NUS artificial intelligence research gets boost

NUS Computing Assistant Professors Dr Bhattacharyya (left), Dr Meel (middle), and Dr Yao (right) received the NRF Fellowship for Artificial Intelligence

A rapid advancement in AI-based technologies is changing the way we work, live and interact as communities. Industries are becoming increasingly automated, and employees are being encouraged to gain new skills so that they can work with these technologies. However, as societies look to embrace the benefits of improved AI, people remain acutely aware of the challenges posed in its deployment, in particular in the protection of data, privacy and even identity. 

The development of AI technologies thus requires a holistic approach, where technological innovations are pursued with societal concerns in mind. Such work is continuing at NUS, and has been boosted with the Singapore National Research Foundation (NRF) recently awarding three researchers from NUS Computing with the NRF Fellowship for Artificial Intelligence

One recipient, Dr Kuldeep Meel, aims to develop a mathematical framework that addresses real-world problems in AI. His work is motivated by a particular paradox of AI systems. He explained, “The paradox is this: while cutting-edge AI systems can now achieve human-level accuracy in their predictions, these same systems can easily fall prey to the simplest of adversarial attacks, unlike humans.”

To solve this issue, Dr Meel and his research group will be studying AI systems in three dimensions: to design formally correct AI systems, ensure verification of AI systems and to explain decisions made by these systems. “Today’s AI systems have achieved significant progress similar to those achieved by software and hardware in the past. However, that is also when incorrect behaviour of computers could lead to catastrophic impact on society. Hence, all these goals we aim to achieve can significantly impact the safe adoption of AI,” Dr Meel added.

Another recipient, Dr Angela Yao, will be conducting cutting-edge research in artificial visual intelligence. Specifically, Dr Yao and her team aim to analyse and develop efficient algorithms for video analysis and explore methods for learning from large-scale video data.

“Video is the next frontier in artificial visual intelligence,” said Dr Yao. “There is now an unprecedented amount of video data coming from CCTVs, internet video platforms, robotics and more. However, most computer vision algorithms treat video as a collection of image frames, and current algorithms are unable to analyse and predict all of these data. The immediate challenge is to develop new AI models that are efficient, flexible and powerful.”

Dr Yao believes that results from her study will help inform the progress of many high-impact applications such as autonomous vehicles, home assistance robotics, surveillance and content-based video indexing. “In these systems, an enormous amount of video data needs to be processed. We hope to develop efficient and intelligent video understanding algorithms for these tasks.”

In another project, Dr Arnab Bhattacharyya will develop efficient algorithms for causal inference. In data analytics, causal inference refers to the process of inferring the cause of an action, event or effect. Dr Bhattacharyya and his research group will study the mathematical foundations of algorithmic causal inference and apply these insights into AI and other domains such as economics and biology.

“Machine learning is routinely used not only to classify data but to also assist in policy making and policy predictions,” said Dr Bhattacharyya. “This can be a dangerous task, as correlations in data do not necessarily indicate that there is causation.”

To solve this problem, Dr Bhattacharyya and his team aim to find a way of incorporating causal inference to machine learning. “Causal modelling is an essential component of human intelligence — we use it routinely to make predictions about our world. Yet it is a core problem for AI systems,” Dr Bhattacharyya explained. “We want to develop a framework that is able to automatically build large causal models through feature extraction, experimentation, and inference, and then apply it to problems in economics, biology, and medicine.”

The NRF Fellowship in AI is given to outstanding young researchers to lead impactful and independent AI research in Singapore. Fellows will receive a five-year research grant of up to S$3 million to fund their studies. This feature has been adapted from the NUS School of Computing News.