External Research Grants

"Learning by doing” (LBD) is the phenomenon where a worker’s productivity rises with cumulative production experience. As LBD requires no additional investment in hiring or equipment investment, it is viewed by many as an important channel for firms to achieve productivity growth. Unfortunately, although conceptually simple and intuitive, the sources and enablers of LBD remain a mystery; as a result, even when a firm intends to facilitate LBD among its employees, it is not clear how to effectively achieve it. This challenge originates from the difficulty in quantifying and isolating the effects of LBD, and even in a few instances where the measurement of LBD effects (in terms of productivity) is made possible by natural events, these measurements are typically only at the aggregate level. In this project, the team aims to build a novel Big Data framework to measure the LBD effects for workers in the transport gig economy in Singapore. Their ambition is to measure LBD effects at not just the productivity level, which is easily tainted by other factors, but also at the skill level. They plan to achieve this by mining drivers’ microscopic movement traces and trip fulfilment (including both taxi and ride-hailing drivers), and quantify drivers’ skills in anticipating demands and competition from other drivers. Their research will provide a rare view into how big data can revamp the understanding of labor productivity and LBD effects at the individual level, and it will help policy makers and platform operators to come up with policies that are more effective in helping workers cope with competitions and sudden changes such as disruptions brought about by the COVID-19 pandemic.

In this multi-pronged initiative, we propose to build a framework for developing certifiable AI systems systematically, i.e. with the help of theories, tools, certification standards and processes. This is motivated by the many recently discovered problems on existing AI techniques and systems, e.g. adversarial samples, privacy and fairness issues, as well as the many ad hoc attempts on fixing them. For AI techniques to truly become part of a wide digital transformation across many industries, it is vital that we have foundational mechanisms to quantify the problems in AI models, and rectify the discovered problems.

This five-year longitudinal study with the Ministry of Sustainability and the Environment (MSE) aims to quantify and qualify the economic and environmental impact of the Extended Producer Responsibility (EPR) framework for e-waste in Singapore. The findings will aid MSE in assessing the effectiveness of the e-waste EPR and refining it for future phases.

This project aims at developing a software engine that enables planners to generate optimised routing plans based on a given set of data inputs subject to constraints. This work is an offshoot from the Collaborative Urban Delivery Optimisation (CUDO) project which was completed under the Fujitsu-SMU Urban Computing and Engineering (UNiCEN) Corp Lab, that was funded by the National Research Foundation (NRF). The project will customise and extend the functionalities of the CUDO engine developed at SMU, and integrate the engine into Y3 Technologies’ enterprise platform system via an Application Programming Interface (API).

Deep learning has enabled significant advances in applications involving image, text and audio data, in applications such as surveillance, machine translation and speech recognition. These successes have positioned deep learning as a promising approach to address critical challenges (reducing defects, design-time and down-time) in the advanced manufacturing and engineering (AME) domain. A major barrier to this broader application of deep learning is its need for large, labeled datasets to obtain good performance. Thus, the team aims to develop novel deep learning methods that can learn with 10 to 100 times less data as compared to current approaches. This project will enable deep learning to be used in a wider range of applications especially where data is scarce or expensive to obtain. The team will also demonstrate their methods using real-world data from the three identified AME applications (tentatively defect identification, predictive maintenance and circuit design) to show progress and applicability to the AME domain. This project is a collaboration between A*STAR’s I2R, SMU, SUTD, NTU and NUS.

There is an increasing need to use technology to address unequal access to justice, however, this is met with a dearth of empirical evidence to guide the design of access to justice solutions and evaluate their effectiveness at meeting actual needs. This project thus aims to address this problem, through the design and testing of a legal needs questionnaire for Singapore and conducting a pilot survey.

This project investigates 1) when and why skilled mid-career employees in Singapore, and particularly those coming from families of lower socioeconomic status (SES), fail to leverage job opportunities, and 2) how can such workers leverage job opportunities more effectively. In stage 1 of the project, we aim to advance the understanding of job preferences and choices of workers from lower-SES families using a longitudinal study, and to compare them with preferences of local hiring organizations to pinpoint ways in which the two can be more effectively matched. In stage 2 of the project, we will develop a new field intervention that can to help workers effectively seek information and resources to leverage job opportunities, as well as boost their individual coping strategies. Together, this project aims to extend the body of knowledge on workers coming from lower-SES backgrounds, who are at present relatively neglected, and at the same time to have a positive practical impact by improving career outcomes of such workers.

Collaborative machine learning is an appealing paradigm to build high-quality machine learning (ML) models by training on the data from many parties. However, concerns over trust and security have hindered the sharing of data between organizations. This projects aims to address the notion of trustability (i.e. trustable sharing for AI) of the research team’s proposed model-sharing framework in terms of three key principles: assurance of quality and security, fair distribution of benefits, and the right to be forgotten. A fourth aspect, namely the legality of the sharing transactions, is also a key trustability consideration for a learning collaborator partaking in a sharing transaction with others, given that there are many emerging regulatory frameworks to protect data privacy globally. To close the loop between the legal and technical aspects, the team will also study the technical feasibility of current legality frameworks in light of the technological developments of model-centric sharing. The team will design and develop a trustable, operational and legally-compliant model-centric sharing platform called XModelHub based on the above research, and validate with real-world data and applications provided by industry partners. Given that many real-world AI solutions required collaborations between self-interested parties, the team believes that such a platform would be instrumental for powering a data-driven AI innovation ecosystem through trusted sharing for a vibrant digital economy.

Through the project, the team proposes an integrative program of fundamental research towards a vision in which every human will have such an AI assistant for daily life and work. The overall aim is to build conceptual understanding of human-AI collaboration, to develop representations, models, and algorithms for situated assistance, and to integrate them in an experimental device platform for evaluation. The research program consists of five thrusts: (i) situated language communication with reasoning, (ii) visual-linguistic situation understanding, (iii) human collaboration modeling, (iv) robust situated teaming, and (v) the integrative showcase project The “Other Me”, which develops and evaluates the experimental platform Tom – a wearable situated AI agent that assists the human in creating novel artifacts or diagnosing faults. The research aims to empower a new generation of AI assistants for situated, just-in-time, and federated assistance. They will reinvent the relationship between the human and the device in our daily life and work. This research/project is supported by the National Research Foundation, Singapore under its AI Singapore Programme (AISG Award No: AISG2-RP-2020-016). 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.

As AI becomes ubiquitous, people’s trust in AI is actually dwindling. The key barrier to adopting AI is no longer technical in nature, but more about gaining stakeholders’ trust. Federated Learning (FL), in which training happens where data are stored and only model parameters leave the data silos, can help AI thrive in the privacy-focused regulatory environment. FL involves self-interested data owners collaboratively training machine learning models. In this way, end-users can become co-creators of AI solutions. To enable open collaboration among FL co-creators and enhance adoption of the federated learning paradigm, this project aims to develop the Trustworthy Federated Ubiquitous Learning (TrustFUL) framework, which will enable communities of data owners to self-organize during FL model training based on three notions of trust: 1) trust through transparency, 2) trust through fairness and 3) trust through robustness, without exposing sensitive local data. As a technology showcase, we will translate TrustFUL into an FL-powered AI model crowdsourcing platform to support AI solution co-creation.