong Chian School of Medicine employs a [Link]
Behaviour
range of experimental techniques in mice to study the mammalian gut-brain axis including sequencing, metabolomics, in vitro microbiota anaerobic culture, in vivo multiphoton neural
imaging/recording, genetically-guided functional interrogation, as well as systems and behavioral neurobiology techniques.

Power inverters play a vital role in renewable energy systems by effectively converting direct current from eco-friendly sources and storage devices into alternating current. This research tackles
significant limitations associated with traditional inverter designs, particularly the heightened output harmonic distortion exacerbated by wide-bandgap semiconductor technology, which affects
performance in motor drive and grid interconnection scenarios. By utilizing asymmetrical cascaded H-bridge and hybrid gallium nitride/silicon topologies, this study aims to produce nearly sinusoidal
High Performance Inverters in Renewable Energy
130 EEE CoE Christopher H. T. Lee output voltage waveforms while reducing the need for large filtering components. While these advanced topologies enhance waveform quality, they also pose challenges in increasing power density [Link]
Systems
and refining control strategies. This study will focus on advancing power density through the integration of printed circuit board (PCB)-embedded magnetics and developing machine learning-based
control algorithms to bolster system robustness and dynamic responsiveness. The goal is to achieve a balance of high efficiency, increased power density, and excellent output quality in the resulting
inverter prototypes. These innovations aim to establish a solid foundation for the next generation of power inverter technologies for both motor drive and grid-connected applications.

Mid-infrared semiconductor laser with a broadband emission spectrum has attracted a lot of attention in recent years due to its broadband feature in the mid-infrared wavelength range, in which there
are many important applications including but not limited to biomedical sciences, imaging, security, spectroscopy, gas/chemical sensing, communications, military applications and astronomy sciences.
High performance electrically pumped
131 CoE CoE Wang Qijie In this project, the prospective student will focus on electrical and optical characterization of high performance broadband mid-infrared lasers from the bandstructure design, to waveguide design and [Link]
semiconductor lasers
optimization, from device processing in the cleanroom to device electrical and optical characterizations. After finishing the project, the student will learn the knowledge about spectroscopy and how to
characterize the performance of mid-infrared lasers, such as light-current, voltage-current, and spectra characteristics.

Optical fiber-based lasers are widely favored for their compact design, excellent beam focusing capabilities, and efficient thermal management. These lasers can produce optical power in the kilowatt
range, making them essential tools in industrial manufacturing, scientific research, and defense applications.

One of the key challenges in deploying high-power fiber lasers is the efficient transmission of laser beams over long distances to workstations. Traditional delivery fibers are typically limited to lengths
High power laser beam delivery through hollow- of about one meter, which significantly restricts their usability. Hollow-core fibers offer a promising alternative, enabling the delivery of high-power laser light over tens of meters from the source to
132 EEE CoE Chang Wonkeun
core optical fibers the target location.

This project focuses on developing effective methods for coupling laser light from a doped fiber laser into a hollow-core delivery fiber using fusion splicing. The work will involve designing and
constructing a doped fiber laser with the desired beam quality, implementing fusion splicing techniques to connect it to a hollow-core fiber, and addressing thermal management challenges at the
splicing interface.
High pressure physics on freestanding In this study, ultrathin membranes will be placed in a high pressure environment up to at least 20 GPa. Subsequently, their magnetic and electric properties will be studied. We expect to discover new
133 PAP Div CoS Wang Xiao [Link]
membranes properties in this high pressure condition.
This project focuses on the design and optimization of a high-efficiency axial-flux permanent magnet motor (AFPM) for compact electric propulsion systems, such as unmanned aerial vehicles (UAVs) or
electric vertical take-off and landing (eVTOL) aircraft. Compared to traditional radial-flux motors, axial-flux motors offer higher torque density and a more compact structure, making them ideal for
space-constrained applications.

Key tasks:
High-Efficiency Axial-Flux Permanent Magnet
134 EEE CoE Christopher H. T. Lee [Link] review on axial-flux PMM designs and their applications in compact propulsion systems. [Link]
Motor for Aerospace Applications
[Link] of an appropriate axial-flux motor topology (single-rotor single-stator vs. dual-rotor single-stator).
[Link] optimization using finite element analysis (FEA) to maximize efficiency and torque density.
[Link] and Thermal analysis to ensure stability and durability under continuous operation.
[Link] evaluation through simulation, comparing with conventional radial-flux designs.

Prompt alerting of ground shaking is critical in Singapore, where underground activities, surface traffic, and local earthquakes generate ground motions that may threaten public safety and
High-frequency Seismic Ground Motion infrastructure. While low-frequency (<1 Hz) ground motions can be reliably simulated using physics-based solvers, high-frequency (>1 Hz) motions are typically modeled statistically and remain
135 CEE CoE Yuan Congcong
Prediction via Deep Learning inaccurate. Moreover, existing ground-motion simulations are computationally intensive and cannot provide timely feedback to the public. To address these challenges, we invite visiting students to
develop a deep-learning-based ground-motion predictor that enables just-in-time, nationwide ground-motion mapping, assessment, and early warning.
GaN?s superior high-power, high-frequency performance thanks to its large critical electric field (EC) and saturation velocity (vsat), has positioned it as an excellent candidate for radar systems,
satellite communications, and electronic warfare. High-power RF devices based on GaN high-electron mobility transistors (HEMT) present a compelling departure from traditional vacuum technologies,
offering the capability to deliver kilowatts of microwave power, while remaining cost-effective and inherently modular. The performance of GaN electronics, however, is practically limited by the
constraints associated with the device structures rather than the intrinsic properties of the material. One major limit in high-frequency GaN electronics is the parasitic contact resistance of tunneling
junctions.

Electronic Metadevice is an innovative concept, in which a microstructured metal is coupled to the two-dimensional electron gas (2DEG) beneath through collective subwavelength electromagnetic
interactions, eliminating the need for electron injection seen in traditional devices [1].

Mohammad High-performance GaN Integrated Circuits based The main objective of this project is to implement, characterize, and model electronic metadevices in form of switches on a standard foundry process, and to explore the potential of three-terminal
136 EEE CoE [Link]
Samizadeh Nikoo on Electronic Metadevices GaN amplifiers based on electronic metadevice concept:

1) Development of ultrahigh-power GaN electronic metadevice switches on UMS foundry process with operation power >10 W mm?1 with operation frequency above 200 GHz.

2) Conducting a complete experimental study on the taped-out devices to evaluate their cutoff frequency, contact resistance, power handling, switching time, and breakdown voltage, and to
benchmark these devices against classic technologies provided by the foundry, to showcase the supremacy of our new device solution.

3) Development of a model for the realized devices for the sake of future MMIC designs.

[1] Samizadeh Nikoo, M. & Matioli, E. Electronic metadevices for terahertz applications. Nature 614, 451?455 (2023).
 Eukaryotic cells compartmentalize biochemical reactions and cell signaling events in membrane-bound organelles. These organelles actively communicate with each other at the membrane-contact
sites (MCSs). To facilitate this, membrane remodeling occurs at the MCSs remains unclear. Recently, I discovered an unexpected contribution of MCS to endosomal sorting. Excitingly, these
How does the organelle contact site contribute
137 SBS CoS Sho Suzuki observations provide new mechanisms for the understanding of membrane dynamics at the MCSs. The PhD candidate will utilize cell biological and biochemical approaches mainly using yeast and [Link]
to membrane remodeling?*
mammalian cells to study how organelle contact sites contribute to membrane remodeling. Loss-of-function mutations of genes encoding the MCS resident proteins are directly linked to neurological
disorders, including Parkinson?s disease. This research program is uniquely positioned toward new mechanisms underlying neurodegenerative diseases.
A novel marine bacterium has been isolated from Singapore water with the ability to change colour from white-green-red. This phenomenon is unknown and we cultivated a number of isolates, some
with stable colouration and other with the ability to change their colour and sequenced their genomes. We have also identified the red (a known antibiotic), but not the green natural product. This
Rebecca Josephine Identifying the genetic basis of colour changing in
138 SBS CoS project will use comparative genomics and bioinformatic approaches to determine: 1) are the colour variants due to genomic rearrangement, point mutation or a plasmid?; 2) does the colour variation [Link]
Case a novel bacterium*
reflect evolution throughout the genome or only in the phenotypic genes? Should these questions be answered, further work can be undertaken to identify the green coloured natural product and
characterise the strain to describe the novel species.
This project investigates how lysosomes interact with other organelles and biomolecular condensates within macrophage cells experiencing metabolic stress. By combining advanced imaging
techniques, proteomics, and cellular assays, we aim to identify key proteins and molecular mechanisms governing these interactions. The Global Connect Fellow will play a key role in this project by
Imaging Lysosome Dynamics in Stressed
139 SBS CoS Claudio Bussi leading the imaging experiments, utilizing cutting-edge microscopy to visualize and analyze lysosome-organelle interactions in real-time. They will also have the opportunity to gain experience in
Macrophages
complementary techniques such as proteomics and molecular biology by collaborating with other lab members. This research will shed light on the role of lysosomes in cellular adaptation to stress,
with implications for understanding diseases like neurodegeneration and cancer, where metabolic dysfunction and lysosomal impairment are prominent features.
This project will focus on simulating the electron beam path inside chiral crystals at experimental conditions. The student will learn the principle of electron beam transmission and scattering from a
quantum physics-based perspective, and apply that knowledge to understand how the electron beam travels at Angstrom length scales inside a chiral twisted crystal structure. Successful completion of
Imaging chiral crystals using sub-atomic electron
140 MSE CoE Prashant Kumar the project will result in new knowledge about twisting of electron beams at sub-atomic levels. [Link]
probes

Mechanical stress at the cellular level can directly damage DNA and disrupt the proteins that repair it. When cells are forced to squeeze through narrow spaces - a process called constricted migration -
their nuclei deform and sometimes rupture, leading to DNA double-strand breaks, chromosomal rearrangements, and mislocalization of DNA repair proteins. Normally, these injuries are repaired, but
in cells with suboptimal or defective DNA repair machinery, the repair response is inefficient, thus causing mutations to accumulate and potentially predisposing cancer formation.
141 LKCMedicine LKCMedicine Kong Li Ren Impact of Mechanical Stress in Tumour Evolution
This process becomes especially relevant in chronic fibrotic diseases, where excessive extracellular matrix (ECM) buildup generates confined spaces that cells must traverse during regeneration.
Clinically, patients with liver cirrhosis or idiopathic pulmonary fibrosis (IPF) face significantly higher risks of developing liver and lung cancers, respectively. This project aims to uncover how fibrosis-
induced mechanical stress accelerates genome instability and cancer initiation. Using transwell chamber, we will design experiments to define how mechanical forces disrupt DNA repair pathways.

Implementing Computer Vision Algorithms on Computer vision and machine learning algorithms play an important role in many applications today. However, most of the existing algorithms are complex and cannot run efficiently on embedded
142 CCDS CCDS Lam Siew Kei [Link]
FPGA for Edge Computing systems that have low computing capabilities. This project aims to design embedded systems for computer vision and machine learning algorithms. The student will explore design options on an FPGA.
CakeML ([Link] is a formally verified compiler, i.e., it comes
with a proof that the compiler will never miscompile input programs. This
strong correctness guarantee adds a layer of trust in developing
safety-critical applications.
Implementing and Verifying a Compiler Pass for [Link]
143 CCDS CCDS Tan Yong Kiam
CakeML This project is about developing a compiler optimization in the CakeML project. en%20label%3A%22student%20project%22
This optimization will need to be proven correct and integrated into the CakeML
compiler. This project is suitable for students who wish to learn compiler
verification and have strong background in functional programming and/or
compiler development. Knowledge of proof assistants would be beneficial.
This project focuses on leveraging meta-learning techniques to enhance the accuracy and adaptability of diagnostic models for skin lesion classification. Skin cancer diagnosis presents unique challenges
due to the diversity of lesion types and variations in patient demographics. Traditional machine learning approaches often struggle to generalize well across different skin conditions and patient
populations, leading to misdiagnosis and delays in treatment. Meta-learning offers a promising solution by enabling models to learn from a variety of skin lesion datasets and adapt quickly to new cases
Improving Skin Lesion Diagnosis through Meta- with minimal labeled data. In this project, we will investigate how meta-learning algorithms can improve the performance of skin lesion classification models by learning from diverse lesion
144 EEE CoE Wang Lipo [Link]
Learning characteristics and patient demographics. Through empirical evaluation using the publicly available ISIC (International Skin Imaging Collaboration) Archive dataset, we will assess the effectiveness of
meta-learning in enhancing diagnostic accuracy and robustness, especially in scenarios with limited annotated data or rare skin conditions. By developing and evaluating meta-learning-based skin lesion
diagnostic models, this project aims to contribute to more accurate and timely detection of skin cancer, ultimately improving patient outcomes in dermatology practice.

Fission nuclear energy is crucial for decarbonizing the global energy system due to its ability to produce reliable, carbon-free electricity. Unlike fossil fuels, nuclear fission emits no greenhouse gases
during operation, making it a key solution for achieving net-zero emissions. It provides a stable, scalable energy source to complement intermittent renewables like wind and solar.

Small Modular Reactors (SMRs) are transforming the nuclear energy landscape by addressing challenges of traditional large reactors. SMRs are more cost-effective, faster to deploy, and safer due to
their modular construction and passive safety features. Their smaller size and standardized design make them ideal for remote areas and regions with limited grid infrastructure. Additionally, SMRs can
Yi-Sheng (Eason) Improving material safety in small modular adapt to fluctuating renewable energy output, enhancing grid stability.
145 MSE CoE
Chen nuclear reactors
However, the adoption of nuclear energy and SMRs faces materials engineering challenges. Reactor materials must endure high temperatures, radiation, and corrosive environments over decades.
Innovations in advanced alloys, ceramics, and coatings are essential to prevent radiation-induced damage, embrittlement, and corrosion. Real-time monitoring and predictive modeling further enhance
safety.

By addressing these challenges, nuclear fission, including SMRs, can play a critical role in achieving a sustainable, low-carbon future, offering reliable, emissions-free energy to meet global climate goals.

The increasing demand for food and valuable biocompounds has triggered the search for novel and sustainable resources. Microalgae are a promising source of sustainable high-value nutrients.
Microalgae requires mainly water, nutrients and light at suitable intensity and wavelength to support its growth. The effect of different light plays a significant role in microalgal cultures and one
Influence of light sources and wavelengths on challenge is to observe how different light source can enhance or limit their photosynthetic efficiency and productivity. In this project, we will investigate the use of different light sources at different
146 CCEB CoS Chew Kit Wayne [Link]
the growth of microalgae* wavelengths to evaluate the growth and composition of microalgae culture. The photosynthetic efficiency of microalgae at different wavelength and light type used will be measured to understand
how changes in lighting can affect the growth and accumulation of functional compounds within the cells. The project will contribute to the development of a more sustainable approach by using
microalgae and its light-absorbing mechanism to naturally utilize CO2.

Antimicrobial resistance (AMR) is a complex, cross-sectoral threat that requires integrated surveillance and risk assessment across human, animal, and environmental health domains. In this project,
we aim to operationalize a One Health risk assessment framework by leveraging existing AMR abundance profiles generated in the laboratory from environmental and clinical samples.
The project will focus on integrating these high-resolution AMR datasets into a structured risk assessment model that captures the transmission dynamics and potential health impacts of resistant
Integrating Antimicrobial Resistance Abundance
organisms across interconnected systems. Using a combination of statistical modeling, network analysis, and machine learning, we will identify critical control points and quantify the likelihood and
147 CEE CoE Yin Xiaole Profiles into a One Health Risk Assessment
consequences of AMR spread through water, food, and animal-human interfaces.
Framework
By embedding AMR profiles currently available in the lab into a broader One Health context, this project bridges the gap between molecular surveillance and actionable risk management, contributing
to national and global AMR mitigation strategies.

As AI-guided systems increasingly assist human decision-making in complex physical tasks, their effectiveness depends on how humans learn from AI. This study examines the cognitive and emotional
Integrating Artificial Intelligence and Augmented
148 ITOM Div CoB (NBS) Ben Choi mechanisms shaping human learning from AI when learning instructions are delivered through augmented reality (AR). Grounded in cognitive load theory, we theorize how visual signaling and visual
Reality: A Pathway to Next-Generation Education
segmenting influence human learning performance. Multiple experiments involving eye tracking, GSR, and an augmented reality setup will be conducted.
 This study seeks to explore how generative AI can be systematically integrated into science inquiry-based learning (IBL) to enhance students? ability to investigate, reason, and draw evidence-based
conclusions. IBL emphasises question formulation, data collection, analysis, and reflection, but students often require scaffolding to engage deeply with these processes. The research will examine
Integrating generative AI into inquiry-based
149 NSSE NSSE Park Joonhyeong three focal areas: (1) how generative AI can support the design of investigation, (2) its role in facilitating interpretation of experimental data and identification of patterns, and (3) how it can guide [Link]
learning
students? construction of explanations and discussion. Through classroom-based interventions and mixed-method analysis, the study aims to produce a practical framework for integrating AI tools into
IBL pedagogy. The expected contribution is both theoretical and practical strategies to foster students' AI literacy and scientific reasoning for scientific practices.
In this project, we focus on the integration of computer vision outputs with large language models to enhance feedback mechanisms in educational settings. Imagine using advanced computer vision
Integration of Computer Vision Outputs with
techniques to analyze student actions and behaviors, and then leveraging powerful language models to provide insightful feedback to both students and teachers. This innovative approach aims to
150 NSSE NSSE Edwin Chng Large Language Models for Feedback in [Link]
personalize learning experiences, improve teaching strategies, and ultimately foster a more effective and engaging classroom environment. If you're eager to make a tangible impact on the future of
Educational Settings
education and want to be at the forefront of AI-driven innovation, we invite you to apply.
Characterizing underground activities is crucial in Singapore, where caverns, tunnels, and faults coexist with dense urban infrastructure and pose risks of mechanical instability. Seismic monitoring
Intelligent Extraction of Arbitrary Weak Coherent enables detection of weak microseismic signals and imaging of subsurface discontinuities such as fractures and faults. However, extracting coherent seismic phase curves across multiple channels
151 CEE CoE Yuan Congcong
Seismic Phases remains challenging due to low signal-to-noise ratios, limited labeled data, and strong anthropogenic noise. We invite visiting students to help develop a novel deep-learning framework for automated
seismic phase-curve picking. This framework will enable robust, scalable processing of large seismic datasets and enhance urban subsurface monitoring and risk assessment.

Diabetic foot infections (DFls) remain prevalent worldwide due to the lack of effective standard care of treatments. DFIs harbouring Enterococcus faecalis are associated with higher rates of chronicity
and amputation but its role in promoting chronicity is not well-studied.

Recent internal findings from Kline-Thibault labs indicate that E. faecalis delays wound healing through evoking the mammalian unfolded protein response (UPR) in cutaneous wounds. With UPR being
dysregulated in diabetes, we hypothesize that UPR could act as a synergistic node to delay healing in diabetic wound infections. As such, this project aims to identify how a diabetic phenotype and E.
Investigating synergetic outcome of host-
faecalis dysregulate the UPR to impaired wound healing. We are recruiting a final-year project student to join a PhD student in these investigations.
152 SBS CoS Guillaume Thibault microbe interactions in wound of diabetic [Link]
models
The FYP student will use cell biology, genetic, and biochemistry approaches to (1) screen for UPR mediated transcriptional dysregulations upon a diabetic phenotype in epithelial cells through gene and
protein expression studies (2) establish the role of these genes in diabetic wound healing through in vitro cell migration assays and (3) access these host UPR's contribution to bacterial infection and
growth to yield foundational knowledge for identifying therapeutically relevant interventions for chronic diabetic infections in vivo.

The student will work together with an experience team of experts.

The human gut microbiome is increasingly recognized as a key regulator of host physiology, with profound implications for cardiometabolic health. Our research focuses on uncovering the mechanisms
by which specific gut microbial communities and their metabolites influence the development and progression of cardiovascular and metabolic diseases. We are particularly interested in understanding
how microbial composition and function interact with host pathways such as lipid metabolism, inflammation, and vascular homeostasis.
To achieve these goals, we employ a combination of advanced experimental systems. Anaerobic bacterial culture platforms allow us to isolate and manipulate gut microbes under physiologically
Investigating the role of gut microbiome in relevant conditions. Using gnotobiotic mouse models, we investigate causal relationships between defined microbial consortia and cardiometabolic phenotypes in vivo. Furthermore, we incorporate
153 LKCMedicine LKCMedicine Kazuyuki Kasahara [Link]
cardiometabolic diseases state-of-the-art molecular biology approaches, including the use of genetically engineered microbial strains, to dissect microbial metabolic pathways and their impact on host tissues. These
complementary models provide powerful tools to establish mechanistic insights into host?microbiome interactions that cannot be achieved through observational studies alone.
This research program requires strong technical expertise and prior training in molecular biology, microbiology, or related disciplines. Experience with microbial genetics, animal experimentation, or
multi-omics approaches is highly advantageous. Through this work, we aim not only to advance fundamental knowledge of the gut microbiome?s role in human health, but also to identify novel
microbial targets and therapeutic strategies for the prevention and treatment of cardiometabolic diseases.

Reproducing results from recent research papers is often slow and error-prone due to missing implementation details, fragmented codebases, and unclear experimental settings. This project will build
an LLM-agent system that converts a target paper (PDF + appendix, and code repository when available) into a structured reproduction plan and a runnable experiment scaffold.

The agent pipeline will: (i) extract key claims, assumptions, experimental settings, datasets, and evaluation metrics; (ii) generate an executable checklist (environment setup, dependencies, configs,
commands, expected outputs, and sanity checks); (iii) propose minimal ablation studies to validate core contributions and isolate the source of reported gains; and (iv) maintain a reproducible ?lab
LLM Agents for Automated Reproduction of notebook? that logs every intermediate step (retrieved evidence, decisions, code diffs, configs, and results) in structured formats (e.g., Markdown/JSON).
154 EEE CoE Chau Yuen
Research Papers (Paper-to-Experiment)
During the fellowship, the student will implement a prototype with a modular agent architecture (e.g., Planner / Retriever / Implementer / Verifier) and apply it to a small benchmark set of 5?10 papers
in one focused domain (e.g., wireless communications or robot perception). We will evaluate the system using practical metrics: reproduction success rate, time-to-first-run, number of human
interventions, and common failure modes. The project emphasizes reliable tooling (containerization, deterministic runs, and standardized logging) and responsible use (grounded citations and explicit
uncertainty when information is missing).

Overview
Side-channel analysis (SCA) has traditionally focused on detecting leakage or extracting secret keys, but bridging the gap between observed leakage traces and the exact leaking implementation
remains an open challenge. This project proposes a novel LLM-assisted framework that not only detects leakage from power/EM traces, but also attributes leakage to specific code fragments in Verilog,
C, or gate-level netlists.
Research Objectives
[Link] a multi-modal learning framework combining side-channel traces and implementation artifacts.
[Link] models to map trace features to leakage sources such as registers, intermediate variables, or logic blocks.
[Link] LLMs to reason over code semantics and correlate them with statistical leakage indicators (e.g., TVLA, CPA results).
LLM-Guided Side-Channel Leakage Attribution
155 EEE CoE Gwee Bah Hwee [Link] generate explanations and candidate leaking code snippets. [Link]
and Code Recovery Overview
Methodology
The project will combine classical SCA methods with deep learning for trace feature extraction. An LLM will ingest annotated code (RTL, C, or netlist representations) together with leakage reports,
enabling structured reasoning about control flow, data dependencies, and hardware behavior. The system will output ranked hypotheses of leakage locations and corresponding code segments.
Expected Outcomes
?A prototype tool for automated leakage localization
?Demonstration on cryptographic hardware (e.g., AES, ECC, ASCON)
?Publishable results.

Medical imaging data analysis is critical for the detection and management of diseases. It is however a tedious and time-consuming process that demands high levels of expertise from clinicians. The
research project introduces an innovative deep learning model leveraging Large Language Models (LLMs), specifically utilizing the capabilities of the MedClip. (Contrastive Language?Image Pretraining)
156 LKCMedicine LKCMedicine Yeo Si Yong LLMs for the Analysis of Medical Imaging Dataset model to automate the analysis of medical scans. This model is adept at comprehending visual information and associating images with relevant textual data, enhancing interpretative accuracy. This [Link]
work will involve models in areas such as region segmentation, image understanding, and text generation. This research will significantly alleviate the burden on radiologists by automatic analysis of
radiology data and reducing the costs associated with training large numbers of specialists in this field.

With the advance of artificial intelligence (AI), the emergence of Google Gemini and OpenAI Q* marks the direction towards artificial general intelligence (AGI). To implement AGI, the concept of
interactive AI (IAI) has been introduced, which can interactively understand and respond not only to human user input but also to dynamic system and network conditions. In this project, we explore an
Large Language Models (LLMs) for Next
157 CCDS CCDS Dusit Niyato integration of Large Language Model (LLM) based IAI in networking. We will investigate the use of LLMs in enhancing network functionality, improving user experience, and promoting efficient network [Link]
Generation Network Design
management. We will also propose LLM-based network optimization framework, which consists of environment, perception, action, and brain units. We also design the pluggable large language model
(LLM) module and retrieval augmented generation (RAG) module to build the knowledge base and contextual memory for decision-making in the brain unit.
 Project Overview
--This project explores how well modern Large Language Models (LLMs) perform when applied to structured financial tasks. While LLMs demonstrate strong language capabilities, their reliability in
financial reasoning and decision-making remains unclear. The goal is to systematically evaluate LLM behavior in controlled financial settings, focusing on both decision rationality and financial
statement understanding.

Key Objectives
--Design financial reasoning tasks (e.g., portfolio choices, trading decisions, or strategic interactions) and evaluate LLM consistency and rationality.
--Analyze whether LLMs exhibit stable, economically coherent behavior across different prompts and scenarios.
--Test LLMs? ability to parse and reason over financial statements combining tables and text.
--Assess performance on discrete tasks such as ratio calculation, cross-table checks, and trend comparison.

Possible Tasks
Large Language Models for Financial Reasoning
158 SPMS CoS Patrick Pun Chi Seng --Prompt engineering and benchmarking across multiple LLMs.
and Information Extraction
--Construction of evaluation metrics for reasoning accuracy and robustness.
--Error analysis to identify systematic weaknesses or biases.

Expected Outcomes
--A structured evaluation framework for LLMs in financial contexts.
--Empirical findings on LLM strengths and limitations in finance-related reasoning.
--A final report with experimental results and insights.

Prerequisites
--Basic Python programming
--Familiarity with machine learning or NLP is helpful but not required
--Interest in finance and AI evaluation

Online education has rapidly become a major part of how people learn new skills, earn qualifications, and prepare for jobs?especially since the COVID-19 pandemic. From online degrees and
professional certificates to short courses and AI-powered learning tools, digital education is now closely connected to the world of work. However, many important questions remain unanswered:
Does online education really improve career outcomes? Do employers value online credentials in the same way as traditional, in-person education? And how are new technologies such as Generative AI
changing learning and skill development?

This project aims to answer these questions by studying the real labour-market impact of online education. We will use a large-scale, real-world workforce dataset, which contains detailed information
Learning Online, Working Ahead: How Digital
159 ITOM Div CoB (NBS) Jiayu Yao on individuals? careers, employers, job postings, and skills across countries and industries.
Education Shapes Future of Work
Using modern data analytics and econometric methods, the project will compare online education and traditional face-to-face education to understand how different learning pathways affect career
progression, job mobility, and skill matching in the labour market. The project will also explore how AI-related skills and online learning platforms influence hiring and promotion decisions.

This project is ideal for students interested in data analytics, economics, information systems, education technology, and the future of work. Participants will gain hands-on experience with large
datasets, causal analysis, and applied research.

There is huge effort continuously to learn from nature. Mimosa pudica is a very common plant found in many Southeast Asia countries, including Singapore, and is best known for its responsive
Learning from Nature: Experimental analysis of
behaviour towards mechanical stimulation by touching its leaves. Therefore, the aim of this project is to grasp a better understanding of this unique plant by studying its response to mechanical [Link]
160 MAE CoE Li Hua mechanical stimulation to the plant (Mimosa
stimulation and its responsive behaviour via experimental work and data analysis. The project will include the following areas: (1) literature survey; (2) data analysis of experimental work done by 662
pudica) movement
others previously; (3) supplement experiment, if necessary.
Large AI models trained on vast amounts of data have shown promise of adaptability to small specialized tasks. New types of 'small' data generated from bioengineering research and emerging
Leveraging foundation AI models for interpreting technologies may also be interpretable by such foundation models, which could accelerate their pathway towards commercialization and adoption. The student will study modern fine tuning AI
161 CCEB CoS Liang Kaicheng
small data from new technologies techniques on state of the art foundation models, and explore their relevance and translatability to new biomedical datasets. Note: medium-high proficiency and confidence in Python coding is
expected, and familiarity with machine learning fundamentals is preferred.

Vertically Aligned Carbon Nanotubes (VACNTs) demonstrate exceptional electrical and thermal properties, positioning them as promising candidates for electrical interconnects and thermal interface
materials in next-generation integrated circuit packaging. However, the high growth temperature of VACNTs (>700?C) presents a critical barrier to their seamless integration with temperature-sensitive
substrates, specifically the flexible substrates essential for wearable electronics. Conventional thermocompression bonding (TCB) processes typically require high processing temperatures (~300?C) to
transfer VACNTs from growth substrates to rigid target substrates, rendering them unsuitable for flexible materials due to the risk of substrate degradation.
To address this fundamental challenge, this research proposes a novel low-temperature TCB methodology to transfer VACNT arrays from growth substrates to flexible polymer substrates. The central
Low-Temperature Thermocompression Bonding
hypothesis posits that gold (Au) nanoparticles, formed through the annealing of an ultrathin Au layer, can facilitate robust, atomic-level diffusion bonding at temperatures strictly below 150?C. This
162 PO PO Tay Beng Kang Transfer of Vertically Aligned Carbon Nanotubes
innovative approach effectively overcomes the thermal incompatibility between VACNT synthesis, the conventional TCB transfer process, and temperature-sensitive flexible substrates. The project
onto Temperature-Sensitive Substrates
methodology employs a multi-technique strategy, integrating traditional Au-Au TCB with nanoparticle-assisted bonding and specific surface treatments to minimize thermal requirements.
The student will get hands-on experience with VACNT growth and transfer in this project. Through an orthogonal experimental design, this study will optimize key parameters?including bonding
temperature, pressure, time, and metal layer thickness. The specific aims are to achieve a high-strength mechanical bond capable of withstanding repeated bending, comprehensively characterize the
electrical properties of the bonded interface, and validate the technology by fabricating a VACNT-based flexible device.

Recently, Ng et al. (2020) proposed a comprehensive framework to redefine world class education/successful education and future-ready learner outcomes. The framework consists of four
dimensions. The first dimension is the three specific purposes of education articulated as development of learners for: learning, living, and lifework (Ng & Wong, 2020). The second dimension is the
importance of time continuum emphasizing that future readiness involves understanding history, meeting current knowledge standards, and preparing for future needs. The third focuses on the
contextual factors, including economic, social, and environmental considerations, which provide a backdrop for defining and understanding the specifics for the educational purposes described in the
MULTI-DIMENSIONAL FRAMEWORK FOR
first two dimensions. The fourth dimension, considered the most critical, involves practical applications such as leadership and management practices, teaching and learning methodologies, community
163 PCL PCL David Ng Foo Seong DEFINING FUTURE-READY LEARNING
engagement, and research and inquiry activities. This multi-dimensional framework aims to ensure that Singaporean students are holistically prepared for future challenges.
OUTCOMES
Ng et al. (2020) redefined the word class education, stating that ?A successful education system is able to develop future-ready individuals who will continue to learn beyond graduation, take on future
lifework, and thrive in a changing society and environment.? The definition will need supporting evidence from each of the four dimensions. The project?s operational aim is to define future-ready
outcomes based on alignment and integration of the four dimensions.

In this project, students will use LLM techniques to solve database system problems, such as index selection, data integration, generating data dictionary, NLP2SQL, etc.

Requirements: This is a research project, and suitable for students who are interested in doing research, and are willing to take up challenges. You will need to read research papers and develop new
solution. It is necessary to be familiar with a programming language. You are also expected to commit time for the project.
164 CCDS CCDS Cong Gao Machine Learning and LLM for Databases [Link]
?It is necessary to be familiar with a programming language.
?Some background on database and machine learning will be helpful
 This research focuses on leveraging Machine Learning (ML) and Artificial Intelligence (AI) to accelerate materials discovery and optimization for next-generation energy and power applications,
particularly in batteries and hydrogen technologies. Traditional trial-and-error approaches in materials development are often slow, resource-intensive, and limited in predictive accuracy. By integrating
AI-driven models with high-throughput simulations and experimental datasets, this work aims to:

1. Predict material properties and performance with high precision, reducing the need for extensive laboratory testing.

2. Identify novel electrode, electrolyte, and catalyst materials for advanced batteries and hydrogen production/storage systems.
Machine Learning/Artificial Intelligence in
3. Uncover hidden correlations and design principles that govern electrochemical stability, conductivity, and durability.
Materials Discovery for Energy and Power
165 NSSE NSSE Edison Ang Huixiang [Link]
Applications in Batteries and Hydrogen
4. Accelerate the design cycle through autonomous optimization frameworks that combine data-driven insights with physical and chemical principles.
Technology.
Candidates are expected to have prior experience in applying ML or AI methods to materials development, with expertise in handling large datasets, developing predictive models, and linking
computational insights to experimental validation.

Ultimately, this approach will pave the way for sustainable, high-performance energy storage and conversion technologies, enabling breakthroughs in renewable energy integration, green hydrogen
production, and large-scale electrification.

This project not only advances next-generation energy technologies but also prepares students for a competitive career at the forefront of AI-driven materials innovation.
This project explores the transformative potential of artificial intelligence in predicting and optimizing steel properties during production. Steel's properties?such as strength, ductility, and
toughness?are influenced by numerous factors, including composition, temperature, and processing techniques. Traditional methods of predicting these properties can be time-consuming and
complex.
Yi-Sheng (Eason)
166 MSE CoE Machine learning in materials industry In this project, you will apply machine learning algorithms to vast datasets from steel production, identifying patterns and correlations that influence final material properties. By accounting for key [Link]
Chen
parameters like chemical composition and processing conditions, you will participate in the development of predictive models that can accurately forecast steel properties, leading to more efficient and
cost-effective production processes.

This interdisciplinary project offers a unique opportunity to blend materials science with data science, equipping you with cutting-edge skills applicable to the future of industrial innovation.

Convolutional neural networks (CNNs) are often referred to as "black-box" models because their inner workings can be difficult to interpret and understand. This lack of interpretability can be a
significant limitation, especially in critical applications where explanations of decisions are essential, such as medical diagnoses or legal contexts. Making CNNs more interpretable is an ongoing
167 EEE CoE Wang Lipo Making Artificial Intelligence More Interpretable research area, and several techniques have been proposed to address this challenge. The student should (1) carry out a literature survey on existing work on the topic, (2) either download an existing [Link]
program or code a new program for an existing approach, and (3) attempt to improve the results by modifying the existing algorithm and/or combining existing algorithms.

This research project involves mapping the evolution of the historical drug trade in China by linking historical GIS to ancient texts, and thence using existing online drug name synonymy tools, to modern
Michael Stanley- Mapping the History of Chinese Materia Medica
168 SoH HASS biodiversity and biochemistry data. This will link the geography and names from the historical tradition to the modern marketplace and future medical agriculture, as well as defend against potential [Link]
Baker to Modern Ecologies
harm to the world drug supply from climate change. Project foundation is well-established, as can be seen in the associated link.

Marine cyanobacteria are prolific producers of chemically diverse secondary metabolites, many of which exhibit unique biological activities with potential biomedical and ecological applications. This
project focuses on the discovery of anti-quorum sensing (QS) molecules from filamentous marine cyanobacteria collected in Singapore?s coastal waters. Quorum sensing is a bacterial communication
system that regulates biofilm formation, virulence, and resistance mechanisms; therefore, disrupting QS pathways represents a promising strategy to combat pathogenic bacteria without applying
selective pressure associated with conventional antibiotics. The project will involve several stages. First, students will participate in the collection of filamentous cyanobacteria from intertidal areas in
Marine Cyanobacteria: Potential Source of Singapore. Crude extracts will be prepared and fractionated to isolate metabolites. The chemical characterization of these specialized molecules will be performed using modern spectroscopic and
169 NSSE NSSE Tan Lik Tong Bioactive Specialized Molecules as Anti-infective spectrometric methods, including nuclear magnetic resonance (NMR) spectroscopy and high-resolution mass spectrometry (HR-MS), to determine their molecular structures and identify novel
Drugs scaffolds. In addition, the bioactivity of isolated compounds will be evaluated using a biosensor strain of Pseudomonas aeruginosa (lasB-gfp), which fluoresces in response to quorum sensing activation.
By monitoring reductions in GFP expression, students will assess the ability of cyanobacterial metabolites to inhibit QS signaling pathways. This approach provides both a rapid screening platform and
quantitative insights into the anti-QS potential of the compounds. Through this project, students will gain hands-on training in natural products chemistry, microbiological assays, and bioanalytical
techniques, while contributing to a timely research area that intersects marine biodiversity, chemical ecology, and antimicrobial discovery. The expected outcomes include the identification of novel
anti-QS molecules and insights into their structural diversity, laying the groundwork for future translational research in anti-infective drug development.

My area of research focus is philosophy of mind, and I?m particularly interested in supporting projects from students on topics in the philosophy of memory.
Memory is a fundamental cognitive faculty. Without it, a creature would be unable to learn from its past and unable to reliably anticipate its future. The pertinence of memory for big questions
concerning personal identity, knowledge, or moral responsibility indicate its importance. It is crucial for us to have a full understanding of memory, its role, and its relation to faculties such as
imagination, an understanding grounded in psychology but which is also philosophically sensitive. And yet it is only in the last decade or so that memory has become a sustained focus for analytic
philosophy in the way that perception or language have long been. So the philosophy of memory is quickly becoming an active and exciting area of inquiry.
In my own current work I am especially interested in (i) the relationship between memory and reference, and (ii) memory errors. Episodic memory is a system that enables us to consciously `relive?
specific events from our personal past. For example, one might remember making coffee this morning and sensorily recall what it was like to hear the kettle reach a boil or to smell the coffee grounds.
Success in such cases will require two things. First, something must `fix? or determine that one?s memory is about that particular event, rather than, say, a similar event the previous day. Second,
Memory successes and memory errors: A new
170 PL HASS James Openshaw one?s memory must be suitably accurate. Though these observations are simple, what we might call the reference-fixing and accuracy conditions of remembering remain obscure. Existing theories face [Link]
philosophical framework
major challenges. The thriving work on memory in philosophy and the sciences requires new approaches to these issues. In part, this is because memory systems, like all systems, can break down or
malfunction, and do so in revealing ways.
Confabulation is a symptom of various neuropathologies in which the patient reports having a memory of an event which did not occur (or occurred at a different stage in their life). A full picture of
memory requires us to understand memory systems not only when they properly function but also when they fail. And, while philosophers have largely assumed that confabulations are defective,
there are reasons to think the story is more complicated, and that confabulations can, along some dimensions, be successful or adaptive. For instance, they may refer to genuine events or places in the
patient?s past, although the events or places are mischaracterised. If so, we will need either a new theory of reference or a new theory of confabulation, for existing theories predict that the two
phenomena should be incompatible.
I would be very happy to work with students interested in philosophical questions concerning memory, reference, imagination, confabulation, and related phenomena.
This research aims to provide a comprehensive meta-analysis of vulnerabilities and attack vectors in machine learning (ML) models, with a particular emphasis on the role of explainable AI (XAI) in
enhancing or compromising security. As ML systems are increasingly deployed in critical applications, understanding their security risks is of paramount importance. While numerous studies have
explored specific attacks such as adversarial examples, data poisoning, and model inversion, there remains a need for a unified analysis that identifies patterns across different attack types and defense
mechanisms. Additionally, the impact of model explainability on security remains a contentious topic; some argue it could expose models to new risks, while others believe it enhances transparency
and helps detect attacks.
Research Objectives:
Meta-Analysis of Vulnerabilities, Attack Vectors, [Link]-Analysis of Vulnerabilities and Attack Vectors: This will involve analyzing trends, identifying commonalities and differences across attack types, and evaluating the effectiveness of various
171 EEE CoE Chang Chip Hong and the Role of Explainability in Machine defense mechanisms.
Learning Security [Link] of Explainability on Security: To investigate how explainability affects ML security by surveying the literature and conducting experiments. This aspect will explore whether explainable AI
methods introduce new vulnerabilities or help in defending against attacks.
[Link] Analysis of Defense Mechanisms and Privacy-Preserving Techniques: To provide a supplementary review of defense strategies against adversarial attacks and privacy-preserving
methods, highlighting their effectiveness and limitations.
Expected Contribution:
This study will provide a holistic view of the current state of ML security, revealing under-explored vulnerabilities and proposing a roadmap for future research that balances the trade-offs between
explainability and security.
 This project delves into the innovative application of meta-learning in few-shot object detection. Traditional object detection algorithms typically require large annotated datasets for training, limiting
their applicability in scenarios where labeled data is scarce or costly to acquire. Meta-learning offers a promising solution by enabling models to generalize from a few examples and adapt quickly to
new object classes with minimal training data. In this project, we aim to investigate the effectiveness of meta-learning techniques in enhancing the performance of few-shot object detection systems.
172 EEE CoE Wang Lipo Meta-Learning for Few-Shot Object Detection [Link]
The student will explore one or two meta-learning approaches tailored to the unique challenges of few-shot learning in object detection. By developing and evaluating meta-learning-based object
detection models, this project seeks to provide insights into the practical benefits and limitations of applying meta-learning in this domain.

Metal complexes supported by unusual boron Complexes consist of metal and ligands, and the nature of the metal complexes highly relies on the property of the ligands. Recently, we have developed an entirely new ligand system based on boron
173 CCEB CoS Rei Kinjo [Link]
cluster ligands cluster. By incorporating metal onto the ligands, in this project, we will develop a series of unprecedented metal complexes and investigate their catalytic activity.

Vibrio cholerae has a complex life history where it can move from marine to fresh water sources, and opportunistically infect humans. Much of its life is spent associated to particles in marine
environments and it is known to attach to chitin, a common polymer in marine invertebrates. V. cholerae is found associated with microalgae and this association is poorly understood. We hypothesis
Rebecca Josephine Microalgae as a marine resovoir for Vibrio
174 SBS CoS that V. cholerae is associated with a specific group of microalgae, diatoms, in which some species are known to have chitin in their outer surface. This project aims test if V. cholerae can specifically [Link]
Case cholerae*
associate with microalgae with chitin compared to those without chitin. This will be done in our co-culturing system we have developed. In addition, you will have the opportunity to explore mutants
deficient in attachment and chemotaxis to chitin and use microscopy to identify changes in host colonisation associated with localisation to chitin.

Precise ion separation underpins many energy and environmental technologies, including water treatment, resource recovery and electrochemical systems. This project introduces students to
microporous polymer membranes for selective ion transport. Students will synthesise representative hydrophilic polymers of intrinsic microporosity and fabricate self-standing and thin-film composite
Microporous Polymer Membranes for Precise Ion
175 CEE CoE Ye Chunchun membranes. Polymer structure will be characterised using NMR, TGA and gas sorption, while membrane performance will be assessed through swelling measurements and ion transport tests with [Link]
Separation
representative salt solutions under controlled conditions. The results will be used to establish clear structure?property relationships linking polymer chemistry, membrane structure and ion selectivity.
The project is highly guided and achievable within two months, providing hands-on training in polymer synthesis, membrane fabrication and sustainable ion separation technologies.

This project focuses on designing and optimizing the microstructures of metallic alloys to enhance their performance and sustainability. By understanding and manipulating the microstructure at the
microscopic level, we can develop materials with improved properties such as strength, durability, and corrosion resistance, which are crucial for sustainable engineering applications.

Yi-Sheng (Eason) Microstructural engineering for sustainable As a participant in this project, you will have the opportunity to use advanced microscopy techniques, including scanning electron microscopy (SEM), to characterize and analyze the microstructures of
176 MSE CoE [Link]
Chen metallic materials various alloys. You?ll explore how different processing techniques influence these structures and, consequently, the material's overall performance.

This hands-on experience in cutting-edge materials science will give you valuable insights into how metals behave and how we can engineer them for future applications, contributing to a more
sustainable and innovative world.

This project focuses on advancing mid-infrared (MIR) silicon photonics by developing heterogeneously integrated semiconductor lasers directly on silicon chips. The mid-infrared spectral region (2?10
?m) is crucial for chemical sensing, environmental monitoring, and free-space communications, but traditional MIR systems rely on bulky, power-hungry components. By integrating efficient III?V
177 CoE CoE Wang Qijie Mid-infrared silicon photonics semiconductor gain materials onto a silicon photonics platform, this project aims to realize compact, low-cost, and electrically pumped MIR laser sources compatible with large-scale CMOS fabrication. [Link]
The student will design, simulate, and characterize hybrid laser structures that combine silicon waveguides for passive light routing with active III?V regions for optical gain and emission. This work will
contribute to building a new generation of on-chip MIR photonic systems, enabling applications such as lab-on-a-chip spectroscopy, gas detection, and integrated optical communications.
Lightweight cellular structures are widely used for impact protection, but conventional honeycombs often face a tradeoff between mass efficiency and stable crushing. Our previous study introduces a
staggered + rotational/lamellar honeycomb (SRH) fabricated in 316L by metal AM, and shows substantially higher specific energy absorption (SEA) in the out-of-plane orientation compared with in-
plane, under both quasi-static and impact conditions.
Modeling Studies for Enhanced Energy
This proposal focuses on modelling studies (validated FE + parametric design exploration) to (i) explain the mechanisms behind SRH performance and (ii) guide next-iteration designs for even better
178 MAE CoE Li Hua Absorption in 3D-Printed Staggered-Rotational
crashworthiness:
Honeycomb (SRH)
O1. Build a validated, reusable FE modelling pipeline for quasi-static and impact crushing of SRH (in-plane and out-of-plane).
O2. Identify deformation/energy-absorption mechanisms (e.g., progressive buckling, localization, layer-wise collapse) and connect them to geometry/orientation.
O3. Run a parametric design study to optimize SEA while maintaining stable crushing and manufacturability for AM.
O4. Quantify sensitivity to boundary conditions, friction, mesh, material model, and printing-induced anisotropy/defects.
The mechanical and structural properties of double helical DNA have been successfully described by models with varying levels of complexity and coarse-graining schemes. In this project, we will work
on a discrete elastic model of the DNA to understand the interplay between base pair interactions and the bending and twisting of the DNA. We will perform Monte Carlo simulations to understand
179 PAP Div CoS Yong Ee Hou Monte Carlo simulations of DNA [Link]
equilibrium structure and shape.

Large language models (LLMs) like ChatGPT have shown remarkable performance for tasks based on human language. However, relatively few work has been done on using these powerful AI models
for the development of medicine. This research will focus on the development of an LLM-based model that represents a novel approach in the field of drug discovery. The implications of this research
Multi-Modal Large Language Model for Drug are profound, offering the potential to revolutionize the way drug formulations are discovered and optimized. By leveraging the power of deep learning to understand and model the multi-faceted
180 CCDS CCDS Chan Guo Wei Alvin [Link]
Development nature of drug formulations, we can accelerate the pace of discovery and enhance the precision of drug development processes. This, in turn, could lead to the creation of more effective and targeted
therapies, ultimately translating to improved clinical outcomes.

This project aims to develop a state-of-the-art multimodal deep learning model that leverages multiple data modalities to enhance the diagnosis and treatment of cardiovascular diseases. The model
Multimodal Deep Learning Model for Medical
181 LKCMedicine LKCMedicine Yeo Si Yong will include data from different scan modality, along with medical report to create a comprehensive, patient-centered analysis tool. The incorporation of the diverse data can provide a more detailed
Dataset.
and accurate assessment of health, supporting clinicians in making more informed decisions.
Supervisors: Atul Parikh and Peter Torok

At the Institute for Digital Analytics and Science (IDMxS), we are developing accessible diagnostic tools aimed to detect small molecules and pathogens. Our approach is to develop and test simple
microscopes based on darkfield microscopy and interferometric scattering microscopy. Leveraging nanoparticle-based assays, wherein DNA probe molecules are conjugated onto various types of
nanoparticles, we will trigger particle clustering in the presence of target molecules. By changing their material composition, shape, and size, we can mix distinct nanoparticles in solution and then
distinguish them by type using image analysis. By programming these different nanoparticle types to detect separate target pathogens, e.g. covid and flu, the aim is to accurately detect the presence of
Multiplexed nanoparticle imaging for disease multiple targets and estimate their concentration by capturing a single image.
182 SPMS CoS Peter Torok
diagnostics The student?s role involves contributing to the development of conjugation protocols for DNA and various types of nanoparticles, using standard wet chemistry techniques. Once these protocols are
established, images will be captured of nanoparticle samples?both with and without target molecules?at different concentrations using a home-built optical microscope setup. The student will run the
acquisition of thousands of images crucial for analysis purposes, primarily using the Python programming language. By applying simple machine learning algorithms, quantitative methods to
differentiate target molecule concentrations will be developed.
The project would expose the student to (1) interdisciplinary research combining tools and techniques from across physical, biological, and data sciences and engineering; (2) collaborative environment,
involving working in small and large teams; and (3) a societally-led problem of global significance and import.
By achieving these objectives, we aim to develop a cost-effective multiplexed diagnostic tool capable of paving the way for at-home disease diagnosis.
In complex surgeries such as the brain, mapping of the tumour field is critical, especially in real-time while the resection is ongoing. However, mapping is usually done based on pre-operative scans, and
Multiview endoscopic microscopy for tumour
183 CCEB CoS Liang Kaicheng there are very limited options for real-time imaging. Endoscopic microscopy is a promising approach to obtain live multi-perspective views. The student will work on optical and mechanical designs
resection guidance
involving fibre optics and miniature lenses/mirrors for multi-view endoscopic probes.
The research is about fabricating nanoporous carbon fiber materials and study its function in capacitive energy storage. The effect of pore size and shape will be compared to elucidate the storage
184 PO PO Fan Hongjin Nanoporous carbon materials for energy storage [Link]
mechanism. The research activity includes material synthesis and basic electrochemistry test.
 Linear temporal logic (LTL, [1, Ch. 5]) and, more generally, ?-regular objectives are alternatives to the traditional discount sum and average reward objectives in reinforcement learning (RL, [2]), offering
the advantage of greater comprehensibility and hence explainability [3]. In prior work, we study the relationship between these objectives. Our main result is that each RL problem for ?-regular
objectives can be reduced to a limit-average reward problem in an optimality-preserving fashion, via (finite-memory) reward machines. Furthermore, we demonstrate the efficacy of this approach by
showing that optimal policies for limit-average problems can be found asymptotically by solving a sequence of discount-sum problems approximately. Consequently, our prior work resolve an open
problem of Alur et al. [3]: optimal policies for LTL and ?-regular objectives can be learned asymptotically.

The purpose of the proposed Global Connect Fellowship project is to conduct an empirical evaluation and to develop more performative learning algorithms which still come with correctness
guarantees. A first milestone would be the implementation of our prior work and to demonstrate that competitors (such as [4]) may fail to find optimal policies. Thereafter, there are various avenues
which can be explored:
? Guide learning by suitably designed reward shaping [5].
? Develop more practical algorithms for limit-average rewards.
? Identify fragments of LTL or assumptions for which more refined approaches are applicable without sacrificing optimality.
New Omega-Regular Objectives for ? Study extensions of the RL framework such as partially-observable MDPs [6] or multi-task environments [7].
185 PO PO Ong Chih-Hao Luke
Reinforcement Learning
The ideal candidate has a strong background in the theory and practice of reinforcement learning, as well as the foundations of automata theory.

References:
[1] Christel Baier and Joost-Pieter Katoen. Principles of Model Checking. 2008.
[2] Richard S. Sutton and Andrew G. Barto. Reinforcement Learning: An Introduction. 2018.
[3] Rajeev Alur, Suguman Bansal, Osbert Bastani, and Kishor Jothimurugan. A framework for transforming specifications in reinforcement learning. 2022
[4] Cameron Voloshin, Abhinav Verma, Yisong Yue: Eventual Discounting Temporal Logic Counterfactual Experience Replay. ICML 2023.
[5] Andrew Y. Ng, Daishi Harada, Stuart Russell: Policy Invariance Under Reward Transformations: Theory and Application to Reward Shaping. ICML 1999
[6] M.T.J. Spaan. Partially Observable Markov Decision Processes. In: Wiering, M., van Otterlo, M. (eds) Reinforcement Learning. Adaptation, Learning, and Optimization. 2012
[7] Rich Caruana. Multitask learning. Machine Learning. 1997.

Synthetic organic chemistry has the ability to enhance the value of simple materials. Through the development of precise chemical reactions, commodity chemicals can be transformed to medicines,
agrochemicals, and other substances which improve human life. This is the power of organic synthesis.

New chemical rearrangements with applications The building blocks of organic synthesis are predominantly derived from the petrochemical industry, originating from benzene and other hydrocarbons. While the transformation of fossil fuels to a life-
186 CCEB CoS Phillip Stephen Grant [Link]
in drug discovery* changing medicine is an incredible achievement, dependence on non-renewable resources is a major issue. The challenge facing organic synthesis today is the sustainable production of valuable
chemicals.

This project involves the development of new chemical rearrangements which aim to improve the efficiency and sustainability of organic synthesis. Come to NTU and get involved!

here is a continuous drive to enhance the operation frequency and power handling of electronics. Numerous applications, ranging from telecommunications and medical diagnosis to aerospace and
defense rely on advanced electronic technologies. The overall performance of electronic chips is ultimately limited by the characteristics of its device building blocks, and as a result, development of
new electronic devices can provide upstream solutions to boost the performance of electronics systems. The main aim of this research proposal is to demonstrate a new type of electronic devices
(namely mesoscale devices) based on Micro-Electro-Mechanical System (MEMS) technology, which can significantly extend the state-of-the-art in terms of power capability and operation frequency.

The specific aims are as follows:

New high-perfromance radiofrequency switches
Mohammad 1) Design and Multiphysics Finite Element Method (FEM) simulation of mesoscale microwave, mm-wave, and terahertz switches.
187 EEE CoE based on micro-electromechanical systems [Link]
Samizadeh Nikoo
(MEMS)
2) Optimization of the parameters based on the simulations and to design the device layout.

3) Fabrication of mesoscale MEMS switches and experimental characterization based on time-domain and frequency-domain techniques.

4) Showcasing the technological potential by integration of mesoscale MEMS switches on a semiconductor platform.

Our new technology has great potentials to be used in satellite electronics, RF front-end of advanced telecommunication chips and radars.
A complex network is a system made up of interconnected nodes (entities) and edges (relationships) that exhibit non-trivial patterns of connection, often displaying features not found in simple
networks like lattices or random graphs, and are used to model real-world systems like social networks, biological pathways, or the internet, where interactions between components are intricate and
188 PAP Div CoS Yong Ee Hou Nonlinear dynamics of Complex Network [Link]
not easily predictable. In this project, we will study how to model nonlinear dynamics on different complex networks (e.g., random, scale free, etc) and see how they can be used to model social
systems.

Two-dimensional (2D) quantum materials, including transition metal dichalcogenides and van der Waals ferroelectrics, exhibit strong and tunable nonlinear optical (NLO) effects such as second- and
third-harmonic generation. These responses arise from their unique symmetry, excitonic resonances, and quantum confinement, enabling light?matter interactions far stronger than in bulk crystals.

Nonlinear optic responses in 2D quantum This project investigates the origin, tunability, and measurement of nonlinear optical responses in 2D materials. Using femtosecond laser spectroscopy and polarization-resolved harmonic generation
189 EEE CoE Chae Sanghoon [Link]
materials and their measurement measurements, it will quantify the nonlinear susceptibilities (??, ??) and explore how factors such as thickness, stacking, and external fields modify NLO efficiency.

The outcomes will provide physical insight into 2D quantum nonlinearities and guide the design of ultracompact on-chip nonlinear photonic devices, including wavelength converters, modulators, and
quantum light sources.

In this project, the student will investigate how to selectively enhance light emission and nonlinear optical effects such as second-harmonic generation in 2D material systems. The work combines
optical spectroscopy and simulation to understand how light?matter interactions can be controlled for brighter, more efficient near-infrared photonic devices, with potential applications in
communications, imaging, and quantum technologies.

190 CoE CoE Wang Qijie Nonlinear optics in 2D materials This project explores how to make 2D material, a promising near-infrared semiconductor, shine much brighter by coupling it with smart light-trapping structures called photonic cavities. The students [Link]
will fabricate or simulate an etch-free photonic cavity on a selected 2D material such as InSe layer, study how cavity resonance affects light emission and nonlinear processes, understand how excitons
interact with cavity modes to boost optical efficiency, and demonstrate a scalable, low-cost approach for improving near-infrared photonic devices. The student will learn optical spectroscopy
(photoluminescence, second-harmonic generation measurements, Finite-difference time-domain (FDTD) or other photonic simulations, and nanofabrications.

This project will look into discovery, synthesis and characterization of new materials for solar cell applications.
191 PO PO Lydia Helena Wong Novel materials for solar cell application [Link]
Applicants should have background in Materials, Chemistry, Physics or Engineering
 In this project we will investigate online learning augmented algorithms for sequential financial decision-making, designed to leverage black-box predictions arriving over time. Financial tasks like
dynamic portfolio management or algorithmic trading inherently operate in an online setting, requiring decisions based on evolving data streams. We propose algorithms where a machine learning
component learns online (by adapting based on performance observed thus far) how to best interpret and integrate potentially imperfect black-box predictive inputs (e.g., market forecasts) revealed at
each step. The core theoretical pursuit is designing frameworks where online algorithms can provably leverage predictive advice to improve performance (consistency) while maintaining guarantees
even when the advice is arbitrarily poor (robustness). In particular, this project will explore new learning-augmented algorithms for online trading under realistic models of price behavior. Drawing
inspiration from recent work on "trading prophets" and learning-augmented decision-making, the project will study how predictive signals can be utilized to improve robustness and performance. The
Themistoklis Online learning augmented algorithms utilizing
192 CCDS CCDS focus will be on developing simple theoretical models that move beyond idealized assumptions and testing preliminary implementations on real financial data.
Gouleakis financial predictions
[1]: Correa, Jose, Andres Cristi, Paul Dutting, MohammadTaghi Hajiaghayi, Jan Olkowski, and Kevin Schewior. "Trading prophets." In Proceedings of the 24th ACM Conference on Economics and
Computation, pp. 490-510. 2023.

[2]: Rajput, Surbhi, Ashish Chiplunkar, and Rohit Vaish. "Trading Prophets: How to Trade Multiple Stocks Optimally." In 2025 Symposium on Simplicity in Algorithms (SOSA), pp. 238-252. Society for
Industrial and Applied Mathematics, 2025.

Olfaction, the sense of smell, is critical for the survival of many animal species. It is a key physiological process that allows for the detection and discrimination of volatile airborne molecules. Insects
use the sense of smell to find food, avoid predators and noxious agents, to find appropriate mating partners, and to find hosts for completing life cycles. There is continued global health interest in
finding effective methods to control insects that harbour infectious agents including mosquitoes, ticks, and flies. The olfactory system has evolved over 500 million years, resulting in evolutionarily
distinct chemosensory olfactory receptors (ORs) in insects compared to vertebrates. These receptors sense odorants or pheromones as a chemical signal and convert it into an electrical signal. The
Optimize the expression, purification and
insect olfactory system is comprised of a diverse odorant specific OR subunit and a highly conserved olfactory co-receptor (Orco) subunit in an unknown stoichiometry. The Orco subunit is conserved
193 SBS CoS Sandip Basak reconstitution of Olfactory receptor (Orco) from
across insect species and has been implicated in trafficking and/or membrane localization of conventional OR subunits. Interestingly, when heterologously expressed, Orco alone can form a cyclic
Drosophila in lipidic nanodisc*
nucleotide activated ion channel without the odorant selective OR. The initial focus of this work will involve a structural and functional description of an olfactory receptor complex from the model
organism Drosophila melanogaster. The D. melanogaster olfactory system is simpler and the sequence similarity between the D. melanogaster Orco and the mosquito Orco is greater than 70% implying
similar architectures. The main goal of this project is to optimize the expression of the Drosophila Orco receptor. The expressed protein will then be used for purification to get a stable homogeneous
sample. Beside that the aim will be to reconstitute Orco protein in native-like lipidic nanodisc to get information about the conformational changes during the receptor activation.

The maritime industry is under increasing pressure to reduce its carbon footprint and improve energy efficiency in response to environmental regulations and sustainability goals. This project aims to
explore innovative solutions for energy management and optimization on vessels by investigating alternative energy sources such as LNG (Liquefied Natural Gas), hydrogen, biofuels, and ammonia. The
research will focus on two key aspects:
1. Life Cycle Assessment (LCA): A comprehensive environmental impact assessment of alternative energy sources, from resource extraction and production to operational use and disposal. This analysis
Optimizing Energy Efficiency in Maritime
will help identify the most sustainable options based on greenhouse gas emissions, energy consumption, and waste generation throughout the energy's life cycle.
Operations: Life Cycle Assessment and Techno-
194 CEE CoE Yan Ran 2. Techno-Economic Analysis (TEA): A detailed evaluation of the economic feasibility of alternative fuels, considering factors such as capital costs, fuel availability, infrastructure requirements, and
Economic Analysis of Alternative Fuels for
operational costs. The study will assess the trade-offs between economic performance and environmental impact to determine viable solutions for commercial implementation.
Sustainable Shipping
The project will involve data collection from industry reports, scientific literature, and case studies of vessels that have implemented alternative energy solutions. Students will use software tools to
model energy systems and perform LCA and TEA analyses.
By the end of the project, participants will develop a framework for energy optimization strategies that balance environmental and economic considerations. The findings will contribute to the growing
body of research on sustainable shipping and provide actionable recommendations for improving energy management practices in the maritime sector.

This research proposal aims to explore and optimize the scheduling of serverless functions to enhance both performance and cost efficiency in cloud environments. By leveraging advanced techniques
Optimizing Serverless Function Scheduling for such as adaptive scaling, resource allocation algorithms, and workload prediction, the project will identify bottlenecks in the current scheduling processes. The goal is to develop innovative solutions
195 CCDS CCDS Dmitrii Ustiugov [Link]
Enhanced Performance and Cost Efficiency that reduce latency, improve throughput, and lower operational costs. The intern will investigate real-world serverless platforms and conduct experiments to benchmark the proposed optimizations,
contributing to more effective serverless function deployment.
Optoelectronic devices based on heterogenously- In this project, 2D layered materials will be heterogeneously integrated with other types of materials, such as oxides. Based on these structures, novel optoelectronic devices with superior functionality
196 PAP Div CoS Wang Xiao [Link]
integrated 2D materials will be developed.
Over the past decade, the escalating impacts of global warming have intensified the need for advanced thermal?management strategies and have catalyzed substantial research into mid-infrared (mid-
Organic conductive polymer based wearable IR) electrochromism as a promising and energy-efficient approach for dynamic thermoregulation. By actively tuning thermal emittance, users can personalize radiative heat transfer with minimal
197 MSE CoE Wu Ronghui electrochromic devices for human body power consumption, thereby mitigating the influence of environmental temperature fluctuations. In this project, we develop wearable electrochromic devices based on organic conductive polymers to
thermoregulation enable adaptive human-body thermoregulation. Through electrochemical modulation of the infrared emissivity of these polymers, the devices dynamically regulate heat exchange between the body
and its surroundings.

Rapid technological advancements, evolving market dynamics, and geopolitical shifts are converging to create a turbulent work environment, placing immense pressure on individuals and
organisations, particularly small medium enterprises (SMEs), to adapt and upskill. While workplace learning (WPL) offers a promising path forward, effective implementation is often hindered by
challenges in securing buy-in from key stakeholders, and the absence of robust WPL strategies can stifle growth and even jeopardize survival. This study will delve into how WPL can empower
individuals and organisations to not just survive but thrive amidst this constant change.
This research will commence by establishing a baseline understanding of the current WPL landscape in SMEs, identifying prevalent practices and perceived value. The research then will delve into the
Organisational Success through Workplace effectiveness of diverse workplace learning modalities, seeking to tailor approaches that address SMEs' unique needs taking in account of the role of emerging technologies such as Gen AI. This
198 LMO Div CoB (NBS) Yu Kang Yang Learning: Unveiling its Value and Dynamics in knowledge will empower policymakers to develop evidence-based recommendations for effective WPL implementation, while practitioners can leverage these findings to make informed decisions
SMEs about their learning and development modalities. A parallel investigation will also quantify the tangible and intangible benefits of WPL from the employer's perspective, providing concrete evidence of
its ROI. This data will highlight the measurable returns of WPL, empowering SME leaders to prioritise investments in workplace learning and development.
By adopting a mixed-methods approach that integrates quantitative and qualitative data, this research aims to generate a rich insight into the multifaceted role of WPL. In addition to new data
collection, this research will also leverage existing data collected by government agencies (e.g. NACE, SSG, WSG, MOM) to enrich our understanding of workplace learning practices and their impact.
The findings will inform evidence-based recommendations for SMEs, policymakers, and HR professionals, fostering a culture of continuous learning and development.

In 1952, Alan Turing published the reaction-diffusion (RD) mathematical framework, laying the foundations of morphogenesis as a self-organized process emerging from physicochemical first principles.
199 PAP Div CoS Yong Ee Hou Pattern Formation in Nature In this work, we will study how to use phenomenological reaction diffusion equations to model different kinds of patterns we see in nature. We aim to follow recent work by the Milinkovitch group [Link]
closely.
The performance-interpretability tradeoff (PIT) in data science and AI refers to the balance between how well a statistical/machine learning model performs, and how readily humans can comprehend
it. While `high stakes? contexts of application like finance and engineering focus on finding the optimal balance between performance and interpretability, nascent ones like discourse analysis provide
valuable opportunities to investigate the domain-specific nuances and implications of PIT. With the rise of data science and AI in the humanities, this project will investigate the nature and
characteristics of PIT in discourse research. The initial focus is on i) the (strategically) indeterminate nature of PIT, ii) PIT in time-based algorithms and discourses, and iii) conceptual relationship(s)
between PIT and `mixed-methods? discourse research, with subsequent developments based on emergent findings. Discourse datasets to be investigated include psychological counseling, social media,
Performance-interpretability tradeoff in
200 LMS HASS Dennis Zhiming Tay academic lectures, and other relevant time-based contexts, while time-based algorithms include LSTM, ARIMA time series, and survival regression models. Modeling and analytic procedures will
discourse research
replicate established procedures from previous relevant work, with implementation in the Python language. The results will bear important implications for discourse research methodology, and open
new pathways to theoretical PIT research in machine learning.

Feel free to contact me at [Link]@[Link] for more information!

The great importance of corrugated two-dimensional (2D) perovskites as the next-generation perovskite materials cannot be overemphasized. Currently, artificial intelligence (AI)-based materials
design and discovery is widely regarded as the fourth paradigm for material science. With the milestone achievement of Materials Genome Initiative (MGI), various databanks such as Materials
Project, JARVIS, NOMAD, Aflowlib, OQMD, etc, have been built up. Together with other well-established materials databanks, they provide a solid foundation for data-driven functional prediction and
new material design and discovery. Even with the progress, the incorporation of materials periodicity and multiphysical information into materials representation and featurization remains to a major
Periodic-geometry-based learning models for
201 MAS Div CoS Xia Kelin challenge.
corrugated 2D perovskite design and discovery
In this project, we propose periodic geometry (PD)-based learning models for corrugated 2D perovskite design and discovery. Periodic-geometry-based material descriptors that are aware of materials
periodicity information and multiphysical properties are developed. Further, we propose cellular complex neural network (CCNN) models for materials design and discovery. Mathematically, CCNNs are
a generalization of graph neural networks (GNNs) and can learn not only the multiphysical higher-order interactions, but also the periodicity information of the material structures.
 This project aims to investigate the role of phase separation in plant immune signaling during plant-microbe interactions, focusing on immune regulators localized at the plasma membrane. We
hypothesize that phase separation plays a critical role in the spatial and temporal regulation of immune responses. By employing interdisciplinary approaches such as advanced single-molecule imaging,
Phase Separation in Plant Immune Signaling
202 SBS CoS Miao Yansong protein structural analysis, biochemistry, biophysics, and plant pathology, we will explore how different immune regulators undergo phase separation on the plasma membrane and influence defense [Link]
During Plant-Microbe Interactions*
mechanisms. We will analyze the biophysical properties of these regulators, assess their structural dynamics, and observe their interactions with microbial factors, including effectors etc. This study will
provide insights into the molecular mechanisms underlying plant immunity and may identify new targets for enhancing disease resistance in crops.

This project investigates how intracellular signaling events, including protein modifications and membrane remodeling, influence the initiation and regulation of actin remodeling?encompassing
polymerization, stabilization, and depolymerization?to support cellular activities under physiological and pathological conditions. The phase separation of different actin-binding proteins is crucial in
coordinating these signaling events, thereby dynamically regulating actin cytoskeleton dynamics. To explore this, we will use molecular cloning to construct actin-binding proteins, express them, and
Phase Separation-mediated Actin Remodeling in understand how signaling events modulate these key regulators via phase separation for involvementmodeling. Protein biochemistry techniques will be used to analyze protein interactions and
203 SBS CoS Miao Yansong [Link]
Cell Signaling* modifications that mediate phase separation. Advanced imaging methods, such as live-cell fluorescence microscopy, will be utilized to visualize the spatiotemporal dynamics of actin and signaling
proteins in real-time. This study aims to provide a comprehensive understanding of the mechanisms underlying actin remodeling in response to rapid cellular signaling linked to cytoskeletal regulation.

This project aims to develop and apply a cutting-edge Physics-Informed Neural Networks (PINNs) framework for modeling complex coupled-field problems in heterogeneous poroelastic media, with a
Physics-Informed Neural Networks for
focus on CO2 sequestration in depleted hydrocarbon reservoirs. The study will explore the interaction between fluid flow and subsurface mechanics to predict caprock integrity and assess the risk of
204 ASE CoS Luca Dal Zilio Poroelastic Media: Enhancing CO2 Sequestration [Link]
induced seismicity. By employing a novel composite neural network architecture tailored to capture the discontinuities at material interfaces, this project will offer improved accuracy and reliability in
and Caprock Integrity*
simulations, advancing our understanding of CO2 storage safety and long-term containment strategies.

Chemical discovery is essential for addressing global challenges such as climate change and green energy, yet the sheer scale of the chemical space?estimated at $10^{60}$ bioavailable
molecules?renders traditional trial-and-error discovery prohibitively expensive and inefficient. While data-driven artificial intelligence offers a solution to bridge the gap between existing structures and
unexplored spaces, current models are significantly limited by their lack of physical grounding. Most existing architectures fail to incorporate the fundamental physics governing molecular behavior,
leading to representations that cannot accurately capture the relationship between structure and function.
Physics-informed sheaf neural network for
205 MAS Div CoS Xia Kelin
chemical data analysis
In this research project, we propose physics-informed sheaf neural networks (SNNs) for chemical data analysis, bridging the gap between algebraic topology and molecular dynamics. We will establish a
mathematical foundation of Normal Mode Analysis (NMA)-based sheaf model. By incorporating NMA eigenspectral information?which reflects the intrinsic dynamics and vibrational properties of a
molecule?directly into the SNN architecture, our model will embed physical constraints into the learning process. This approach will allow the SNN to learn chemical structures and functions more
accurately and efficiently, providing a superior framework for predicting molecular properties and accelerating the discovery of new chemical entities.

There is extensive research focussed on motion planning for robot manipulators. Part insertion is one such scenario that is common in both daily life and industrial settings. The contact-rich
manipulation tasks, for example, inserting USB port and socket, are complex and hence can lead to failures or damage to the manipulated object. In literature, these problems have been solved using
vision-based methods and learning-based methods. Both these methods have their own limitations. In the sole vision-based method, inaccuracy in determining the pose may lead to a mismatched
trajectory and hence a failure. In the learning-based method, extensive sampling is required to gather sufficient data. This is time-consuming and expensive. Training in a simulator and then testing in
real world doesn?t always lead to accurate results. There is extensive research in reducing this sim-to-real gap. Moreover, gathering a dataset covering all possible cases can be challenging.
Planning Robotic Contact-Rich Tasks via
206 MAE CoE Domenico Campolo [Link]
differentiable Physical Interaction models For more technical details, pls refer to:
Campolo, D., & Cardin, F. (2023). Quasi-Static Mechanical Manipulation as an Optimal Process. In 2023 62nd IEEE Conference on Decision and Control (CDC) (pp. 4753-4758). IEEE
[Link]

Plastic pollution is negatively impacting ecosystems, human and planetary health, livelihoods and economies worldwide. Small Island Developing States (SIDS) are disproportionally affected and highly
vulnerable while their contribution to plastic production and trade is minimal. This interdisciplinary research explores the intersection between plastic waste management, environmental policy and
Plastic waste management strategy in the Global
207 CEE CoE Fei Xunchang circular economy in Pacific SIDS (PSIDS). Often overlooked and small, data on imports, waste generation, composition and management are inadequate. To address this gap, this study contributes to
South
data transparency of the lifecycle of plastics by examining the current waste and policy landscape and assessing various ongoing initiatives. In this FYP project, you may conduct a case study on a PSIDS
to collect data on waste flows, analyse policy pathways or map out circular economy initiatives ? based on your interests and knowledge.

Modern optical communication relies on the ability to transmit information through complex environments. Recent research has highlighted the potential of Orbital Angular Momentum (OAM) states -
characterized by phase singularities - for high-capacity data transmission. These states are often utilized because they exhibit a greater degree of robustness compared to standard Gaussian beams
when encountering certain types of environmental perturbations. However, as light travels through dense scattering media like biological tissue or atmospheric turbulence, these structured phase
fronts can still face significant degradation that limits their effectiveness.

This project investigates the comparative performance of Polarization Singularities in these challenging environments. Unlike scalar OAM states, polarization-structured beams, such as radial,
azimuthal, and cylindrical vector vortex beams, possess spatially varying polarization states. This added complexity offers a unique degree of freedom that may offer distinct advantages or different
degradation profiles when compared to traditional OAM states. Whether these vector features provide superior robustness against scattering remains an open research question that this project aims
Polarization singularities in scattering media for
to explore.
208 EEE CoE Matthew Foreman biological imaging and free-space [Link]
communications
The work will begin with simulation and modelling, where you will utilize an in-house Python-based random matrix codebase to simulate the propagation of vector light through random scattering
media. Following this, you will perform a comparative analysis by generating various structured light beams, including both OAM and vector vortex states. The goal is to quantify how their spatial and
polarization properties evolve as they encounter increasing scattering density. Finally, you will establish performance metrics to evaluate the relative robustness of these states by analysing the
degradation of their singular points and the preservation of information-carrying features over distance.

This project is relevant to the development of next-generation in-vivo biological imaging and secure free-space communication, where maintaining beam structure through "cloudy" media is a primary
challenge. You will gain experience in vector electromagnetics, matrix-based simulation techniques, and advanced data analysis in a research-active environment.

Project Overview: Variational inference (VI) is central to modern Bayesian machine learning, providing scalable approximate inference in models that are analytically intractable. However, its
foundations rely fundamentally on probability theory?expectations, integrals, and additive measures of uncertainty?which may be limiting in scenarios where information is sparse, incomplete, or
inherently epistemic. This internship explores a new research direction: variational inference based on possibility theory, an alternative framework in which uncertainty is expressed through maximum-
based rather than additive operations. The goal is to develop principled approximation methodologies that extend VI beyond the probabilistic paradigm.
Possibilistic variational inference for Machine
209 SPMS CoS Jeremie Houssineau [Link]
Learning
Motivation: Many real-world problems present forms of uncertainty that are not naturally captured by probability distributions: i) sparse or weak observations, ii) severe model misspecification, and iii)
requirements for conservative or robustness-oriented inference. Possibility theory offers a mathematically coherent means of describing plausibility without invoking probabilistic assumptions, making
it relevant for models where epistemic uncertainty dominates. This perspective suggests the need for new optimisation principles
akin to VI, but not tied to integrals or classical divergences.
 Model-based reinforcement learning (MBRL) relies on learning a ?world model? (latent dynamics + reward) and using it for planning or policy improvement. In practice, most uncertainty-aware MBRL
methods conflate aleatoric uncertainty (stochasticity intrinsic to the environment) with epistemic uncertainty (lack of knowledge due to limited data), which can lead to suboptimal exploration: agents
may repeatedly probe inherently noisy transitions rather than targeting knowledge gaps.

This project investigates possibility theory as an alternative uncertainty calculus for world models, with the goal of making exploration depend solely on epistemic uncertainty. We will represent the
Possibilistic world models for epistemic-only
learned dynamics as a possibility function over next-state predictions (or latent transitions), where the spread of the possibility function reflects epistemic ambiguity, while stochastic variability can be
210 SPMS CoS Jeremie Houssineau exploration in model-based reinforcement
separately captured via an explicit noise model. Planning will then use a possibilistic rollout that propagates epistemic uncertainty through time, enabling intrinsic motivation signals such as ?epistemic
learninr
novelty? or ?possibilistic information gain? without being inflated by irreducible randomness.

The expected deliverables are: i) a short technical report clarifying the epistemic/aleatoric decomposition under possibilistic modelling, ii) a prototype MBRL agent integrating a possibilistic world model
(e.g., learned from ensembles or Bayesian last-layer surrogates) with planning, and iii) empirical validation on standard continuous-control benchmarks, comparing exploration efficiency and
robustness against noise-heavy environments.

Membranes play a central role in redox flow batteries by controlling ion transport, suppressing active-species crossover, and determining electrochemical efficiency and stability. This project focuses on
the rational design and evaluation of ion-sieving microporous polymer membranes for aqueous flow battery systems. Students will develop a clear understanding of flow battery operation and
Powering Flow Batteries with Ion-Selective
211 CEE CoE Ye Chunchun membrane transport mechanisms, then fabricate membranes via solution processing and controlled post-treatment to systematically tune pore structure and ion selectivity. Membrane properties will [Link]
Polymer Membranes
be quantified through swelling, ionic conductivity and ion diffusion measurements, while electrochemical performance will be evaluated using cyclic voltammetry, crossover tests and long-term flow-
cell operation. The project aims to establish robust structure-property-performance relationships to guide the design of next-generation membranes for scalable redox flow batteries.

Genetic exchange helps us understand how microbes adapt, affecting their behavior and potential to cause disease. New genetic exchange mechanisms, like CRISPR systems (Jinek et al. 2012) and
bridge RNAs (Durrant et al. 2024), have transformed genome editing and engineering. Previous studies (Smillie et al. 2011; Groussin et al. 2021) focused on genome exchanges of uniform genomic
regions, assuming exchanged regions remain largely unchanged. However, bacterial genomes are highly diverse due to mutations, gene deletions, horizontal gene transfer, and recombination. These
varied regions often encode important ecological functions. Traditional methods like BLAST (Camacho et al. 2009) and multiple sequence alignment (Edgar 2004) mainly detect changes such as point
mutations and short insertions/deletions (<20bp). Our preliminary data show that genetic exchange regions can evolve to include diverse long insertions/deletions (50-100bp), which are linked to
spacer changes in CRISPR systems, a bacterial immunity mechanism. These varied regions have been overlooked in previous methods.
Predicting DNA-Protein Binding via AI to Enhance
Therefore, it is crucial to design computational methods that can detect flexible genetic exchanges allowing for heterogeneous changes. Our long-term goal is to create an AI-guided platform to predict
212 SBS CoS Anni Zhang Bacteria-Based Cancer Therapy Safety and
emerging health risks in Singapore. The objective of this proposal is to design computational approaches to predict flexible microbial genetic exchanges and uncover new exchange mechanisms. We
Efficacy*
hypothesize that flexible computational algorithms, such as x-mers and language models, can identify various genomic patterns. To achieve this, we will:

Aim #1: Develop a machine learning model to detect flexible genetic exchanges.
Aim #2: Discover new mechanisms of microbial genetic exchange.

Upon completion, we expect to have developed new computational models that can reveal under-explored mechanisms of microbial evolution, aiding in the early prediction and preparedness for
emerging health risks in Singapore.

In this project, we aim to build deep neural networks that can predict characteristics of materials from their chemical formulae or 3D lattice structures. Example characteristics include stability, band
gap (for semi-conductors), voltage and current (for solar cells), etc. We will also study the adversarial robustness of the built predictors.
Predicting Inorganic Material Characteristics
213 CCDS CCDS Li Boyang
Using Deep Learning This project is in collaboration with world-class material scientists from NTU and the Hebrew University of Jerusalem.

Primary science learning requires learners to interaction and explore their physical environment. Evidence has shown that unguided exploration could lead to misconceptions in their explanations and
understanding of how things work. This suggests that for learning to take place, the spontaneous concepts of learners must be guided to become legitimate and formal theoretical scientific concepts.
Primary Science Learning: Role of Curricular
214 NSSE NSSE Tan Aik Ling Curricular resources such as textbooks, activity books, manipulatives and online materials can be used by teachers to mediate the process of transforming students' spontaneous ideas to theoretical [Link]
Resources in Mediating Science Learning
knowledge. In the project, students will work with a team of researchers to examine how grade 3 and 4 teachers and students interact with an integrated suite of curricular resources to learn science in
classrooms in Singapore.
The development of printed, flexible and biocompatible electronics that are composed of non-toxic materials and can completely disappear in a controlled fashion after fulfilling its duty will enable
215 EEE CoE Leong Wei Lin Printed Electronics [Link]
many novel applications such as sensors for food monitoring and integrated health diagnostics. This project aims to demonstrate a flexible, printed transistor based on bioinspired materials.

The macromolecular condensation driven by the liquid-liquid phase separation has more and more been recognized as a key principle to organize the intracellular macromolecules from nanoscale to
mesoscale for their functionality. Early studies extensively focused on understanding phase separation from the perspective of the molecules per se, exploring mechanisms such as multivalent
interactions, condensation propagation, and molecular organization. However, it remains unclear how the condensation-engaged cellular microenvironment regulates biomolecular phase separation.
Probing the cellular microenvironment for Indeed, the cellular microenvironment could vary a lot depending on the cell types and development stages. In this project, genetically encoded nanoparticle- or/and chemical dye-based sensors will be
216 SBS CoS Miao Yansong
molecular condensation during plant signaling* designed and applied to characterize the cellular microenvironment by dissecting and quantitatively analysing different physical properties of the cytoplasm microenvironment including viscosity and
macromolecular crowing. The condensation of several well-studied markers that undergoes phase separation either in 3D cytoplasm solution or 2D panel of plasma membrane will then be
characterized in different cell types with distinct cellular microenvironment. Furthermore, in vitro reconstitution will be conducted to recapitulate the condensation in simple solutions that mimic
different cellular physical properties. This project is proposed to understand how the cellular microenvironment regulates phase separation condensation assembly and plant signaling.

Our cells are known to alter the organization of their nucleus during aging and cancer. However, many of the changes are on the nanometer scale and thus challenging to probe. In this project, using
Probing the nanoscale nuclear organization in
217 CCEB CoS Zhao Wenting designed nanochips fabricated by high-resolution lithography, we will examine the changes in the nuclear deformability and nuclear envelope organization in aging and cancer cells. The student will
aging and cancer cells
learn frontier topics in cancer biology, cellular aging, nuclear mechanics, and nanobiotechnology, and gain fundamental research experience related to this topic.

Immersive narrative has been increasingly employed for educational purposes and awareness promotion. Both national agencies and international organizations are utilizing 360-degree videos to
Promoting Sustainability with Immersive
present immersive narratives to garner social attention toward sustainability issues. Despite the prevalence of virtual immersion, there is a lack of understanding regarding how immersive narratives
218 ITOM Div CoB (NBS) Ben Choi Narratives: Discovering the Underlying
can facilitate knowledge acquisition and promote behaviors. Drawing on the narrative transportation literature, we consider the effects of immersive narratives on sustainability knowledge acquisition
Mechanisms with Physiological Measurements
and pro-environmental behaviors. We further investigate the underlying mechanisms of the relationships. We tested our hypotheses in an experiment involving physiological measurements.
2025 is the International Year of Quantum Science and Technology. In this project, the candidate will study possible topics on quantum theory that can be covered in a secondary school syllabus,
219 NSSE NSSE Kwek Leong Chuan Quantum Mechanics for Secondary Schools
including hands-on activities. This project may involve collaboration with Prof. Myungshik Kim at Imperial College.
Join our project on Quantum Sensing, where we use atom interferometry to tap into the wave-like nature of atoms. By harnessing this revolutionary method, we can create unrivaled inertial navigation
Rainer Helmut
220 SPMS CoS Quantum Sensing / Matter Wave Interferometry sensors?pushing the boundaries of modern science and engineering. This hands-on experience will deepen your understanding of quantum mechanics and open doors to the rapidly expanding field of
Dumke
next-generation sensor technology. Don?t miss your chance to be at the forefront of tomorrow?s breakthroughs!
Quantum state tomography (QST) is a standard process of reconstructing quantum information of an unknown quantum state through measurements of its copies. QST is used to verify state
preparation, examine state properties such as correlations, and calibrate experimental systems. In this project, the candidate will join a group working on integrated photonic chip and look into the
possibility of implementing quantum state tomography on a photonic chip.
Quantum State Tomography on an Integrated
221 NSSE NSSE Kwek Leong Chuan
Photonic Chip
Prerequisite:
A good understanding of quantum mechanics as well as good programing skills with Python
 When two atomic layered materials are twisted together, they form an emergent moire pattern
that radically reconstructs their electronic bandstructure and component electronic wavefunction. Such quantum matter possess behavior more than the sum of their individual layers; for instance,
while graphene itself is non-superconducting, when two layers of graphene are twisted together, a strong superconducting state can form [1]. Similarly, while graphene does not possess an anomalous
Hall effect (often used as a signature of non-trivial geometric phase behavior), when a twisted bilayer graphene heterostructure on boron nitride quantized Hall resistance [2].

This computational and theory project will explore how geometric phases emerge in moire quantum matter. It will concentrate specifically on how broken symmetries (e.g., broken time reversal and
Justin Song Chien Quantum geometry in strongly interacting moire inversion symmetry) enable pronounced values of quantum geometric quantities such as the Berry curvature, quantum metric, Berry connection polarizability that may dominate the linear and
222 PAP Div CoS [Link]
Wen quantum matter* nonlinear transport behavior in moire quantum matter. The candidate will perform bandstructure calculations of moire matter as well as Wilson loop computations of quantum geometric quantities.
The candidate should be familiar with basic bandstructure of solids, the adiabatic theorem, as well as have an ability to code.

[1] Cao, Yuan, et al. "Unconventional superconductivity in magic-angle graphene superlattices." Nature 556.7699 (2018): 43-50.

[2] Serlin, M., et al. "Intrinsic quantized anomalous Hall effect in a moire heterostructure." Science 367.6480 (2020): 900-903.

Quantum sensing technologies operate with quantum mechanics as the basis and provide extremely high efficiency for various sensing applications. One such quantum sensing platform is based on
Quantum sensing with two dimensional
223 EEE CoE Song Peng quantum photo emission in semiconductors. This project aims to develop quantum sensing devices based on exciton emissions in two-dimensional semiconductor heterostructures. In this project,
semiconductors
students will be exposed to cutting-edge quantum sensing sciences and technologies, to gain hands-on experience fabricating state-of-the-art quantum optoelectronic devices.

AI-based materials design is widely regarded as the fourth paradigm of material science, supported by the milestone achievements of the Materials Genome Initiative (MGI) and massive databanks like
the Materials Project, JARVIS, and NOMAD. While these repositories provide a solid foundation for data-driven discovery, traditional geometric deep learning models?specifically Graph Neural
Networks (GNNs)?face significant structural bottlenecks. GNNs often struggle to fully capture periodic symmetry and the higher-order, many-body interactions prevalent in crystalline systems. Because
they primarily rely on pairwise atomic representations, they fail to assimilate the complex materials periodicity and multiphysical information necessary for accurate functional prediction, leaving a
Quotient complex transformer (QCformer) for major gap in the digital modeling of periodic structures.
224 MAS Div CoS Xia Kelin
material design and discovery
In this research project, we propose Quotient Complex Transformer (QCformer), a novel framework that models material structures as quotient (cellular) complexes (QCs). This representation naturally
incorporates material periodicity via a quotient operation and captures both pairwise and many-body interactions through simplices of varying dimensions. By utilizing efficient higher-order features
within a simplex-based Transformer module, the QCformer learns the intricate relationship between a material?s topology and its physical properties. We will demonstrate its efficacy by pretraining on
benchmark datasets and evaluating its performance on the bandgap properties of Hybrid Organic-Inorganic Perovskite (HOIP) datasets.

Cyber-physical systems are a class of systems that combine both cyber and physical components (e.g., sensors and actuators) to monitor and control the physical world. Such systems are often
responsible for control in safety-critical applications, such as in automotive systems, and thus, impose strict timing and safety requirements. In modern cyber-physical systems, deep neural networks
(DNNs) are widely used for planning, perception, and control tasks. Specifically, to ensure a greater explainability and a more modular architecture, pipelines of DNN models are used. These pipelines
can often be divided into planning, perception, and control stages, with data dependencies between the components. Furthermore, in such a pipeline, the performance of one component may impact
Real time scheduling of deep neural networks on
225 CCDS CCDS Arvind Easwaran the scheduling decisions of the subsequent components. Thus, a comprehensive scheduling framework considering performance as well as data dependencies among different components of the [Link]
heterogenous hardware
pipeline is required. Furthermore, considering the heavy memory footprint of the contemporary DNN models, it is essential that the scheduling algorithms are aware of the memory limitation.
Additionally, contemporary computing platforms consist of heterogeneous computing units (specifically, CPUs, GPUs, and NPUs/TPUs) to execute workloads of different characteristics efficiently;
however, with corresponding challenges. Thus, we aim to study the problem of risk-aware scheduling of real-time deep neural network workloads on heterogeneous resources, while considering the
time-memory-accuracy trade-offs and satisfying the system design constraints.?

Mobile robots and autonomous vehicles need to know their position in a map during navigation. This is problem called localization. Unfortunately they cannot rely on GPS for localization in indoor
Real-time Visual Localization System on an
226 CCDS CCDS Lam Siew Kei environments or in places with tall buildings. An alternative is to use visual localization where computer vision algorithms are used to both localize the robot or vehicle, while simultaneously generating [Link]
Embedded Platform
a map of the environment. The aim of this project is to develop a high-speed and light-weight algorithms for visual localization on an embedded platform.

At the Institute for Digital Analytics and Science (IDMxS), we are studying antibiotic resistance. This project explores how bacterial cells respond to antibiotics by analyzing microscope images. Using
Recognizing Antibiotic Effects in Bacteria Using freely available image datasets, students will investigate whether computer-based image analysis can distinguish between bacteria that are susceptible or resistant to treatment. The project
227 EEE CoE Wang Lipo
Machine Learning and Image Analysis emphasizes hands-on learning with image processing and pattern recognition, but it does not require prior knowledge of biology or machine learning. Students will work with the data, explore analysis
methods, and compare results to understand how image features can reveal important biological information.
Recent advances in Large Language Models (LLMs) have demonstrated remarkable capabilities in natural language understanding and generation, raising growing interest in their potential for emotion
Yuvaraj Recognizing emotions with Large Language
228 SoLEC SoLEC recognition. This study explores how LLMs can be utilized to develop more robust, context-aware emotion recognition systems (using EEG, facial expressions, or other physiological signals) for [Link]
Rajamanickam Models (LLMs)
applications in education, mental well-being, and human?computer interaction.
Even though alkyl C(sp3)-H functionalization has seen significant improvements in the past few years, long chain fatty acids have yet to be investigated as starting substrates for C(sp3)-H activation
Regio and Stereoselective Functionalisation of
229 CCEB CoS Tan Choon Hong reactions. Our group believes functionalization of C(sp3)-H bonds at specific positions along the alkyl chain of long chain fatty acids will generate a library of novel, unique and unnatural fatty acids [Link]
Fatty Acids*
which are likely to have their own unique properties and bioactivities.
Accurate and reliable pose estimation is essential for applications such as autonomous navigation, robotics, and intelligent vehicles. Ultra-Wideband (UWB) technology, known for its high precision and
robustness in indoor environments, has become a promising solution for localization. While UWB measurements provide accurate range or angle-of-arrival, they are not sufficient to fully determine the
Relative Pose Estimation with Ultra-Wideband
230 EEE CoE Chau Yuen pose of a vehicle due to limited spatial observability. A common strategy to enhance pose estimation is to deploy multiple UWB nodes on the vehicle. This project aims to develop a robust UWB-based
Ranging and AoA
pose estimation algorithm that operates reliably under both line-of-sight (LOS) and non-line-of-sight (NLOS) conditions by leveraging convex optimization techniques. The developed method will be
evaluated against conventional nonlinear least-squares approaches in terms of accuracy, while maintaining computational efficiency suitable for real-time navigation.
Visual signals occupy number one position among all the signals at the input of any AI system or product. Hence, the ability of understanding the meanings from visual signals is very important to the
Research Toward Achieving Cognitive Binocular
231 MAE CoE Xie Ming development of advanced robots and AI systems. The objective of this research project is to investigate physical principles which will enable a binocular vision system to produce meaningful outputs [Link]
Vision
about the detailed knowledge behind a 3D scene.
AI systems are knowledge-based systems which depend on the mental abilities of acquiring and organizing knowledge derived from signals and data. In general, knowledge include both conceptual
Research Toward Achieving Large Knowledge
232 MAE CoE Xie Ming meanings as well as physical meanings. The objective of this research project is to investigate the design and development of large knowledge models which include the organized knowledge coming [Link]
Models
from both the physical world and the conceptual worlds.
Resource provisioning and scheduling for edge and cloud architecture have gained a lot of attention in the recent past both in academia and industry. This interest is primarily driven by the enormous
success of two technologies in the field of computing: 1) Artificial intelligence (AI), and 2) Internet of Things (IoT). Cloud architectures play a central enabling role for AI due to the significant demand
that this technology imposes on computation resources. Orthogonally, IoT has enabled an unprecedented decentralization and deployment of applications across application domains, including several
Resource Scheduling for Deadline-Constrained in which AI is gaining prominence. As a result, the edge and cloud hardware architecture are poised to dominate the field of computing across application domains. A key challenge in this edge and
233 CCDS CCDS Arvind Easwaran [Link]
Applications in Edge-Cloud Systems cloud architecture is the efficient scheduling of time-critical applications while satisfying various constraints and optimization objectives. This is a research project, which aims to address this challenge
by developing decentralized and online scheduling algorithms that can provide deadline guarantees for time-sensitive applications while minimizing the power consumption of end devices and response
time of non-time-sensitive applications. These capabilities will be enabling technology for the deployment of various safety-critical and time-sensitive applications across domains such as
manufacturing, robotics, urban energy system, autonomous driving system, etc.

Cancer has been recognized as a human ailment for centuries. With the advances in medical sciences, people have revealed that Ras oncogenes are the major cause of the progression of various cancer
types. However, no effective anti-Ras pharmacological inhibitor has reached the clinic despite 30-year efforts. The major challenge for anti-Ras therapeutics attributes to the complicated biochemical
Revealing the Nanoscale Organization of
234 CCEB CoS Zhao Wenting interactions between Ras and other signaling players. To solve that, biophysics have pointed out a novel potential of cells to adjust their behaviors by sensing the physical signals from the
Oncogenic Ras Protein in Cells
neighborhood. In this project, we aim to explore the effect of surface topography on Ras functions and cancer progression. The student will be expected to prepare and image cancer-related proteins on
nanochips in cells, as well as to practice lab techniques in nanobiotechnology and molecular biology.
 This project aims to explore the effects of hydrogen exposure on various structural materials, particularly steels and other metallic alloys commonly used in critical infrastructure. Hydrogen is known to
cause embrittlement in metals, significantly reducing their mechanical strength and increasing the risk of failure in components exposed to hydrogen environments. This phenomenon is particularly
relevant in industries such as energy, transportation, and chemical processing, where hydrogen is used or produced.

Yi-Sheng (Eason) Risk assessment of structural materials exposing In this project, you will be involved in assessing the susceptibility of different metallic materials to hydrogen embrittlement. Using advanced testing methods and analysis techniques, you will
235 MSE CoE [Link]
Chen to hydrogen investigate how hydrogen affects the microstructure, mechanical properties, and overall integrity of these materials. Your work will help identify which materials are more prone to failure, and under
what conditions, providing valuable insights for the safe design and maintenance of hydrogen-exposed structures.

This research offers an exciting opportunity to work at the intersection of materials science and engineering, with real-world applications that contribute to the development of safer and more reliable
infrastructure in hydrogen-related industries. Join us in making a tangible impact on the future of sustainable and secure energy solutions.
Deep Neural Networks (DNNs) playing the role of controllers have demonstrated impressive empirical performances on challenging control problems like autonomous vehicles. However, the potential
adoption of DNN controllers in real-life applications also gives rise to a growing concern over the safety of these DNN-controlled systems, especially when used in safety-critical applications. It is worth
noting that simulation-based testings [1] cannot provide formal safety guarantees such as "no system trajectory from the initial state set will lead to an obstacle collision". Recently, Dalrymple et al. [2]
gave a high-level picture of formal AI safety, advocating AI systems that are guaranteed safe in the sense that they are equipped with a quantitative safety guarantee.

In previous work [3], we proposed a novel framework for unifying both qualitative and quantitative safety verification problems of DNN-controlled systems under perturbations. To achieve this, we
formulate the safety verification tasks as the synthesis of valid neural barrier certificates (NBCs) that can serve as proof certificates of system safety. Initially, the framework seeks to establish almost-
sure safety guarantees through qualitative verification. Whenever qualitative verification fails, our quantitative verification method is invoked, yielding precise lower and upper bounds on probabilistic
safety across both infinite and finite time horizons.

Though our approaches provide formal safety guarantees (in theory) and show the effectiveness over a set of benchmarks (in practice), there are still some remaining questions to be addressed.
[Link] of different hyperparameters.
We have lots of hyperparameters in [3], like the number k used in the k-inductive barrier certificates, and the granularity used to split the state space. Next, an interesting direction is to explore and
Safety of Systems controlled by Deep Neural discuss the effects of these hyperparameters on the final results.
236 PO PO Ong Chih-Hao Luke
Networks [Link].
In this work, we represent barrier certificates (BCs) [4] in the form of neural networks (due to the powerful expressivity of NNs), and adopt the learner-verifier architecture [5] to synthesize our NBCs. A
common problem is to scale our approaches to large neural networks and high-dimensional system states.
[Link] forms of quantitative safety bounds.
In this work, we can derive quantitative safety bounds through our NBCs. However, the bounds over finite time horizons are restricted to certain fixed forms (i.e., linear or exponential w.r.t. the time
horizon), for technical reasons in Probability Theory. An interesting question is how to generate safety bound functions in other forms, e.g., polynomials, or even arbitrary forms (if exist).
[Link] of quantitative safety probability bounds.
One drawback of BCs is that BC conditions have nothing to do with the tightness of quantitative bounds, i.e., although we can find a BC satisfying all its BC conditions, the tightness of the derived bound
via this BC is unknown. However, when we focus on quantitative analysis, the tightness of the quantitative results is non-trivial. To address this, we proposed a simulation-guided loss term (see Section
6 in [3]) to improve the tightness of the bound during training. Next, we will consider other methods about improving the tightness of quantitative bounds.

References

[1] Zhang et al., Machine learning testing: Survey, landscapes and horizons. IEEE Transactions on Software Engineering. 2020.
In the era of large language models (LLMs), high-quality data has become a valuable commodity, with companies increasingly collecting, curating, and synthesizing datasets for sale. However, a major
Secure and Private Data Valuation for Large challenge in the data marketplace is that buyers cannot assess the true effectiveness of the data before making a purchase, leading to trust issues and inefficiencies. To address this, we need a novel
237 CCDS CCDS Dong Wei
Models framework for private data valuation that enables fair pricing and verification without exposing sensitive information or undermining the profits of both buyers and sellers. Our research explores
cutting-edge techniques such as zero-knowledge proofs, differential privacy, and secure multi-party computation to design a secure and transparent data valuation mechanism.
Most civil infrastructure is large-scale, e.g., tunnels, viaducts, dikes. To monitor the conditions of these structures, we need cost-effective sensing technologies. In contrast to state-of-the-art solutions,
Self-deployable robotic IoT for large-scale
238 CEE CoE Fu Yuguang like fiber optic sensing, which is static, this project aims to develop mobile sensors (i.e., robotic nano-drone) that are self-deployable for contact-based sensing, e.g., acceleration and ultrasonics. This
structural health monitoring
will involve prototype development and programming.
Ultraviolet microscopy is a new technique in medical imaging which promises high resolution images with cellular details for doctors to use in diagnosis. However, this technique has yet to be translated
Signal and image processing for ultraviolet
239 CCEB CoS Liang Kaicheng into an endoscopy system for more demanding clinical applications. A challenge lies in the real-time processing of raw signals into meaningful and high quality images. This biomedical project will see
endoscopic microscopy
the student leveraging principles of signal and image processing to enable real-time imaging for future healthcare applications.

Large-scale high-dimensional vector data has become ubiquitous in contemporary times. For instance, various forms of unstructured data, such as images, videos, texts, and speeches, are typically
transformed into vectors using deep learning techniques (e.g., word2vec, node2vec, item2vec, etc.). These vectors are subsequently employed in downstream analytical tasks. K nearest neighbor (KNN)
search in high-dimensional vector space constitutes a fundamental problem with a wide array of applications in information retrieval, recommendations, and retrieval-based large language models.
Due to the curse of dimensionality, exact KNN queries often result in unacceptable response times. In pursuit of a better balance between time and accuracy, many researchers have turned to its
Similarity Search on High-Dimensional Vector
240 CCDS CCDS Long Cheng relaxed version, known as approximate K nearest neighbor (AKNN) search. Various algorithms have been proposed to address the AKNN problem, encompassing quantization-based, graph-based, [Link]
Data
hashing-based, and tree-based approaches. However, popular AKNN algorithms such as the quantization-based ones do not provide theoretical guarantees and may fail in some scenarios. In addition,
most of these algorithms primarily focus on vector data and may fall short in real-life applications involving more than just vectors. This project aims to develop new quantization-based algorithms that
provide theoretical guarantee as well as algorithms to address various AKNN problems, including the attribute-filtering AKNN problem, the AKNN problem for sparse vectors, and the multi-index AKNN
problem. These would help to bridge the gap between AKNN solutions and real-world applications featuring diverse data types.

This project aims to explore how fluid injection can trigger earthquakes and aseismic slip in the Earth?s crust. Using advanced hydro-mechanical models, we will simulate the complex interactions
Simulating Earthquakes Triggered by Fluid
241 ASE CoS Luca Dal Zilio between fluid flow and fault mechanics. The study will focus on understanding the transition from slow aseismic creep to rapid seismic slip, influenced by fluid pressurization. By comparing model [Link]
Injection*
results with data from in-situ experiments, the project will shed light on the mechanisms behind fluid-induced seismicity, offering insights into seismic hazard assessment and geo-energy exploration.
Climate change is melting Arctic ice, opening the Northern Sea Route (NSR) as a potential shortcut between Asia and Europe. Will this new route replace the traditional Suez Canal Route, which is the
lifeline of Singapore?s logistics economy?

This project answers this question by building a simulation platform. Instead of static comparisons, we will use dynamic data, including ship operational data, future weather patterns, oil prices, and
Simulating the Future of Global Shipping
242 CEE CoE Kim Youngrong carbon taxes, to predict the most profitable and eco-friendly route. This research is perfect for students interested in climate change, data science, maritime transportation, and marine technology, [Link]
Focusing on the Era of Northern Sea Routes
which offers a unique perspective on how Singapore and global shipping should prepare for the future.

Students will engage in modeling ship voyages under future climate scenarios, comparing several shipping routes in terms of energy consumption, fuel costs, travel time, and environmental impact.
Potentially, our results will tell us "When does the Arctic route become a threat to the Suez route?", and "How will it affect the global maritime trade?

The rapid integration of renewable energy sources has significantly reduced the inertia of modern power systems, posing challenges to system stability and reliability. Grid-forming converters have
emerged as a key enabling technology for future low-carbon power systems by allowing power electronic converters to actively regulate voltage and frequency, mimicking the behavior of synchronous
generators.
Smart Grid-Forming Control for Future Low-
243 EEE CoE Tang Yi This project investigates advanced grid-forming control strategies for inverter-based resources. The study will explore fundamental control approaches such as virtual synchronous machines, droop-
Carbon Power Systems
based control, and virtual oscillator control. Emphasis will be placed on understanding dynamic performance, stability margins, and robustness under grid disturbances such as load changes and weak-
grid conditions. Simulation-based analysis will be conducted to evaluate transient response, frequency stability, and interaction with other grid-connected converters.

The project provides strong exposure to modern power system challenges and is well suited for students interested in renewable energy integration, smart grids, and advanced control systems.
 Wireless power transfer (WPT) technologies are increasingly critical for enabling reliable and autonomous operation of mobile platforms, including autonomous vehicles, mobile robots, and automated
guided vehicles. By eliminating physical connectors, WPT enhances system robustness, operational safety, and maintenance efficiency, particularly in dynamic or challenging environments.

This project focuses on the design and analysis of inductive wireless power transfer systems tailored for autonomous applications. Key areas of investigation include resonant power transfer topologies,
Smart Wireless Power Transfer for Autonomous
244 EEE CoE Tang Yi magnetic coupling and coil design, and compensation network optimization to ensure efficient energy transfer under varying alignment and load conditions. Control and regulation strategies will also
Systems
be developed to maintain stable and efficient power delivery in the presence of positional misalignment and dynamic operational conditions.

Simulation-based studies will be conducted to evaluate system performance, including transfer efficiency, power regulation, and robustness under realistic operating scenarios. The outcomes of this
project aim to advance reliable wireless charging solutions for autonomous systems, supporting applications in robotics, automated logistics, and intelligent transportation.

Juvenile arthritis is not a singular disease but rather a term encompassing various joint pain, inflammation, and joint-related conditions. The current treatment approach is bolus injection of steroids,
which are associated with severe side effects, such as osteoporosis and uncontrolled drug release which would hinder the growth of children and cause aversion to the drug. To address this issue, a
responsive drug delivery system is needed. Based on research conducted in our lab, we propose a responsive drug delivery platform for inflammatory diseases, utilizing an inflammation-sensing module
Smart drug delivery system for management of
245 CCEB CoS Dang Thuy Tram and a drug module. In the presence of inflammation, the platform undergoes cleavage, leading to the targeted release of the drug. Conversely, in the absence of inflammation, the system remains
arthritis*
inactive, preserving drug. This project entails screening various anti-inflammatory drugs for incorporation into the drug release platform, developing different polymer systems to meet specific clinical
requirements, and investigating their impact on arthritis. The student working on this project will gain valuable technical skills in polymer synthesis, polymer microgel characterization techniques,
experimental planning as well as exposure to entrepreneurship in biotech towards commercialization.
This project is mainly about chemistry and materials, and it focuses on energy?something that quietly supports almost everything we use every day, from transportation and industry to electronics.
Solid electrolytes are interesting because they can make electrochemical systems safer and more stable, and they are being explored for many applications, such as fuel cells, and solid-state metal
batteries.
Solid state electrolytes for high performance
246 EEE CoE Zhang Weiran The problem is that when solid electrolytes are in contact with electrodes, things often go wrong. Mechanical stress, chemical reactions, and electrochemical processes interact with each other, leading [Link]
solid-state batteries
to instability and performance loss. In this project, we want to understand what actually happens at these solid-solid interfaces and learn how to control them. The goal is to use that understanding to
design better solid-state energy systems that can eventually work in the real world

As a student, you will work hands-on with materials and battery experiments and learn how chemistry and materials come together in real energy systems.
There has been a marked surge of recent research interest in developing DNN-based computational methods for solutions of differential equations. Compared with classical methods, such plug-and-
play approaches are applicable to more general classes of problems with a uniform framework. However, practitioners are plagued with the accuracy and efficiency in the involved learning and training.
Spectral Neural Networks for Solving Differential
247 MAS Div CoS Wang Li-Lian How to properly integrate classical tools with these emerging techniques becomes essential to surmount these obstacles. The aim of this project is explore the construction of spectral NNs that can
Equations*
excel the capabilities of high accuracy of spectral methods and approximability of NNs. The essential idea is to design single layer NN and build in spectral basis and then tensorise the resulting
structures for multiple dimensions.
Moore's law based growth of conventional electronics has stagnated decades ago. In order to push the electronics technology beyond, various alternatives have been considered. Spintronics materials
have played a supplementary role and magnetic random access memory (MRAM) has already entered the market as a replacement for e-Flash. In order to push the spintronics materials further, it is
Spin torque Materials for energy efficient essential to reduce the energy consumption. In this sense, spin-orbit torque and orbital torque play an important role.
248 PAP Div CoS S. N. Piramanayagam [Link]
electronics*
In this project, the student will work on growth of materials and investigation of materials and devices. The student will learn fabrication and characterization techniques, which are important aspects
of experimental condensed matter physics research.

The removal of introns and linkage of exons via splicing is an essential step in the expression of human genes, and certain changes of this process cause many genetic diseases as well as cancer. In
Francesc Xavier Roca Splicing mechanisms and their implication in addition, alternative splicing is a key regulatory mechanism whose alterations are also associated to disease. In this project, the student will study the mechanisms of splicing in relation to one of the
249 SBS CoS
Castella human disease* projects in the lab, in the context of either human myeloid cells and leukemia, or the regulation of antiviral genes, or the diagnosis of novel genomic variants affecting splicing and found in the
Singaporean populations. The student will use standard molecular biology such as cloning, cell culture and transfection, RNA extraction, RT-PCR, and perhaps bioinformatics.
In chemical synthesis, application results from fundamental research, and many basic reactions frequently involve highly reactive and transient intermediates. Isolation of such species leads to
Stable version of reactive main group species for
250 CCEB CoS Rei Kinjo dramatic advance in the area. In this project, we challenge to synthesize, isolate and characterize the highly reactive main group species, namely, p-block elements. We will further investigate their [Link]
small molecule activation
reactivity for small molecule activation as well as catalyst application.
WebAssembly is a widely used virtual machine, now supported by all major Web browsers. WasmCert-Isabelle is a machine-checked encoding of the WebAssembly specification in the Isabelle-HOL
tool, created in order to analyse the WebAssembly type system and prove it correct - this technique is known as "mechanisation", and it is one of the most reliable ways to build a strong certification of
trust in the system under analysis. A new form of control flow, called stack switching, is being experimented with in WebAssembly. This project would aim to extend WasmCert-Isabelle with stack
251 CCDS CCDS Conrad Watt Stack switching for WasmCert-Isabelle switching, in order to analyse the correctness of this new feature. This would involve becoming familiar with Isabelle/HOL and the basics of the WebAssembly semantics, and the basics of delimited
continuations, the underlying primitive for stack switching. The project member will work towards updating the Isabelle-HOL model of WebAssembly itself, and associated certifications - namely the
proof that the WebAssembly type system is correct, and a verified reference interpreter built on top of the model.

Stress granules (SGs) are dynamic cytoplasmic aggregates containing mRNA, RNA-binding proteins, and translation factors that form in response to cellular stress. Evidence indicates that SGs are
intended to be transient structures, but the chronic stresses associated with aging lead to persistent SGs that appear to act as a nidus for the aggregation of disease-related proteins. This project aims
to characterize the dynamics of stress granules that form in microglial cells under different types of stress, such as oxidative stress, pro-inflammatory stimuli, and lysosomal damage. We will combine
human microglial cell lines and induced-pluripotent stem cell-derived microglial cells with confocal and super-resolution imaging modalities to obtain a quantitative overview of how SG dynamics are
Stress Granule Dynamics During Inflammatory influenced by environmental stressors and what interactions these structures establish with other cellular organelles.
252 SBS CoS Claudio Bussi and Oxidative Stress Conditions in Human References: [Link]
Microglial Cells* Bussi, C., Mangiarotti, A., Vanhille-Campos, C. et al. Stress granules plug and stabilize damaged endolysosomal membranes. Nature 623, 1062?1069 (2023)
Wolozin, B., Ivanov, P. Stress granules and neurodegeneration. Nat Rev Neurosci 20, 649?666 (2019)
Herman, A.B., Silva Afonso, M., Kelemen, S.E., et al. Regulation of stress granule formation by inflammation, vascular injury, and atherosclerosis. Arteriosclerosis, thrombosis, and vascular biology,
39(10), pp.2014-2027 (2019).

This project seeks to employ state-of-the-art electro-fabrication technology for the nanoscale engineering of chiral structures to tap on the potential of chiral induced spin selectivity (CISS) to
Study of Nanoscale Chirality Using Electro-
253 CCEB CoS Tan Choon Hong discriminate enantiomers based on electronic effects. We also seek to investigate how chiral nanostructures may be applied in asymmetric catalysis probing the question of whether structural chirality [Link]
fabrication*
can be transferred to molecular chirality. This will shed light into the mechanism behind CISS as well as the nature of chirality in various forms namely, structural and molecular chirality.
The rapid development of portable electronics, electric vehicles, and smart grids creates large demand for energy storage devices with not only high energy density but also low cost. Si anode is one of
Study of solid state electrolyte for
254 EEE CoE Lee Seok Woo the most promising candidates due to their ultrahigh theoretical energy density and the advantages of environmental benignity and the low cost of silicon. The candidate will study the mechanical [Link]
electrochemical energy storage
interaction of Silicon anode and solid-state electrolyte during charging and discharging.
Critical metals such as lithium, cobalt, and nickel are essential for batteries and clean energy technologies, yet their selective recovery from complex resources remains highly challenging. This project
focuses on sub-nano channel ion sieving membranes, where angstrom-scale confinement allows ions to partially dehydrate and interact with specific chemical binding sites. Students will fabricate and
Sub-nanochannel membranes for critical metal characterize membranes by incorporating tailored functional groups and investigate how ion charge density, hydration energy, coordination geometry, and intersite spacing influence selective ion
255 MSE CoE Shen Jie
recovery transport within confined channels. Experimental work will examine ion partitioning, competitive transport, and the separation performance of representative Li+/Na+/Mg2+ or Co2+/Ni2+ salt
mixtures. The project offers hands-on training in advanced materials processing, membrane testing, and analytical techniques, while contributing to practical membrane-based strategies for
sustainable resource recovery.
This project focuses on optimizing quantum gate operations in superconducting quantum processors to minimize error rates and improve fidelity. The student will begin with a review of existing gate
optimization techniques, followed by using simulation tools to model and identify potential improvements. They will then implement these optimized gate designs on superconducting qubits, working
Rainer Helmut Superconducting Quantum Processors
256 SPMS CoS closely with the research team. Comprehensive testing and characterization of the optimized gates will be conducted to evaluate performance. The student will document their findings and
Dumke Optimization*
methodologies, culminating in a detailed final report and presentation. This project provides hands-on experience in quantum computing, emphasizing gate optimization crucial for advancing reliable
quantum processors
 This research explores how imaginative alternative universes can be used to enhance students? critical thinking and conceptual understanding in science. Through carefully designed ?what-if?
scenarios, we aim to challenge students to apply scientific reasoning in novel contexts, pushing beyond memorisation into deep, systems-level thinking. A key innovation in this project is the use of AI
tools, such as large language models (LLMs) and interactive simulations, to co-create and explore these hypothetical scenarios. We?ll examine 1) how AI can assist in generating rich, scientifically
Support Student Science Learning through
257 NSSE NSSE Edwin Chng grounded alternate worlds, 2) how these worlds can be used in the science classroom to foster curiosity and deeper engagement, 3) how students navigate scientific principles when the rules of the [Link]
Alternative Universes and AI
universe are altered, and 4) how AI can support student learning and critical thinking when dealing with alternate universes. We are seeking enthusiastic students who are interested in science
education, cognitive science, speculative design, and/or artificial intelligence. Your work may involve 1) designing alternative universe scenarios, or 2) experimenting with AI tools like LLMs to scaffold
student inquiry.
This project aims to examine how consumers value specific sustainability claims in everyday consumer packaged goods?and how those valuations change over time and across demographic segments.
A number of sustainability related claims, e.g., Fair Trade, Rainforest Alliance, recyclable/compostable packaging, organic/natural ingredients, ethical sourcing, among others, are typically used in
consumer packaged goods. Using household scanner panel data and UPC-level sustainability claims extracted from packaging data, we propose to use a structural demand model adapted for multi-unit,
258 MKT Div CoB (NBS) Sadat Reza Sustainability claim valuation in consumer goods repeat purchasing to recover willingness-to-pay for distinct sustainability claims and test whether these valuations evolve over the last few years. We also wish to examine whether these valuations
vary systematically by income, education, age, and related characteristics. To explain observed differences and the intention?behaviour gap, the project complements the observational analysis with
longitudinal surveys and experiments, measuring attitudes, norms, values, and perceived control repeatedly over two years. Findings will inform firms? segmentation and messaging strategies and
guide policymakers on targeted nudges and standards to accelerate sustainable consumption
Project Overview

Lithium is a critical material for next-generation batteries, but supply is limited and recycling remains a major challenge. This project aims to develop a selective membrane system capable of recovering
lithium ions efficiently from complex waste streams such as spent lithium-ion batteries or brine solutions. By tailoring the membrane?s chemistry and structure, we will enhance ion selectivity,
durability, and recovery efficiency, while minimizing energy use and secondary pollution.

Objectives

1. Fabricate and optimize nanostructured membranes with selective lithium-ion transport properties.

2. Test separation performance in simulated waste battery leachate solutions.

3. Evaluate stability and recyclability of the membranes across multiple recovery cycles.
Sustainable Lithium Recovery from Spent
259 NSSE NSSE Edison Ang Huixiang [Link]
Batteries Using Advanced Membrane Technology 4. Compare cost-effectiveness with conventional lithium recovery techniques.

Skills Students Will Gain

Knowledge of sustainable materials development for circular economy.

Exposure to real-world energy and environmental challenges.

Candidate Requirements

Prior experience in membrane fabrication and electrochemistry.

Strong interest in sustainable energy and materials science.

One fascinating example of cooperative behavior in nature is the swarming of honeybees whereby honeybees self-assemble into tightly packed structures to protect the queen bee. The physics of
collective behavior is well described by the emerging framework of active matter, now a major paradigm in statistical physics.. We are interested in understanding how local behaviorial rules at the
260 PAP Div CoS Yong Ee Hou Swarming behaviors in honey bees individual level affect organization patterns and functions at the population scale for biological systems such as honeybees. In this work, we will model the dynamical process of honeybee swarming as [Link]
active particles (or agents) to understand the physics of honeybee self-organization.

Organic synthesis is a fundamental activity that provides access to compounds which may have a natural origin, i.e. previously isolated from a living organism, or are unnatural, i.e. designed by people.
Synthesis is underpinned by spectroscopic methods, especially high field NMR, to ensure that the structures of the synthesized molecules are confirmed. Organic synthesis provides access to the
compounds, acts to confirm structural assignment and generates new chemical knowledge.
Roderick Wayland
261 CCEB CoS Synthesis of Natural and Unnatural Products
Bates In this project, you will synthesise molecules of current interest. This may be because they are natural products, suggested to be natural products or due to anticipated biological or materials
properties. Organic synthesis in this laboratory has special emphasis on the control of stereochemistry and the use of transition metal catalyzed reactions such as Pd-catalysed coupling, carbonylation,
alkyne cyclotrimerisation and metathesis.

Smart contracts have shown great potential in revolutionizing the legal environment, by streamlining processes, reducing costs, and increasing transparency and efficiency. A smart contract can be
viewed as a self-executing agreement with the terms of the contract written directly into code. When certain conditions are met, the contract automatically executes, ensuring the agreed-upon actions
Synthesising Provably Correct Smart Legal
262 CCDS CCDS Li Yi are carried out without the need for intermediaries or manual intervention. One of the prominent attempts on smart legal contract is the Accord Project ([Link] Smart legal
Contracts with LLMs
contracts have been applied in domain such as Finance, insurance, purchase agreements, etc. In this project, we would like to build a technique to translate legal documents into smart contracts
automatically with the help of Large Language Models (LLMs). The success of this project will enable average layman users to create their smart legal contracts without deep technical knowledge.

In this project, we focus on the creation of synthetic data to train AI models for computer vision tasks in educational settings. Synthetic data, generated through advanced simulations and models,
Synthetic Data for Computer Vision Tasks in offers a powerful tool for overcoming limitations of real-world datasets, including privacy issues, data scarcity, and bias. In this work, you will engage in groundbreaking work that aims to enhance AI's
263 NSSE NSSE Edwin Chng [Link]
Education ability to recognize and interpret visual information accurately. We are seeking motivated students with a strong background in computer science or artificial intelligence (preferred but not necessary).
Familiarity with programming languages such as Python and a passion for exploring new ideas are essential. Previous experience in research is a plus but not required.

This project aims to engineer synthetic proteins to create immune regulatory hubs, effectively rewiring plant signaling pathways to enhance growth or strengthen defense against pathogens. By
leveraging synthetic protein engineering, we will design novel proteins that can integrate into plant signaling networks, modifying their responses to environmental cues and microbial attacks.
Structural biology techniques will be used to elucidate the three-dimensional structures of these engineered proteins and predict their interaction sites. Systems biology methods will enable us to
model the impact of these synthetic hubs on overall plant signaling networks. In combination with protein biochemistry, we will assess the functionality of these proteins in vitro. Transgenic plants
expressing these engineered proteins will be generated to validate their effectiveness in vivo, either to boost growth or enhance immune responses. This approach offers a promising strategy to
Synthetic Protein Engineering for Plant Signal
264 SBS CoS Miao Yansong develop crops with improved resilience and productivity under stress conditions. [Link]
Rewiring*

Terahertz semiconductor laser has attracted a lot of attention in recent years due to its potential applications in biomedical sciences, imaging, security, spectroscopy, gas/chemical sensing,
communications, military applications and astronomy sciences. Quantum cascade laser is a semiconductor laser based on intersubband transitions. Through the bandstructure engineering, one can
design the quantum cascade laser to cover the Terahertz regime from 1-5THz.
Terahertz quantum cascade laser
265 CoE CoE Wang Qijie In this project, the prospective student will focus on electrical and optical characterization of high performance mid-infrared quantum cascade lasers and use it for gas sensing applications and compare [Link]
characterizations
it with the existing Fourier Transform Infrared Spectroscopy technologies. The student will gain skill training in instrumentation, characterisation, and measurements of quantum cascade lasers. After
finishing the project, the student will learn the knowledge about spectroscopy and how to characterize the performance of mid-infrared lasers, such as light-current, voltage-current, and spectra
characteristics.
 DDX3X is an ATP-dependent RNA helicase. While germline mutations in the DDX3X gene are the underlying cause of a neurodevelopmental disorder in affected females called DDX3X syndrome, DDX3X
somatic mutations have been associated with human cancers. DDX3X plays tumor suppressive or oncogenic role depending on the tumor type. We discovered DDX3X mutations as an important
prognostic and chemosensitivity predictive marker in non-Hodgkin lymphoma subtypes. Most DDX3X mutations are missense, truncating, or single-base substitutions distributed within the ?helicase [Link]
266 CCEB CoS Navin Kumar Verma The DEAD-box helicase DDX3X in cancer
ATP-binding? and the ?helicase C-terminal? domains of the gene. We hypothesize that the types and positions of mutations within the DDX3X gene may inform on the nature of mutagenic mechanisms 021-01437-0
involved in tumor development, progression, and aggressiveness. The aim of this project is to summarize current knowledge on DDX3X gene variations observed in human cancers. The study will
uncover tumor-type specific mutations in the DDX3X gene, which will contribute to better understanding of the significance of DDX3X as a target for therapeutic strategies.

Generative Artificial Intelligence (GenAI) is transforming online labor markets by enhancing worker capabilities and altering work demand. Despite its rapid adoption, the impact of GenAI on matching
efficiency?the ability of platforms to effectively pair employers with workers?remains underexplored. This study investigates how GenAI affects the employer?worker matching process, focusing on AI-
assisted proposal creation. While access to GenAI may improve proposal clarity and increase the likelihood of a successful match, it can also obscure workers' unique attributes, potentially hindering
effective matching. To identify these dual effects, we design controlled lab experiments that vary GenAI access, allowing us to track workers' behavior and output, as well as employers' decisions
267 ITOM Div CoB (NBS) Jiayu Yao The Impact of GenAI on Online Labor Markets
throughout the matching process. We further analyze how these dynamics differ between creative-intensive and routine-intensive jobs. This research offers an understanding of GenAI?s role in the
matching process, highlighting its potential to reduce information asymmetry while simultaneously weakening signaling effectiveness.
This project is well suited for students interested in Information Systems, Business Analytics, Economics, Psychology, Human?Computer Interaction, and the Future of Work. Students will gain hands-on
experience in experimental design, behavioral data analysis, and AI-related research.

The gut microbiota plays a crucial role in human health, actively interacting with the host to influence numerous biological processes and disease outcomes. Extensive research has uncovered its
complex interactions with the digestive system, and alterations in this microbial community has been linked to a range of disorders, including inflammatory bowel diseases and colorectal cancer. A key
mechanism of microbiota-host interaction is through bacterially derived short-chain fatty acids (SCFAs), primarily butyrate, acetate, and propionate. While the effects of SCFAs on the microbial
The Impact of Gut Bacterial-Produced Short-
268 LKCMedicine LKCMedicine Yi Liu community itself are well-documented, their specific impact on host intestinal epithelial cell plasticity and lineage composition remains poorly defined. This study proposes to address this gap using [Link]
Chain Fatty Acids on Colon Epithelial Lineages
colon organoids derived from a spectrum of donor tissues, ranging from healthy to metastatic colorectal cancer (CRC) harboring somatic mutations. We will investigate how specific SCFAs alter
epithelial cell-type compositions (lineages) within these models. The outcomes of this research will advance our understanding of microbiota-epithelial interplay and provide new insights into the role
of gut microbial metabolites in intestinal homeostasis and cancer dynamics.

Halide perovskite emitters have garnered significant interest for their applications in optoelectronic devices, particularly in light-emitting diodes (LEDs), lasers, and display technologies. These materials
The Science of High INtensitY (SHINY) Perovskite exhibit exceptional photophysical properties, making them ideal candidates for next-generation light sources as well as scintillators. This project explores the synthesis, characterization, and
269 IAS IAS Sum Tze Chien [Link]
Emitters* optimization of perovskite materials to engineer stable, highly luminescent next-generation emitters. The student will investigate the photophysics of these systems using a range of state-of-the-art
optical spectroscopy techniques. Importantly, The student will have opportunities to work in a fast-paced frontier research field and interact with a highly dynamic research team.
This study examines the impact of synthetic music on human decision-making and task performance. Building on existing music and acoustics literature, the study identifies key structural and
perceptual aspects of music and examines the underlying psychological mechanisms that shape responses to AI-generated sound. Multiple experiments employing computer vision, galvanic skin
The Sound of AI: Exploring Synthetic Music
270 ITOM Div CoB (NBS) Ben Choi response (GSR), and eye-tracking will be conducted to capture cognitive, emotional, and behavioral effects. Additionally, transitional work will focus on developing prototype systems that integrate
Beyond Curation
synthetic music into interactive contexts. Together, this research advances our understanding of how AI-driven music moves beyond curation to shape human experiences actively, offering insights for
both theory and application in human?AI interaction.
Almost all photosynthetic CO2-fixation is mediated by the slow and non-specific carboxylase Rubisco. Our group tries to understand how plants and algae have compensated for these problematic
properties. A key strategy used in almost all algae involves the condensation of Rubisco into dense protein droplets, which are then flushed with CO2 gas. This is known as a biophysical CO2
concentrating mechanism. In the past we have used purified proteins to reconstitute in vitro the Rubisco droplets of diverse algae [1,2]. This happens by a mechanism known as liquid-liquid phase
separation. We have recently discovered multiple additional Rubisco binding proteins that we believe contribute to the phase separation of Rubisco in the diatom Phaeodactylum tricornutum. In this
project we will focus on one of our putative Rubisco linker proteins, purify it, and characterize its properties. The project will involve protein biochemistry and fluorescence microscopy.
Oliver Martin
271 SBS CoS The stickers and spacers of Rubiscondensation [Link]
Mueller-Cajar References:
Oh ZG, Ang WSL, Poh CW, Lai SK, Sze SK, Li HY, Bhushan S, Wunder T, Mueller-Cajar O. A linker protein from a red-type pyrenoid phase separates with Rubisco via oligomerizing sticker motifs. Proc Natl
Acad Sci U S A. 2023 Jun 20;120(25):e2304833120. doi: 10.1073/pnas.2304833120. Epub 2023 Jun 13. PMID: 37311001; PMCID: PMC10288592.

Wunder T, Cheng SLH, Lai SK, Li HY, Mueller-Cajar O. The phase separation underlying the pyrenoid-based microalgal Rubisco supercharger. Nat Commun. 2018 Nov 29;9(1):5076. doi: 10.1038/s41467-
018-07624-w. PMID: 30498228; PMCID: PMC6265248.
In this project, we are interested in developing a multi-scale understanding of biopolymers that can explain how different levels of biological organization can arise in a hierarchical manner. We plan to
utilize emerging geometrical and topological data analysis tools such as persistent homology, Betti numbers, Ricci curvature, and Hodge Laplacian to extract meaningful geometrical and topological
272 PAP Div CoS Yong Ee Hou Topological Data Analysis of G-quadruplex information, with the goal of uncovering previously unknown features. These persistent structural features are related to biological functions of the biopolymers, from which we can better understand [Link]
structure-activity relationship as well as structure-function relationship.

Enormous power has been demonstrated by geometric deep learning (GDL) in molecular data analysis. In particular, the recent great success of AlphaFold 3 shows the great advantage of GDL in
analyzing biomolecular structural information. However, there still remains great challenges in traditional GDL. In particular, its failure to characterize high-order interactions/information pose a major
issue for its applications in AI for sciences.

Topological Deep Learning (TDL) leverages novel topological tools to characterize data with complicated higher-order structures. Different from graph-based data representation, TDL uses topological
representations from algebraic topology, including simplicial complexes, cell
Topological deep learning for material data
273 MAS Div CoS Xia Kelin complexes, sheaves, hypergraphs, and combinatorial complexes to model not only pair-wise interactions (as in graphs), but also higher-order interactions among three or more elements. In fact, these [Link]
analysis
algebraic topology-based molecular representations have already achieved great success in molecular data analysis, including protein flexibility and dynamic analysis, drug design, virus analysis,
materials property analysis. Mathematically, TDL uses a generalized message-passing mechanism thus enables the communication of information from simplices of different dimensions. In contrast to
GNNs, where information is passing among nodes or edges, TDL allows information to propagate through any neighborhood relations.

In this project, we will study the TDL-based models for material data, from Material Project, Javis, etc. We will focus on the incorporation of multiscale and periodic information into TDL models. We
will also design special novel TDL architectures that can be aware of the general physical laws, i.e., physics-inspired TDL models.

This project working in collaboration with National Dental Centre of Singapore addresses the growing need for advanced patient-specific dental implant solutions by integrating topological optimization
Paulo Jorge Da Silva Topological optimisation and metal 3D printing and metal 3D printing to enhance implant performance. In clinical practice, dental implants must withstand complex biomechanical forces while promoting osseointegration and long-term stability. By
274 MAE CoE
Bartolo of dental implants optimising implant geometry through computational modelling, this approach aims to reduce stress shielding, improve load distribution, and enhance biocompatibility. Metal 3D printing enables the
fabrication of intricate structures that mimic natural bone, potentially leading to improved healing, reduced failure rates, and better overall clinical outcomes.
Underground material property changes can evolve into destructive events such as rock fracturing, cavity collapse, fluid leakage, and earthquakes. Scattered seismic waves are widely used to monitor
Tracking Spatial Elastic Material Property subtle material property variations in both laboratory and field settings. However, the spatial resolution of such observations is strongly constrained by survey coverage and assumptions about wave
275 CEE CoE Yuan Congcong
Variations via Scattered Acoustics types, limiting precise localization of evolving changes. To address these challenges, we invite visiting students to develop a novel time-lapse seismic methodology capable of accurately inferring the
spatiotemporal evolution of elastic material properties.
Plant cells accumulate oil in the form of triacylglycerol (TAG) in their seeds as a main resource of carbon and energy for seedling development. Plant oils are not only essential for the human diet but
also provide renewable feedstocks for many industrial uses.

Transcriptional regulation of plant oil Research in my lab focuses on mechanistic understanding of transcriptional regulation of plant oil biosynthesis and translation of knowledge gained from our research into benefits for the
276 SBS CoS Ma Wei [Link]
biosynthesis* agroeconomy.

Student will participate in the research projects in elucidating the regulatory mechanism of these essential plant oil regulators. Student will have an opportunity to learn molecular biology, genetics,
protein-protein interactions, developmental biology, and lipid biochemistry from the research work.
 The genus Parasponia has garnered scientific interest in the past few decades due to its unique ability to form nitrogen fixing root nodules with Rhizobia as the only species outside of the family
Fabaceae (Legumes). The simplistic method, through which the rhizobia infect Parasponia roots, is believed to have a simple genetic background and hence be inducible in other plants, including crop
species, using genome editing techniques. Parasponia is closely related with the genus Trema, a group of physically very similar but non-nodulating species. Among Trema, Trema tomentosa is a unique
species as it is possibly an allopolyploid between Trema orientalis and a non-nodulating Parasponia-like species. Studying Trema tomentosa could, hence, allow a better understanding of the differences
between the nodulating Parasponia and non-nodulating Trema. However, research into Trema tomentosa has been limited by the poor genome assembly the species has.
Trema tomentosa genome assembly as a tool for
277 SBS CoS Jarkko Salojarvi
understanding nodulation in Parasponia In this project, you will produce a high-quality genome assembly for Trema tomentosa. You will first extract High Molecular Weight DNA from the species and send part of it to be sequenced with
PacBio HiFi sequencing, known for its high quality. In the meantime, you will also perform an ultra-long read library prep for Nanopore and run the resulting ultra long reads with PromethION sequencer
present at the lab. You will then extract more HMW DNA and produce an Omni-C DNA library to be short-read sequenced. Using these resources, you will then carry out a genome assembly utilizing the
HiFi reads, the ultra-long Nanopore reads and the Omni-C reads to produce a high-quality genome assembly for Trema tomentosa. You will then phase the assembly to separate the parental
subgenomes and use mapping of reads from Trema orientalis and Parasponia species to verify the subgenome origins.

This project aims to develop an innovative deep learning-based framework for melanoma skin cancer diagnosis that combines cutting-edge classification accuracy with enhanced interpretability
through the ABCD rule (Asymmetry, Border, Color, and Dermoscopic structures). While deep learning models have demonstrated exceptional performance in detecting melanoma from dermoscopic
images, their "black-box" nature limits clinical adoption due to a lack of explainability and trustworthiness. To address this gap, our approach integrates explainable artificial intelligence (XAI)
Trustworthy Deep Learning for Melanoma
278 LKCMedicine LKCMedicine Fan Xiuyi techniques to align model predictions with established clinical diagnostic criteria.
Diagnosis Using the ABCD Rule
The proposed framework comprises two key components: a robust convolutional neural network (CNN) for classifying skin lesions and an explainability module that visualizes and quantifies the
model's decision-making process based on ABCD features. This approach bridges the gap between AI-driven automation and human expertise, offering a transparent, trustworthy diagnostic tool.

We are interested in developing novel reinforcement learning (RL) and assessment methods to enable the design of trustworthy control systems in cyber-physical computing. In particular, we are
279 CCDS CCDS Arvind Easwaran Trustworthy Reinforcement Learning interested in developing programmatic policies for RL, deriving sample complexity bounds for efficient learning, as well as develop formal verification techniques for them. We are also interested in the [Link]
application of such methods in the domain of multi-energy systems, particularly for control/coordination of building energy systems.
This project will develop Machine Learning System (MLSys) technologies for trustworthy and explainable AI using large language models and federated learning with a wide range of applications ranging
280 CCDS CCDS Chee Wei Tan Trustworthy and Explainable AI for MLSys from health informatics to combat epidemics to AI-assisted software engineering for 6G cybersecurity. The project deliverables are either MLSys algorithm and software implementation or else [Link]
fundamental machine learning theory for MLSys.
This project will develop Machine Learning System (MLSys) technologies for trustworthy and explainable AI using large language models and federated learning with a wide range of applications ranging
281 CCDS CCDS Chee Wei Tan Trustworthy and Explainable AI for MLSys (II) from health informatics to combat epidemics to AI-assisted software engineering for 6G cybersecurity. The project deliverables are either MLSys algorithm and software implementation or else [Link]
fundamental machine learning theory for MLSys.
The quantum confinement and exotic physics in two-dimensional materials and their heterostructures make them appealing candidates for constructing electronic and optoelectronic devices with
Two dimensional heterostructures for quantum
282 EEE CoE Song Peng unprecedented performances. This project will explore those materials for quantum optical devices that have the potential to achieve extremely high energy conversion efficiency. In detail, the student
optical devices
will work on the fabrication, and characterization of a type of optic device based on 2D materials for bulk photovoltatic applications.

Quantum computing stands as a transformative paradigm in computational capabilities, holding the promise to unlock solutions for problems that have long challenged classical computers. A
significant step in the quantum revolution was the realization of Quantum Supremacy?, marking the point where quantum computers outperform their classical counterparts in solving specific tasks.
However, this remarkable achievement also unveiled a formidable challenge: the need to scale up quantum computers by increasing the number of entangled qubits?. The manipulation and readout of
these qubits have typically relied on separate interconnects, giving rise to a significant thermal load on cryogenic systems, which ultimately imposes limitations on the scaling-up process?. Using optical
links to connect qubits to the room temperature environment has been investigated as an approach to mitigate the thermal load of RF interconnects, however, the added noise of optical read-out is
Mohammad Ultrafast Metadevice-Circuit QED Interface:
283 EEE CoE significantly higher than that can be achieved with microwave electronics?. [Link]
Samizadeh Nikoo Towards Scaled-up Quantum Computers
To address this important challenge, this projects centres on the development of high-speed, energy-efficient cryogenic electronics to be able to directly control and read out qubits without relying on
significant number of radiofrequency interconnects. We aim to use recently proposed electronic metadevices? on Gallium Nitride (GaN) at cryogenic temperatures, to establish a high-performance
interface with qubits based on circuit quantum electrodynamics (QED). This research project paves the way towards the realization of scaled-up quantum computers, as it introduces integrated high-
performance electronics capable of directly controlling and reading out qubits.

Hollow-core fibers guide light through their central hollow region. This unique ability to confine and transmit light along a narrow channel makes them an ideal platform for facilitating light-matter
interactions. By filling the hollow core with matter particles, a significant overlap between light and matter can be achieved over an extended interaction length. Over the past decade, numerous
studies have demonstrated that pumping gas-filled hollow-core fibers with femtosecond laser pulses can produce a range of fascinating phenomena, including pulse compression to the sub-cycle
Ultrafast light-matter interactions in hollow-core regime and nonlinear frequency conversion into the deep-ultraviolet and mid-infrared spectral regions.
284 EEE CoE Chang Wonkeun
fibers
This project aims to establish a novel fiber-based platform for investigating ultrafast light-matter interactions in hollow-core fibers and to explore new physics that are inaccessible in conventional
experimental settings.

The future of humanity relies on the exchange of unprecedented amount of data. In response to this demand, integrated telecommunication circuits must operate at higher frequencies, reaching the
terahertz band, to provide enhanced bandwidths and data rates. However, the performance of semiconductor devices degrades at frequencies above 100 GHz, which limits the capacity of wireless
Ultrahigh-speed Mm-wave and Terahertz networks. Electronic metadevice is a new concept inspired by optical metamaterials, which was first demonstrated in form of switches that could challenge the limitations of traditional semiconductor
Mohammad
285 EEE CoE Electronic Metadevices for 6G devices [1]. This project aims to experimentally investigate some the critical aspects of electronic metadevices, and to build integrated circuits based on this new technology, which can offer a solution [Link]
Samizadeh Nikoo
Telecommunications for 6G telecommunication networks. The selected candidate will be granted access to cutting-edge facilities at NTU for the fabrication of 6G telecom chips and conducting experiments.

[1] Samizadeh Nikoo, M. & Matioli, E. Electronic metadevices for terahertz applications. Nature 614, 451?455 (2023).

As we age, our cells lose their ability to cope with stress, a process that is linked to many age-related diseases. Our lab investigates how a key cellular stress sensor, IRE1, helps cells maintain a healthy
balance, known as homeostasis. While IRE1's role during acute stress is well-known, we've recently uncovered its crucial function in day-to-day cellular maintenance.

This project will focus on the fascinating connection between aging and a specialized form of cellular recycling called ER-phagy. You will get hands-on experience in a dynamic research environment,
using cutting-edge techniques to explore how a decline in IRE1 activity in aging cells leads to a breakdown in ER-phagy.

What you will do:

Uncovering the Link Between Cellular Aging and
286 SBS CoS Guillaume Thibault 1. Work with human epithelial cells to model cellular senescence. [Link]
Stress
2. Use advanced microscopy to visualize and quantify the defects in cellular organelles.
3. Perform gene knockouts (using CRISPR-Cas9) and overexpressions to dissect the role of key proteins.
4. Analyze protein and gene expression levels to support your findings.

This project is a perfect fit for a student interested in cell biology, aging, or molecular mechanisms of disease. You'll gain valuable skills in experimental design, data analysis, and scientific
communication, all while contributing to a high-impact research project.

You will work together with an experience team of experts.

Genetics, lifestyle, and behavioral risk factors are important determinants of cancer risk. Our team works on profoundly understanding the contributions genomics play in cancer predisposition in Asia
with the following ongoing projects
Joanne Ngeow Yuen 1) Understanding genetic, lifestyle, and behavioral risk factors for childhood cancer and early onset cancers in Asia
287 LKCMedicine LKCMedicine Understanding Cancer Predisposition in Asia [Link]
Yie 2) Lifestyle and reproductive history modifiers of known cancer predisposition syndromes comparing Asia with Europe/ America datasets
3) Clinical utility of long-read sequencing for hereditary cancer testing
4) Improving access to genomic medicine in Asia: understanding and overcoming systemic barriers to testing
 In this project, we will collect and annotate a set of meme images that require cultural backgrounds to understand. If time allows, we will explore the training of vision-language models to understand
Understanding Cultural Memes in Low-resource such memes.
288 CCDS CCDS Li Boyang
Languages
We will focus on languages other than Western languages or Mandarin Chinese, which already have plenty of resources. Working knowledge in such languages would be an advantage.

The immune system plays a key role in wound healing. Regulating immune function in the wound microenvironment is an attractive approach to combat excessive inflammation and promote healing.
Understanding Immunomodulatory Impact of Unfortunately, delayed wound healing remains a challenge. Existing therapies have not fully solved complications arising from immune dysregulation. Advances in bioengineering have led to the
289 CCEB CoS Navin Kumar Verma Bioengineered Scaffold Topologies in the Skin development of 3D scaffolds mimicking dermal matrix supporting wound healing. We and others have demonstrated that scaffold fibres arranged in a parallel alignment improve healing; however, the
Wound mechanisms are not fully understood. We hypothesize that the topologies of the scaffolds implanted on the wound influence both innate and adaptive immune responses in the wound environment.
Our goal is to understand how fibre alignments in the scaffolds impact tissue-resident/infiltrating immune cells that guide wound repair and healing.

Acquisition of mutations in the tumor suppressor gene, p53, is the most common event across all cancer types. While un-mutated wild-type p53 exerts its tumor-suppressive functions through the
regulation of various process, mutations in p53, which in over 90% of the cases occur in the DNA-binding domain, essentially abrogate many of these functions, leading to loss of activity. As a
conventional tumor-suppressor, mutations in p53 lead to loss of its tumor-suppressive functions. However, mutant p53 is unique amongst all tumor suppressors in that it acquires novel oncogenic
290 SBS CoS Kanaga Sabapathy Understanding Mutant p53 Gain of Function ?Gain-of-function? (GOF) properties. As such, mutant p53 is often associated with aggressive cancers, metastasis and resistance to various forms of therapies. However, when and how mutant p53 [Link]
gains GOF is an unresolved mystery, as not all p53 mutants exhibit this activity. The goal of this project is to start deciphering at which point in the journey of a cancer cell does mutant p53 gains its
oncogenic functions. Using cellular systems and biochemical approaches, the goal is to evaluate the temporal sequence of events leading to the mutant p53 GOF. Students will employ cell culture and
biochemical and molecular biology techniques during the project. Some prior experience is advantageous.

The protein cage, Ferritin, is being explored as components for molecular electronics. The exploration begins by investigating the charge transport of the Ferritin which requires its deposition on solid
substrate, such as silicon. However, how the Ferritin interacts with silicon surface is not well understood. How many Ferritin can be deposited? What are the best conditions to ensure deposition?
Understanding the Deposition of Protein Cages
291 CCEB CoS Lim Sierin
on Solid Surface The objective of this project is to determine the best conditions for deposition of Ferritin on silicon surface.

Preferred skill-sets: bacteria culture, protein purification

Enzymatic polymer degradation is being explored as a possible complement for the existing mechanical and chemical recycling approaches. The enzyme PETase, which has natural ability to
depolymerize PET, has been engineered for better performance. However, the interactions between the PET polymer and the enzyme activ is not fully understood.
Understanding the Interaction between the
292 CCEB CoS Lim Sierin Polymer-degrading Enzyme PETase and Its
The objective of the project is to understand the interaction between the substrate, PET, and the enzyme, PETase, using molecular dynamics simulation.
Substrate
Prefered skillsets: PyMOL
How are emotions represented in the brain and in deep neural networks? The Nair lab at LKCMedicine investigates how neural circuits in the brain and AI models control and represent emotion states,
including aggression and fear. This project will utilize new advances in explainable and interpretable AI to reveal hidden neural computations in high-dimensional brain data and in foundation AI models
Understanding the neural computation of
including vision transformers. Building on previous efforts from the lab (see Nair et al., Cell 2023, Vinograd*, Nair* et al., Nature 2024 and Nair et al., Science 2025), the student will work at the
293 LKCMedicine LKCMedicine Aditya Nair emotion in foundation AI models and brain [Link]
intersection of neuroscience and machine learning to unravel the degree of similarity in how deep networks and brain circuits process emotion. This project has the potential to advance the
circuits
development of ?digital twins? of the brain using deep neural networks, which could eventually enable personalized therapies based on individualized brain models capable of accurately predicting
activity under disease and disorder states.

Enhanced Rock Weathering (ERW) is an emerging strategy for climate mitigation that involves the application of finely ground silicate minerals to land, where they react with CO? to form stable
carbonate minerals. This project aims to explore the potential of different silicate minerals, specifically wollastonite, forsterite, and natural volcanic ash, to improve soil quality and increase plant
productivity through pot experiments.

The student will have the opportunity to participate in controlled experiments where these minerals will be applied to various soil types, and their effects on plant growth and soil health will be
Unlocking Climate Solutions: The Role of Silicate
carefully monitored. The project will focus on assessing the rates of silicate mineral dissolution and their capacity to sequester carbon through chemical weathering processes.
294 ASE CoS Simon Redfern Minerals in Enhanced Rock Weathering and Soil
Fertility
Advanced analytical techniques will be employed to measure the dissolution rates of these minerals in soil environments. These measurements will be correlated with changes in soil pH, nutrient
availability, and carbon content to evaluate the potential of ERW for both enhancing soil fertility and achieving significant carbon drawdown.

Additionally, the project will investigate how these mineral amendments influence plant productivity, providing insights into the broader environmental benefits of ERW beyond carbon sequestration.
The results of this research could contribute to the development of scalable strategies for climate mitigation while simultaneously enhancing agricultural productivity.

This project aims to investigate the fundamental role of water structure at the charged interfaces of single-atomic catalysis. The catalytic activity of single metal atoms immobilized on surfaces is highly
dependent on their electronic structure and their interaction with water molecules in the surrounding environment. Understanding the effect of potential on the water structure is crucial for unraveling
the mechanism of single-site catalysis.
Unraveling the Impact of Applied Voltage on
295 MSE CoE Li Shuzhou This study will employ state-of-the-art computational techniques to explore the interplay between water structure and applied potential in the vicinity of a catalytic site. The effects of various
Water Structure in Single-atomic Catalysis
potentials on water structure and the reactivity of the catalytic site will be examined systematically, providing insights into the underlying physicochemical processes governing single-site catalysis.
This research will advance our understanding of the role of water structure at charged interfaces in catalytic site reactivity, with broad implications for fields ranging from materials science to biology.

Chirality is a ubiquitous structural property determined by the non-super-imposability of mirror image of an object. Simplest chiral objects are helical screws which have a handedness in the turns.
Unraveling the role of short range order in
Materials can be designed at the atomic scale to have a helical pitch in the arrangement of atoms. Determining the handedness of the helix at the atomic scale is non-trivial since electron diffraction
296 MSE CoE Prashant Kumar symmetry breaking at the nanoscale through [Link]
patterns have subtle differences for a left-handed crystal vs a right-handed crystal. This project will provide the student with hands-on experience on the TEM to acquire and analyze the electron
automated imaging
diffraction pattern in order to create a 3D reciprocal space map of a chiral crystal, thereby enabling structure solution of such materials.

As we age, the tissues that protect our bodies, like the lining of our gut, begin to decline. This is driven by a process called cellular senescence, where our cells get old and stop functioning correctly.
Inside each of these aging cells, tiny structures called organelles begin to break down, but this process is a "black box"?we know it happens, but we lack the tools to see it in detail.

This project aims to build a powerful new platform to watch this breakdown happen in real-time, using 3D "mini-guts" (intestinal organoids) that we grow in the lab from human stem cells.

Your Role & What You'll Learn

This is a highly multi-disciplinary project at the cutting edge of cell biology, microscopy, and artificial intelligence. As an FYP student, you will be an integral part of the team and will gain hands-on
experience in a range of state-of-the-art techniques.

Using Next-Generation Microscopes and AI to You will:

297 SBS CoS Guillaume Thibault
Understand How Our Cells Age 1. Learn to grow and maintain human induced pluripotent stem cells (iPSCs) and differentiate them into complex 3D intestinal organoids.
2. Employ CRISPR/Cas9 to genetically engineer cells, making specific organelles like the endoplasmic reticulum (ER) glow with fluorescent proteins.
3. Gain experience with advanced imaging techniques, including confocal microscopy and our collaborator?s novel label-free interferometric scattering (iSCAT) microscope.
4.: Work alongside our team and collaborators in physics and computer science to analyze complex image datasets that will be used to train an Artificial Intelligence (AI) model to predict how cells are
aging.

We are looking for a highly motivated and curious student with a strong interest in fundamental cell biology and cutting-edge technology. A background in cell biology, molecular biology, or genetics is
preferred. No prior experience with advanced microscopy or AI is necessary?just a passion for discovery and a willingness to learn!

The student will work together with an experience team of experts.

 Autonomous mobility in sea, land and air is important demand in industry and society. So far, a lot of efforts has been dedicated to the achievement of autonomous mobility in land. Future challenge
Vision-Guided Planning and Control by
298 MAE CoE Xie Ming and competition will be toward achieving autonomous mobility in sea. The objective of this research project is to investigate vision-guided planning and control which will enable autonomous surface [Link]
Autonomous Surface Vehicle
vehicles or under-water vehicles to navigate in a complex water area in which there are presence of many other fixed or mobile targets.
Humanoid robot is gaining astonishing level of popularity in recent years. For humanoid robots to be widely acceptable in industry and society, it is necessary to make these robots to achieve human-
Vision-Guided Planning and Control of Motion by
299 MAE CoE Xie Ming like intelligence. The objective of this research project is to investigate vision-guided planning and control of motions which will enable a humanoid robot to achieve autonomous biped walking in [Link]
Humanoid Robot
complicated environments, autonomous grasping of soft objects, and autonomous manipulation of complex objects.
Money laundering is widely seen in existing cryptocurrency transactions. Prior research has been focused on developing automated approaches to detect money laundering behaviours in
Visual Identification of Money Laundering in cryptocurrency transactions. However, such approaches are often unable to be adaptive enough to identify various and evolving money laundering behaviours in the increasingly popular
300 CCDS CCDS Wang Yong
Cryptocurrency via Human-AI Collaboration cryptocurrency transactions. This project aims to combine AI and human domain knowledge and develop a novel visual analytics approach to allow users to interactively identify money laundering
behaviours in a scalable and adaptive manner.
Despite the excellent performance of Large Language Models (LLMs), they still suffer from the lack of explainability and other critical issues like haullucination. This project aims to develop novel
Visualization for Explainable Large Language
301 CCDS CCDS Wang Yong visualization techniques to enhance the explainability of LLMs, making it easier for common users to easily understand how LLMs work. Hopefully, it can also benefit the mitigation of hallucination of
Models
LLMs.
The Golgi is a subcellular organelle that functions as a trafficking hub within a eukaryotic cell. It comprises serially stacked membrane sacs called cisternae, roughly divided into four zones ? the cis,
medial, trans-Golgi, and trans-Golgi network. However, it is still unclear how the Golgi maintains its unique organization and how cargos, such as proteins and lipids, move within the Golgi. One of the
Visualizing the dynamic cargo movement within
302 SBS CoS Lu Lei greatest challenges is that the details within the Golgi are beyond the resolution of conventional and even super-resolution light microscopy. We have recently developed novel super-resolution [Link]
the Golgi
methods and demonstrated their capabilities in visualizing sub-Golgi organization and dynamics (Tie et al., 2016; Tie et al., 2018; Tie et al., 2022). This project will quantitatively examine the intra-Golgi
trafficking of secretory and endocytic reporters using our cutting-edge microscopic tools. It involves cell culture, transfection, microscopic imaging, and image analysis.

This project aims to offer a sustainable solution to biomass waste utilization for great sustainability. The solid and gaseous wastes we produce from consumption of plants (biomass) go into the
environment as carbon. The carbons in the air and soil are utilized by plants and turn into biomass and fuel. In this carbon cycle, biomass provides renewable fuel that help to mitigate greenhouse
problem. It utilizes biodegradable waste to produce fuel resources and energy, which doesn?t result in pollution as burning fossil fuel does. Cellulose, an important structural component of the cell wall
303 MAE CoE Li Hong Waste to Green Hydrogen and Green Chemicals of green plants, is the most abundant biopolymer. Therefore, utilization of cellulose is an important part of biomass refinery. Cellulose can be converted to sugars, alcohols, or other chemicals. The [Link]
main challenge of cellulose refinery originates from its stubborn crystalline structure that typically needs harsh process conditions including high temperature, high pressure, and high acid
concentration to break it down to useful products. In this project, we will explore a green and cost-effective process for biomass waste (e.g., sawdust and seafood waste) upcycling to biofuels, green
hydrogen, and green chemicals.

This research project focuses on the study and optimization of water electrolysis, a process that splits water into hydrogen and oxygen using electrical energy. The project aims to investigate various
304 MSE CoE Jason Xu Zhichuan Water electrolysis for hydrogen production catalysts and electrode materials to enhance the efficiency and cost-effectiveness of hydrogen production. Students will gain hands-on experience with electrochemical techniques, material [Link]
characterization, and data analysis. This project is ideal for students interested in renewable energy technologies and sustainable engineering solutions.

Accurate weather and climate models are, for the most part, based on physical principles and lead to the formulation of differential equations whose solution is very expensive to compute numerically.
On the other hand, machine learning models use expressive data-based functional families that are computationally feasible and describe some qualitative behaviors but are hard to interpret and
eventually violate physical principles. These features have motivated the emergence of hybrid or physics-informed machine learning models in a larger field called theory-guided data science [1]. That
research aims to systematize the use of scientific first principles in constructing machine learning algorithms.
Weather and climate forecasting using hybrid
305 SPMS CoS Juan-Pablo Ortega In this project, we shall construct physics-informed weather and climate models in which the data-based part will be implemented by a paradigm called reservoir computing, and the knowledge-based
machine learning schemes
will be constructed using simplified versions of standard physical models. Successful implementations of this approach already exist in the literature [2]-[5], but all the methods that have been utilized
so far are, for the most part, purely empirical, and an adequate error analysis for them does not exist. Hybrid forecasting methods have, however, been extensively studied in statistics [6]-[54], and
statistical guarantees have been developed for them. This project specifically aims to implement the well-known hedge algorithm in [8]-[9] using families of reservoir systems combined with knowledge-
based models. This method provides a rational strategy to combine the modeling approaches and a framework for adequate error analysis.

This project is about applying eXplainable AI (XAI) for business intelligence on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that
perform various sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models
(such as semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns
and extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI
model from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

1. choose a popular topic, e.g., marketing strategies, customer segmentation, or brand positioning
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
306 CCDS CCDS Erik Cambria XAI for Business Intelligence [Link]
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report
The successful completion of this project consists of the following steps:
1. choose a popular topic, e.g., cryptocurrencies, marketing strategies, or brand positioning
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report

This project is about applying eXplainable AI (XAI) for climate resilience on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform
various sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

307 CCDS CCDS Erik Cambria XAI for Climate Resilience [Link]
1. choose a popular topic, e.g., sustainability, ESG, or extreme weather events
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report
 This project is about applying eXplainable AI (XAI) for education on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform various
sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

308 CCDS CCDS Erik Cambria XAI for Education [Link]
1. choose a popular topic, e.g., e-learning, student engagement, or personalized learning
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report

This project is about applying eXplainable AI (XAI) for finance on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform various
sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

309 CCDS CCDS Erik Cambria XAI for Finance [Link]
1. choose a popular topic, e.g., cryptocurrencies, stock market trends, or ESG investing
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report

This project is about applying eXplainable AI (XAI) for healthcare on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform various
sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

310 CCDS CCDS Erik Cambria XAI for Healthcare [Link]
1. choose a popular topic, e.g., mental health, telemedicine or aging population
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report

This project is about applying eXplainable AI (XAI) for online safety on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform
various sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

311 CCDS CCDS Erik Cambria XAI for Online Safety [Link]
1. choose a popular topic, e.g., cyberbullying, cyberharassment, or fake news
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report

This project is about applying eXplainable AI (XAI) for science on a topic of your choice. In particular, you will be using Sentic APIs, a suite of application programming interfaces that perform various
sentiment analysis tasks in different languages ([Link] All the APIs are based on the sentic computing framework and, hence, leverage an ensemble of symbolic models (such as
semantic networks and knowledge graphs) to encode meaning in an explainable way and sub-symbolic paradigms (such as foundation models and generative algorithms) to identify patterns and
extract insights from data. One of the goals of this project is also to test functionality, reliability, and performance of Sentic APIs. If you have time and interest, you can also explore building a XAI model
from scratch. For more information or inspiration, please check [Link]

The successful completion of this project consists of the following steps:

312 CCDS CCDS Erik Cambria XAI for Science [Link]
1. choose a popular topic, e.g., AI for social sciences, chemistry, or cognitive science
2. crawl and label data about such topic from X or Reddit in English or any other language
3. process the collected data using the API
4. report any encountered issues, e.g., mislabeled tweets
5. deduce possible reasons of such issues, e.g, missing concept in SenticNet
6. come up with possible solutions for the encountered issues
7. visualize and discuss results in your project report
Humans do not perceive the entire scenes at once. Instead, they move their eyes and pay attention to certain parts of the visual scenes. How do we come up with AI models capable of predicting
313 CCDS CCDS Zhang Mengmi intention inference using eye movements human eye movements? Students get to program computational models of eye movement predictions. Alternatively, students also get to learn how to use advanced eye tracking devices in the lab, [Link]
design eye tracking expeirments with programming languages, collecting human data, and analyzing these eye movement data.
Humans do not perceive the entire scenes at once. Instead, they move their eyes and pay attention to certain parts of the visual scenes. How do we come up with AI models capable of predicting
314 CCDS CCDS Zhang Mengmi multi-agent eye movement prediction human eye movements? Students get to program computational models of eye movement predictions. Alternatively, students also get to learn how to use advanced eye tracking devices in the lab, [Link]
design eye tracking expeirments with programming languages, collecting human data, and analyzing these eye movement data.
Humans do not perceive the entire scenes at once. Instead, they move their eyes and pay attention to certain parts of the visual scenes. How do we come up with AI models capable of predicting
315 CCDS CCDS Zhang Mengmi predicting eye movements with AI models human eye movements? Students get to program computational models of eye movement predictions. Alternatively, students also get to learn how to use advanced eye tracking devices in the lab, [Link]
design eye tracking expeirments with programming languages, collecting human data, and analyzing these eye movement data.

Report Generated on : 15-Jan-2026 08:37:49 V1.1