NTHU Team Steals the Show at the ASC Student Supercomputer Challenge
A team from NTHU's Department of Computer Science has recently won the Championship in the online group of the 2020-2021 ASC Student Supercomputer Challenge, outperforming more than 300 teams representing prestigious universities around the world. This was the fourth time in recent years that a major computing competition has been won by a team from NTHU’s Department of Computer Science.
The three most important worldwide supercomputer competitions for university teams are those held by the ASC of China, the SCC of the United States, and the ISC of Germany. The ASC competition was suspended last year due to the pandemic, but resumed this year, albeit partially online. A total of 28 teams advanced to the finals. The on-site competition was held in Shenzhen; 21 teams from China participated, and seven overseas teams participated online.
NTHU is a force to be reckoned with in the circle of supercomputer competitions. At the ASC competition, NTHU team won the Championship in 2019, the Highest LINPACK Award in 2018, and the First Class Award and the Innovation Award in 2017. At the SCC competition in the United States, NTHU team won the Championship in 2010 and 2011, and won the Highest LINPACK Award in 2007, 2008, and 2014.
The team’s advisor was Prof. Jerry Chou (周志遠)of the Department of Computer Science, who for many years has been training students to participate in similar competitions. He said that supercomputer competitions hone students’ ability to use software and hardware in solving various real-world problems. In addition to basic skills, participants must also have a strong knowledge of various related fields and know how to apply it in different situations—a strength which his teams have consistently demonstrated over the years.
This year’s team consisted of juniors Wang Tzuwen(王子文), Huang Wenyuan(黃文遠), and Chang Chenghsun(張承勛), and sophomores Mou Chanyu(牟展佑) and Chiang Liyuan(蔣立元). Team captain Wang Tzuwen said that his team began preparing for the competition in the summer of 2019, with training provided by senior classmates Hsiao Yicheng(蕭亦程) and Lin Ente(林恩德). During this year’s winter vacation they began preparing for the preliminaries, training for more than ten hours a day.
Wang said that this year's ASC finals were held on May 8. During the five-day marathon-style competition, the first two days are for installation, the next two days are for competition, and the final day is for reporting the results. In comparison to previous years, competing online was more challenging, because the virtual machines on the cloud platform weren’t made available until the competition began. In addition to deciding on the optimal division of labor, the team had to quickly install and become familiar with the software so that they could start their calculations.
In this year’s finals there were four topics, three of which were announced in advance. The topics were: using artificial intelligence to understand natural language, searching for pulsars, quantum computer simulation, and weather simulation. There was also an independently scored topic requiring cooperation between teams: devising an effective calculation method for use in predicting the evolution of the coronavirus.
Wang and Huang tackled the topic on artificial intelligence by using the BERT language model to enable the computer to do a cloze test. They finished within an hour, and their work was scored at 85% for accuracy.
Mou and Chang handled the topic on pulsar searching. A pulsar consists of the remnants of a fixed star of huge stellar mass. Amongst all celestial bodies, pulsars have the highest density, the strongest magnetic field, and the fastest rotation. The finalists were tasked with using open source software to search for pulsars. The Tsinghua team successfully completed the task in two hours by searching about 160GB of data from an astronomical telescope.
“A pulsar is rather like a lighthouse in outer space; the regular signals it emits can be used in time calibration and astronomical research,” explained Mou. As it turns out, Mou took a course in astronomy during his sophomore year, and thus had the background necessary for understanding the topic and devising an efficient formula.
Chiang was tasked with the topic on quantum computer simulation. He said that two days before the competition he learned that the parameters were different from the ones they were expecting, and the amount of memory required to run the program was far more than anticipated, so he had to rewrite the program on the spot, which he found both challenging and exciting.
Chou gave a special word of thanks to NTHU and the Department of Computer Science for the support provided to student teams over the years; he also thanked the National Center for High-performance Computing (NCHC) for its technical guidance, and Quanta Cloud Technology and Nvidia computer systems for providing funding and equipment.
