External Research Grants

AI for Program Reasoning is a research program that uses generative AI to advance the science of verifying that software behaves as intended. This programme is Co-Led by Prof Cristian CADAR from Imperial Global Singapore (IGS) and Prof Abhik Roychoudhury from National University of Singapore (NUS). It pursues three connected goals: (1) automatically generating program analysis tools for under-served domains such as scientific languages and floating-point-heavy code; (2) training large language models to genuinely reason about program verification rather than just predict next tokens; and (3) building AI agents that automate formal verification by inferring specifications and jointly repairing faulty code and failed proofs. SMU partners IGS and NUS in this research program and focuses on the second goal.

This research/project is supported by the National Research Foundation, Singapore under its Artificial Intelligence (AI)-for-Science (AI4S) Challenge (Award ID: NRF-AI4SCH-2025-0003).

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.

The HOPE (Human-Centered Operational Pandemic Resilience) programme aims to develop the next generation of AI-enabled decision systems that strengthen national preparedness and response to pandemics and other large-scale crises. Integrating Artificial Intelligence, Operations Research, and Cyber-Physical-Social (CPS) systems, HOPE seeks to transform crisis management from reactive response to proactive resilience.

The programme is organized around three pillars: Resilient Supply Chains and Logistics, Resilient Emergency Response, and Resilient Healthcare Operations, supported by a unifying CPS platform that integrates data, predictive analytics, optimization, simulation, and human-centered decision support. Through digital twins, learning-enabled planning, and multi-agent coordination technologies, HOPE will enable policymakers and operational agencies to evaluate interventions, allocate resources efficiently, and respond adaptively under uncertainty.

Working closely with government agencies, healthcare providers, and industry stakeholders, HOPE will translate research into deployable solutions that enhance societal resilience, protect lives and livelihoods, and establish Singapore as a global leader in AI-driven crisis preparedness and operational resilience.

This research aims to improve how delivery routes are planned in complex, real-world cities like Singapore, where traffic rules, road layouts, and constraints make routing much harder than simple map problems. The team is developing a new AI system that goes beyond current tools by (i) handling realistic and uneven travel conditions; (ii) understanding practical constraints (like delivery rules) more intelligently; and (iii) allowing users to simply describe their needs in plain language instead of using technical inputs.

The significance of this research is that it could make delivery and transport systems faster and cheaper, while also making advanced routing technology accessible to non-experts – helping smaller businesses and improving overall urban efficiency.

This project studies how large language models can help experts understand software when only the compiled program, or binary code, is available. Binary analysis is important for finding hidden security weaknesses, studying malware, and examining old or third-party software when the original source code cannot be accessed. It is also very difficult, especially when software has been deliberately disguised to resist inspection.

The project will first evaluate how well modern AI models perform on key binary analysis tasks. It will then develop two new AI powered tools. The first will help recover readable assembly code from heavily disguised software. The second will compare two binaries by generating pseudocode and identifying meaningful differences. Together, these tools aim to make software analysis faster, more accurate, and more effective.

The findings could help cybersecurity professionals detect malware and software vulnerabilities more efficiently, strengthen trust in software supply chains, and improve digital security more broadly. The project will also support education and future innovation through open source tools, datasets, and training opportunities.

This project proposes the development of GeneralGAD, a next-generation system for detecting anomalies in complex networks such as financial transactions, social interactions, and molecule networks. Unlike existing approaches that are often domain-specific and heavily reliant on large labeled datasets, GeneralGAD aims to operate as a general-purpose solution capable of identifying unusual patterns across diverse settings with minimal or no task-specific training data.

Technically, the project will integrate graph neural networks, which capture structural relationships in data, with large language models that provide contextual reasoning capabilities. This combination will enable the system to learn transferable representations of normal and abnormal behaviors across heterogeneous graphs and to generalize effectively to previously unseen domains. Building on promising preliminary results, the project will scale these models to enhance robustness, adaptability, and interpretability.

The potential impact is broad and tangible. GeneralGAD could transform how organizations detect fraud, mitigate cyber threats, and monitor harmful online activities, while also supporting discovery in scientific and medical research. By reducing dependence on costly labeled data and frequent retraining, the system offers a more efficient and scalable approach to safeguarding digital and real-world systems, ultimately contributing to greater security, trust, and innovation across industries.

PerFormRect aims to detect the misconceptions of students and utilize these insights in a feedback loop for crafting personalized formative assessment questions in programming education using Large Language Models (LLM). This project addresses Theme 4 (Leveraging Technology to Enhance and Personalize the Learning Experience) with efficient analysis of individual coding misconceptions and timely intervention through personalized coding questions for both practice and self-assessment. It tackles the pressing need to efficiently identify coding misconceptions of students and provide timely, personalized questions for both practice and self-assessment by students – tasks which are difficult for human instructors to perform at scale. PerformRect leverages on LLM to identify students’ misconceptions and generate personalized feedback (WP1) based on their code submissions. The identified misconceptions will also be used to generate formative assessment coding questions (WP2) tailored for their repetitive learning. The effectiveness of PerFormRect on students’ learning will be evaluated using randomized control trials (RCTs) in two institutions (for generalizability). In all, PerFormRect offers on-demand access, personalized feedback, tailored assessments and scalable programming skills development opportunities, allowing for a broader reach without sacrificing quality.

This project empowers educators to effectively leverage Generative AI as a transformative tool that not only identifies and addresses student misconceptions but also deepens understanding, fosters critical thinking, and enriches the overall learning experience. It proposes a novel AI-enhanced pedagogical approach called Debunkr, which utilises a cognitive conflict instructional approach to actively debunk misconceptions for university courses, potentially adaptable for any subject in the age of Generative AI.

TITAN 2.0 is a 2-year project to build an advanced AI-driven framework that identifies and fixes complex security flaws in software code. By combining traditional static analysis with the reasoning power of large language models (LLMs) orchestrated in an agentic fashion, the system extends analysis beyond individual functions to comprehend complex interactions across multiple files and modules. This allows it to catch interprocedural vulnerabilities that simpler LLM-powered tools often miss. Designed to support multiple programming languages such as Java, C#, Python, and JavaScript, the framework does not just flag risks; it acts as a digital security partner by providing automated CWE labeling, validated code patches, and developer-friendly reports. By integrating these smart agents, the project significantly improves vulnerability remediation to ensure that digital services remain secure and resilient.

This research / project is supported by the National Research Foundation, Singapore, and Ministry of Digital Development & Information under its Smart Nation and Digital Government Translational R&D Grant (Award No: TRANS2026-TGC01).

This research proposal focuses on the intersection of trustful artificial intelligence (AI) and the digital economy, aiming to develop frameworks and technologies that ensure AI systems are reliable, transparent, and beneficial to economic growth.

Hosted by the Singapore University of Technology and Design (SUTD), in collaboration with Singapore Management University (SMU), this project aims to create breakthroughs in wearable fabrics for health and wellness applications, with a special focus on tracking the mobility and joint motion of individuals with frailty challenges. The project aims to develop self-sustaining, energy-efficient smart textiles that seamlessly integrate sensors and electronics into knitted fabrics. The research collectively addresses two critical challenges: developing wearable fabric materials that can both sense movement and harvest energy, and creating ultra-low-power, on-board data processing mechanisms. The project combines all-knitted energy harvesting (using advanced yarns and knit architectures) techniques with ultra-low-power spiking neural network (SNN) based approaches for data processing, thereby maximising personal comfort while significantly extending the operational lifetime of the wearable sleeves. The research outputs will lay the foundations for scalable, long-term deployment of smart textile-based wearables for healthcare, rehabilitation, and preventive monitoring. Scientifically, the work shall generate globally competitive advances in materials science and neuromorphic computing for next-generation wearables; economically, it advances the marketability of textile-based wearables; and societally, it supports healthier ageing through quantified tracking of frailty and mobility-related impairments.

This research is supported by the National Research Foundation, Singapore under its 33rd Competitive Research Programme (NRF-CRP33-2025-0007).