The team from National Cheng Kung University (NCKU) won the 2019 first prize in the 2019 Asia Pacific HPC-AI competition, organised by the HPC-AI Advisory Council and the National Supercomputing Centre (NSCC) Singapore.
NCKU’s Team Bowhibolar bested 25 other teams from 21 other universities across the Asia-Pacific region, based on a decision by judges from the University of Massachusetts in the United States, Pawsey Supercomputing Centre in Australia, NSCC Singapore, and the HPC-AI Advisory Council.
Through the competition, student teams demonstrated technological creativity while overcoming performance challenges of the different HPC and AI tasks.
The artificial intelligence challenge of the 2019 competition focused on using the TensorFlow framework for image recognition model training. The high-performance computing challenge focused on the SWIFT application.
SWIFT is an acronym for SPH WIth Fine-grained inter-dependent Tasking, where SPH means Smooth Particle Hydrodynamics. It is a gravity and SPH solver designed to run cosmological simulations.
In the AI challenge, the teams optimised the training performance of the ResNet-50 Neural Network model using the ImageNet-2012 image database, reaching a training speed of as high as 11,720 images per second using 16 GPUs to achieve an accuracy of 92.3% in just 90 minutes of training.
The HPC challenge results demonstrated a performance improvement of up to 2.64 times compared with the baseline of a 32-node cluster.
Other prize winners include National Cheng Kung University’s AnonymousPanda team in second place while Taiwan’s National Tsing Hua University in third. China’s Fudan University won the Merit Prize while Singapore’s Nanyang Technological University earned the HPC honorable mention and Hong Kong University of Science and Technology the AI honorable mention.
