National Center for High-performance Computing (NCHC) of NARLabs, the DeepQ, (a medical care business of HTC), and Changhua Christian Hospital have joined hands to use the Taiwan Computing Cloud (TWCC), with its close to 7-fold faster computational speed, to help Changhua Christian Hospital upgrade Dr. Lan. Dr. Lan is not a human being but a talking medical care robot and, as part of a comprehensive medical network, brings AI care to each and every patient at the hospital.

The core technology used in AI medical care is a strengthened deep-learning AI trained by a supercomputer. This works much like the game AlphaGo, quickly finding the best solutions from a large number of possibilities. In Go, there are 361 possibilities for where the next piece can be placed, but Dr. Lan comprises more than 800 possible choices for the next action, taking into consideration various symptoms, medical tests, and diseases. Training AI under these highly complex conditions has been a considerable challenge.

Fortunately, the TWCC can pROVide computational speeds nearly seven times faster than DeepQ to aid in AI training, as each training session demands more than 10 exa floating-point operations (ExaFLOPS). This unceasing process of exploring and learning to accumulate experience creates more intelligent and more precise AI. The result of this – strengthened deep-learning AI – takes just 0.057 seconds to respond to a patient. With the TWCC’s help, large-scale training exercises that used to take a week to complete now just take one day, while related algorithms have also been impROVed.