External Research Grants

This project builds upon algorithms previously developed by the Prof Lau Hoong Chuin under the Quantum Engineering Programme 2.0 initiative - specifically in addressing variants of the News Vendor and Knapsack Problems - to tackle the increasing complexity of consumer demand and fluctuating market dynamics in logistics. In collaboration with ST Logistics (STL), the project will develop a hybrid quantum-classical model capable of jointly performing demand forecasting and inventory optimization. The goal is to deliver a proof-of-concept (POC) solution with computational efficiency for complex, real-world logistics scenarios provided by STL.

The overall objective of this project is to develop systematic and rigorous ways of safeguarding foundation models, including large models such as large language models (LLMs) as well as large multi-modal models (LMMs), against state-of-the-art and future security attacks. While there have been many bandage-like mitigation approaches on mitigating security attacks on LMs, they are far from having a lasting effect. The reason is that these mitigation approaches are treating the symptoms rather than fixing the causes of the problems. The team aims to develop techniques and systems which can detect and defeat a variety of security attacks on large models, through either prompting, finetuning or instruction- tuning, for the goal of jailbreaking, or embedding backdoors. 

This research / project is supported by the National Research Foundation, Singapore and Infocomm Media Development Authority under its Trust Tech Funding Initiative.

Instructors today face increasing challenges in designing and delivering courses that effectively balance cognitive load, align with intended learning outcomes, and actively engage diverse learners. Traditional lecture slides and assessments often lack structure, personalization, and interactivity leading to passive learning, reduced motivation, and inconsistent achievement of educational goals. Furthermore, evaluating and improving teaching delivery remains largely subjective, with limited tools to analyze real-time classroom engagement or instructional clarity. This project offers AI-driven analysis and nudging mechanisms that align content with learning objectives through semantic and topical modeling, while embedding cognitive design strategies to manage learners’ mental load effectively. This research addresses these challenges by exploring how AI and analytics can enhance course content design, assessment, and delivery in a data-informed and scalable manner.

Eye tracking has emerged as a powerful, non-invasive window into neurological and ocular health, offering early biomarkers for conditions such as Parkinson’s disease, Alzheimer’s disease, and glaucoma. However, current RGB camera–based systems are bulky, power-intensive, and limited in their ability to capture the subtle, high-frequency micro-movements of the pupil that are critical for early diagnosis. To overcome these limitations, this project introduces SynapSee, a novel end-to-end wearable system that integrates event cameras with a multi-light active probing setup and computationally optimised algorithms for real-time, fine-grained pupil tracking. Unlike conventional eye trackers, event cameras operate at sub-microsecond latencies and asynchronously capture changes in light intensity, making them uniquely suited for high-velocity saccades and micro-movements. By exploiting “dark” and “bright” pupil effects through multi-light probing, SynapSee reduces extraneous event volume, enabling low-power and efficient processing. The system is further enhanced by hybrid spiking neural networks, adaptive sensing algorithms, and collaborative offloading to nearby devices, achieving both accuracy and energy efficiency. We will validate SynapSee in two exemplar clinical contexts: (i) detecting early neurodegenerative changes in Parkinson’s disease and (ii) identifying the onset of low-vision conditions such as macular degeneration, cataracts, and glaucoma. Longitudinal user and patient studies, conducted in collaboration with clinical partners, will establish discriminative ocular biomarkers and benchmark the system’s sensitivity and specificity. By enabling unobtrusive, continuous, and large-scale monitoring via smart glasses, SynapSee has the potential to transform preventive healthcare, offering clinicians powerful tools for early intervention and personalised disease management.

This project focuses on creating self-adaptive embodied agents capable of perceiving and planning in dynamic real-world environments, addressing current challenges like hallucinated plans, poor object tracking and inflexible execution. It employs retrieval-augmented planning, fine-grained environment understanding, and adaptive plan refinement using large multimodal models, validated through simulations and real robots in household tasks. Expected outcomes include new methods for adaptive planning and perception, a kitchen activity video dataset, and demonstrations in domestic scenarios, with broad applications in autonomous vehicles and assistive devices. The initiative aims to impact daily living and healthcare, especially eldercare in Singapore, aligning with national priorities to enhance AI leadership and support the Smart Nation agenda.

This proposal presents OMNICON, a comprehensive framework for generating realistic long-term multi-human motions with environmental context. By designing novel motion representations with generative solutions, OMNICON addresses critical challenges in long-term motion generation, multi-human interactions, and motion-with-context synthesis. Designed to advance applications across animation, gaming, virtual reality, and robotics, OMNICON leverages principles from physics and spatial reasoning to produce temporally consistent, contextually adaptive, and socially coherent motion sequences.

This project, conducted in collaboration with HTX, explores the use of Generative AI (GAI) to advance scientific computing and strengthen cloud security and cybersecurity resilience. This project looks to address deep research challenges in building intelligent, domain-specific automation. This is done by leveraging LLMs for computational chemistry, cloud configuration security and developing robust defence strategies to protect AI systems for use in mission-critical settings.

The increasing realism and accessibility of AI-generated and AI-edited videos threaten public trust, information integrity, and digital security. From misinformation campaigns to identity fraud, such manipulated content can cause real-world harm. Current detection systems are limited: they often focus narrowly on facial deepfakes, lack cultural and linguistic diversity, offer little interpretability, and struggle to adapt to new manipulation techniques. Additionally, most systems emphasize passive detection, without offering mechanisms for content traceability or origin verification. This bilateral research project between Singapore Management University (SMU) and Sungkyunkwan University (SKKU) aims to address these challenges by developing an interpretable, adaptive, and globally deployable deepfake detection and protection system, tailored to the languages, dialects, and socio-cultural contexts of Singapore and South Korea. 

This research/project is supported by the National Research Foundation Singapore under the AI Singapore Programme (AISG Award No: AISG4-TC-2025-018-SGKR).

Singapore and New Zealand both use interRAI, a standardised assessment tool that supports the care of older adults. While interRAI is reliable and effective, integrating Artificial Intelligence (AI) presents a transformative opportunity to enhance healthy ageing and support older people to live longer, more independent lives. Our project brings together clinicians and researchers from the University of Otago, Singapore Management University, University of Canterbury, and University of Auckland. We will identify how to effectively integrate AI into the interRAI assessment, risk prediction, and care planning process to improve efficiency, consistency, and personalisation of care. We will achieve this with a three-pronged approach: 1. AI-assisted Assessments: By partially automating the currently manual interRAI process, we can reduce assessment time by 50% while improving accuracy. We will integrate structured health data and multimedia inputs to generate enriched assessments. 2. AI-enhanced Risk Prediction: We will develop predictive models for outcomes such as fracture risk, cognitive decline, and depression. These models will be embedded into interRAI software to support timely, targeted interventions. 3. AI-driven Personalised Care Plans: We will create dynamic, user-friendly care plans using a knowledge-based AI system enhanced by large language models. These plans will be tailored for patients, families, and clinicians, ensuring clarity and actionable guidance. With support from New Zealand’s Health NZ and ACC, and Singapore’s Agency for Integrated Care, Kwong Wai Shiu Hospital, NWC Longevity Practice, and 59 Socio-Techno Ventures, this initiative will augment existing systems to deliver scalable, cost-effective improvements to aged-care while growing our respective AI sectors. 

This research/project is supported by the National Research Foundation, Singapore under its AI Singapore Technology Challenge – Leveraging AI for Healthy Ageing (AISG Award No: AISG4-TC-2025-015-SGNZ). Any opinions, findings and conclusions or recommendations expressed in this material are those of the author(s) and do not reflect the views of National Research Foundation, Singapore.

Mild Cognitive Impairment (MCI), a precursor to dementia, offers a chance for intervention to delay dementia onset and facilitate advanced planning. In our ongoing project Sensors In-home for Elder Wellbeing (SINEW) with Sengkang General Hospital (SKH), we have the first longitudinal cohort in Asia from which clinically meaningful behavioural and digital biomarkers are captured with consumer-grade low-cost sensors installed in the homes of older adults. We have also successfully demonstrated proof-of-concept using sensors and machine learning to obtain digital phenotypes for accurate classification of MCI in a home-based setting. In this SINEW+ project, we shall build on the SINEW cohort to validate and refine our explainable predictive modelling approach for early detection of MCI. By leveraging advanced AI models and multi-modal data, we aim to implement and evaluate these solutions in real-world community settings, focusing on their cost-effectiveness and scalability to ensure broad adoption.