The Swan-geese team of USTC won once again the championship of the application track in the PAC Competition

发布时间:2020-10-10浏览次数:192

After more than a year of preparation and competition, the finals of the eighth National Parallel Application Challenge (PAC2020) ended successfully in Zhengzhou on September 30th. More than 200 teams from colleges and universities and scientific research institutions across the country signed up for the competition, which were divided into two tracks for optimization and application, and went through two competition tracks: the preliminary competition and the finals. From September 28 to 29, 16 optimization track final teams and 12 application track final teams from National University of Science and Technology, Shanghai Jiaotong University, Zhejiang University, Ocean University of China, Shandong University, University of Science and Technology of China, Computing Center of the Chinese Academy of Sciences and Institute of High Energy Physics of the Chinese Academy of Sciences competed in Zhengzhou for one gold medal, two silver medals and five bronze medals on each track. The two teams in Swan-geese team, University of Science and Technology of China both entered the final of the application track and finally won one gold and one bronze medal.

 

USTC won the Gold medal of the Application Group


All the award-winning teams of PAC2020 took a group photo with the Organizing Committee.


Members of Swan-geeze application team took part in the on-the-spot defense in the final competition.

 

This year, the application team of USTC, which won the gold medal on the application track, is composed of Jiang Qingcai (captain), a master student of the School of computer science and teconology, Chen Junshi, a postdoctor of the School of computer science and teconology, Zhao Minfan, a senior student of the School of computer science and teconology and Wan Lingyun, a PhD candidate of the National Research Center for Micro-scale material Science in Hefei. It is jointly guided by Professor An Hong of the computer School and Hu Wei, a researcher of the National Research Center for Micro-scale material Science in Hefei. The title of their work is Design and Optimization of first-principles excited State Software for E-scale supercomputing system.

 

The first principle theory is a theoretical calculation method based on quantum mechanics, which can design new advanced functional materials and predict novel properties. it is widely used in the fields of materials, energy, quantum computing, environment and biology. The computing software related to the first principle occupies most of the running time of the supercomputing platform at home and abroad, but the copyright of these software is mostly concentrated in Europe and America, and there is not a domestic independently developed scientific computing software. In order to solve this problem, the interdisciplinary research team composed of National Research Center for Micro-scale material Science in Hefei and school of computer science and technology independently designed the first-principles excited state calculation software based on linear response time-dependent density functional theory (LR-TDDFT), then redesigned and optimized the scalability of the software for E-scale supercomputing system (which can perform tens of billions of mathematical operations per second). By using the international advanced interpolation separable density fitting (ISDF) and the conjugate gradient descent (LOBPCG) method based on local optimal block, a series of problems encountered in super-large-scale expansion are solved from the system level. Combined with a variety of program optimization methods, it has reached an unprecedented scale in the calculable material system. The software has completed large-scale tests on the Tianhe-2 supercomputer in the National Supercomputing Guangzhou Center, the Tianhe-3 E-scale prototype in Tianjin Center, the Shenwei E-sclae prototype in Wuxi Center and the supercomputing cloud of paratera Company. The results show that the software can show good scalability and parallel computing efficiency on various platforms. The team members' rich theoretical knowledge, good applied design paradigm and wonderful present defense left a deep impression on the judges of the competition and the experts attending the competition.

 

The application team 2 is composed of Ling Min (captain), Li Mingfan, Yi Huite, and Qiao Liang, students of school of computer science and technology, under the guidance of Professor an Hong of the school of computer science and technology. The Title of their work is Rapid diagnosis of cervical cancer pathological sections based on weakly supervised deep learning.

 

Since its inception in 2013, the PAC Competition has attracted more than 12000 teachers and students from 40 cities, more than 300 universities and colleges. Since the holding of the PAC contest, the Super calculation Swan-geese team of USTC has organized 15 teams to participate in the PAC contest and won all the prizes, winning a total of 5 gold, 3 silver, 5 bronze, and 2 special awards, reflecting the comprehensive qualities of the students in the fields of high performance computing, big data and artificial intelligence cross research, in computing methods, parallel algorithms, parallel computer systems and software implementation and other related technical fields. It fully demonstrates the leading advantages of USTC in computer basic education and the cultivation of innovative practical ability in interdisciplinary fields.

 

Translator: Qingcai Jiang