Getty Images

NUHS to tap edge supercomputing to crunch AI workloads

Singapore’s National University Health System will tap edge supercomputing resources to train AI models in a range of healthcare-related AI applications by mid-2022

Clinician researchers at Singapore’s National University Health System (NUHS) will soon be able to use edge supercomputing to speed up training of artificial intelligence (AI) models in areas such as radiology.

When ready by the middle of next year, the petabyte-scale edge supercomputing infrastructure, dubbed Prescience, will reside at NUHS’s premises to address data privacy and security requirements.

Speaking at a media briefing on the collaboration between NUHS and Singapore’s National Supercomputing Centre (NSCC) to build Prescience, Ngiam Kee Yuan, group chief technology officer at NUHS, said the edge supercomputing capabilities can be used in predicting patient health trajectories, healthcare robotics and genomics variant calling.

In addition, Ngiam said NUHS will be using Prescience to train conversational chatbots that can converse with people in a more natural way. “These conversational chatbots have many use cases – for example, advising patients on chronic disease care,” he said.

Prescience will be powered by graphics processing units, which are faster and more efficient than central processing units in crunching AI workloads in parallel.

At NUHS, training AI models with large amounts of data usually takes days, but Ngiam said the supercomputing infrastructure will reduce training times to hours, allowing its medical and para-medical staff to optimise patient trajectories and improve patient care.

Ngiam said Prescience will also enable NUHS to undertake AI projects at scale, by making it more efficient to anonymise data that is used to train AI models.

Read more about high performance computing in APAC

NUHS has been using a platform called Discovery AI to anonymise patient data, but doing so creates overheads in training AI models.

“If I wanted to send my data to the central supercomputer, I first have to do a large amount of data pre-processing before I can release it,” said Ngiam, noting that having supercomputing capabilities on-premise enables data to be processed natively at its source.

Bernard Tan, NSCC’s director of strategy, planning and engagement, said the technical details of the edge supercomputing infrastructure are being finalised with NUHS and that more information will be shared later.

Separately, NUHS is also working with Singtel to deploy an indoor 5G network with multi-access edge compute capabilities at the National University Hospital’s operating theatres and wards.

As a first for a public hospital, the 5G deployment would address limitations in latencies and bandwidth, and open up possibilities for better healthcare experiences, such as smoother teleconsultations, augmented surgical navigation using mixed reality devices, and robot AI capabilities using cloud and edge computing.

Read more on Artificial intelligence, automation and robotics

Data Center
Data Management