Last year, defeated the United States Brigade, surpassing yourself this year, and the calculation of the Huazhong University team map to win the first one in the world

The Yangtze River Daily Big Wuhan Client August 19 (Reporter Chen Xiaotong Correspondent Wang Xiaoxiao) "Congratulations to Yu Fei and Qingang Gang and won the champion of GraphchalLenge this year!" On the morning of August 17th, the WeChat group of the Computing Team of the Computing Team of the University of Huazhong University of Science and Technology A jubilant, their two entries won the global champion of Graph Challenge, the 2022 map calculation challenge.

Graphchallenge is one of the most influential international events in the field of graph computing, and it is also the "place for soldiers" of the world's top map computing team. Since its establishment in 2017, the championship of the competition has been monopolized by well -known American scientific research units.

In 2021, the average age of 24 -year -old team from Huazhong University of Science and Technology broke this "tradition" and achieved the "zero breakthrough" for China. This year, they reached the top again and won the championship.

The average age of the team is 24 years old

Figure calculation team members took a group photo.

The picture is the organizational structure that expresses the complex relationship between things. Many application scenarios in real life need to be used. Figure calculation uses computing power infrastructure and big data algorithms to dig hidden information from a huge graph data. As the core technology of the next generation of artificial intelligence, the figure computing has been widely used in many fields such as medical, military, and finance, and has become a new strategic high point for global science and technology competition.

Graphchallenge is founded by IEEE, MIT, and Amazon. The technical difficulties encountered in the actual application scenarios are used as test questions. It calls for global scientific researchers to develop new solutions and continuously improve the maps and sparse data from real life and sparse data. Treatment efficiency.

The competition will attract many well -known companies and universities around the world every year, including Nvidia, Cylini, Huawei, Carnegie Mellon University, Washington State University, etc.

Facing strong opponents, the map computing team of Huazhong University of Science and Technology has accumulated a lot, winning the championship in 2021, and reaching the top again in 2022. The two entries won this year were completed by Sun Yufei, Wang Qingang, Xu Shaoxian, and Wu Minkang. Their average age is only 24 years old. Among them, Sun Yufei is also a senior student, and the other three are doctoral and master students.

American opponents have "good equipment", they have "technology"

Figure calculation team members took a group photo.

Wang Qinggang, a team member and a doctoral student who has won the championship twice, introduced that in this game, we and the United States did not even stand on the same running line. For example, Nvidia, they use their most advanced equipment to participate in each year. Other US national laboratories can allow hundreds of GPUs to run together.

Why can we win the championship when there is a gap between the hardware conditions?

Wang Qinggang played an analogy: top sports cars and general sports cars must be a top sports car strong in performance. But the final decision of the game depends more on the driver's level. "Through strong compilation technology, we maximize the existing hardware performance potential, and let our 'car' at the forefront."

There are three tracks in the Graphchallenge competition, focusing on different typical application scenarios of map calculations. This year and last year, the Huazhong big team chose the "sparse neural network reasoning" track, which provides new solutions for accelerating artificial intelligence applications.

"Last year, we were facing a new topic. Without reference, we started from scratch. This year, we have to challenge ourselves that have reached the top of the world last year, and then find problems from a solution with strong performance and optimize it." Wang Qinggang said Essence

Since the official preparation of the competition early this year, the team members have repeatedly discussed and optimized the plan last year. From the already perfect solutions, finding minimal progress may be found. The two entries that won the award start from different angles and adopt different solutions to improve the performance of sparse neural network reasoning.

In 2021, compared with the same track champion in 2020, the Huazhong team has increased its speed by 13.74 times and 22.29 times. This year, the two works they participated in the performance once again increased their performance by 8.12 times and 6.37 times, winning the global championship.

Their map computing technology has been applied to multiple industries

The Central China team won the champion of Graphchallenge this year.

Young is the characteristic of the large -scale computing team of Huazhong. "Our team has a valuable spirit. You pull me, I pull you, everyone can make progress together." Wang Qinggang said that in the figure computing team, the teacher and students added up to more than 40 people. The instructor team includes Professor Zheng Long, Professor Shao Zhiyuan, , Professor Liao Xiaofei, Professor Jin Hai and other maps to calculate experts.

"Teachers are leading the way in front, doctoral students, master's degree students, and undergraduate division of labor. For example, doctoral students are responsible for technical research and propose methodology. They are implemented by master and undergraduate students.

Want to break through, hard work is normal. Since the beginning of the year, the team has arrived at the laboratory at 8 am every morning and left at more than 10 pm. "Last year's plan was very optimized. We needed to make progress in a narrower space this year, and we needed the team's continuous running -in and continuous efforts. The workload is huge." Wang Qinggang said.

Zheng Long, an associate professor at the School of Computer Science and Technology of Huazhong University of Science and Technology, introduced that at present, the team has cooperated with national backbone enterprises such as the State Grid, Huawei, Baidu, Ping An Technology, Alibaba and other national backbone to provide key technical support.With the increasingly complicated relationship of the map, the traditional map computing technology is increasingly difficult to meet the actual needs of complex scenes. In the future, the Central Daping Calculation Team will bravely break into the deep water area of the map computing field, explore universal computer technology that adapts to complex map calculation scenarios, and develop intellectual property rightsAutonomous and controllable high -performance chart computing chips and systems.【Edit: Zhang Jing】

For more exciting content, please download the "Da Wuhan" client in the major application markets.

- END -