(This is additional funding to SMU with a project extension.)
Funded by the Ministry of Law, this initiative, known as the Singapore Global Restructuring Initiative, is housed under the SMU Centre for Commercial Law in Asia, and it seeks to demonstrate international thought leadership on restructuring and insolvency.
This exploratory study investigates the impact of Singapore Management University’s (SMU) interdisciplinary Core Curriculum in developing lifelong learners. It focuses on three interrelated questions: (1) How do educators in SMU’s Core Curriculum approach interdisciplinarity, and (if relevant) lifelong learning, in their curricular design, course delivery and assessment?; (2) To what extent does SMU’s interdisciplinary Core Curriculum develop lifelong learning among its graduates?; and (3) What are the best practices in interdisciplinary education that develop lifelong learning that can be shared with Singapore IHLs?
This project tackles critical challenges in the development and deployment of Multimodal Large Foundation Models (MLFMs), which are capable of understanding and generating content across text, image, video, and audio modalities. While current MLFMs exhibit impressive performance, their trustworthiness, accuracy, and high operational costs limit their accessibility—especially for smaller research groups and organizations.
To address these gaps, the project focuses on two key innovations: (1) developing fine-grained "super alignment" techniques to reduce hallucinations and ensure model outputs align with human values, and (2) creating a scalable, low-cost multi-agent framework composed of smaller specialized models (8-15 billion parameters) that work collaboratively on complex tasks. These innovations will be powered by Reinforcement Learning with Human Feedback (RLHF) and Reinforcement Learning with AI Feedback (RLAIF), enabling continuous refinement and adaptation.
The research will be validated through real-world applications such as video generation and multimodal chatbots, demonstrating both practical utility and cross-domain adaptability. Ultimately, this work aims to democratize access to advanced AI, supporting Singapore’s strategic goal of building inclusive, trustworthy, and globally competitive AI capabilities.
This research/project is supported by the National Research Foundation, Singapore under its National Large Language Models Funding Initiative (AISG Award No: AISG-NMLP-2024-002). 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.
Aircraft surface inspection is typically done in an open environment, where the inspection model can be challenged by a lack of annotated defect samples, incomplete knowledge about possible defect types, and varying nature conditions or surface appearance. Vision-and-language foundation models have shown unique advantages in handling these challenges in various vision tasks. This project aims to develop innovative approaches to adapt such foundation models for addressing these challenges in aircraft surface inspection. The resulting models will help largely reduce aircraft maintenance cost and alleviate the risk of having unnoticed defects due to time/manpower limitation in aircraft inspection in Singapore.
Many economic studies seek to estimate the causal effect of a key variable on an outcome, but simple regressions often fail due to unobserved confounding factors. Instrumental variable (IV) regression helps address this issue by using instruments that affect the key variable but not the outcome directly. However, IV methods face several challenges: weak instruments, a large number of instruments and controls, and clustered data, all of which can impact the accuracy of inference. Previous research has addressed these issues separately, but this project aims to tackle them simultaneously, focusing on cases where many instruments have small but non-negligible effects rather than a few strong instruments. This work will enhance the reliability of IV regressions with complex data, improving causal inference in empirical studies.
This proposed project, FoCo, aims to develop Fast, cOmmunication-, and memory-effiCient Optimizers, specifically targeting the main identified problems of currently popular optimizers like Adam and AdamW in training large AI models, as follows:
a. FoCo will develop faster optimizers to reduce the training time and thus the cost of large AI models.
b. FoCo will design an Adaptive and COMpensate compression Approach called Acoma to reduce the communication costs of our faster optimizer and other optimizers like Adam.
c. FoCo will develop a memory-efficient approach called MeMo to lower the GPU memory of our faster and communication-efficient optimizer and other optimizers like Adam.
Although FoCo’s three main objectives focus on different aspects of training large AI models, they all work towards the common goal of making large AI training more efficient and faster. Improvements in one area will positively impact the others. Given the increasing importance and widespread use of large AI models, addressing their current training challenges is crucial. High training costs, long development times, and significant energy consumption and emissions are major concerns. By making AI training more efficient, FoCo will not only advance the field of AI but also contribute to a more sustainable and resource-efficient future. This project will benefit academia, industry, and society by enabling faster and more cost-effective development of advanced AI technologies.
The Intergovernmental Panel on Climate Change (IPCC) is a United Nations body tasked with assessing climate science and informing policymakers about climate change, its impacts, and options for adaptation and mitigation. Coordinating Lead Authors (CLAs) are critical actors in the preparation of the IPCC reports as the coordinate inputs from Lead Authors. Their role is enabled by Chapter Scientists including managing references, enhancing coordination, and fostering inclusivity within diverse teams. However, the IPCC does not formally recognize or fund CS roles, leaving their provision dependent on CLA’s access to external funding. This disproportionately affects the equitable participation of Global South (GS) CLAs who often lack access to external funds.
This project supported through the CLARE - CLimate Adaptation & REsilience programme, will address these challenges by recruiting and strengthening the capacity of Chapter Scientists to support Global South CLAs involved in the preparation of the IPCC Special Report on Climate Change and Cities (SRCities). It aims to ensure effective participation and contributions of CLAs from the global South and enhance the inclusion of knowledge, evidence and perspectives from the Global South. The project will support Chapter Scientists in their role through training, peer networking opportunities, and equipping Chapter Scientists as knowledge brokers to support CLAs in disseminating IPCC findings effectively. This initiative bridges gaps in equity and representation, fostering a more balanced and inclusive assessment process that addresses priorities relevant to stakeholders in the Global South.
(This is additional funding to SMU with a project extension.)
This project aims to study how better care options can be provided and developed for the local community. The first study will centre around the awareness and preferences of Singapore residents aged 50-76 regarding Assisted Living, with a set of survey questions to be designed and fielded through the Singapore Life Panel. Other focus areas will be developed over the course of this 2-year collaboration based on up-and-coming topics as they emerge.
This project shall (i) enhance the generalizability of hierarchical multi-agent learning and control framework for heterogeneous agents in a range of scenarios and (ii) develop algorithms to analyse and explain the learned behaviour models at the various levels.
In this project, the problem being addressed is to provide an accurate, low-latency, minimal maintenance indoor localisation solution to locate organisational resources. Our goal is to achieve this without any form of Wi-Fi fingerprinting, without any knowledge of the location of the Wi-Fi Access Points (AP), and without the availability of any maps of the indoor spaces being used. We plan to achieve this by leveraging the new 802.11mc Wi-Fi Fine-Time Measurement standard in pure 1-sided mode that allows time-of-flight measurements to be made between a client device and any AP. These measurements will then be used with inertial data to jointly optimise both the location of the device and the location of the APs.
