Undergraduate Students of CUHKSZ Won the Championship of Tianchi Global AI Challenge on Meteorology

Created: Friday, 28 December 2018

Undergraduate Students of CUHKSZ Won the Championship of Tianchi Global AI Challenge on Meteorology

      Recently, the team composed with four undergraduate students from School of Science and Engineering, CUHKSZ won the first prize in the final of the Tianchi Global AI Challenge on Meteorology.

      Tianchi Global AI Challenge on Meteorology was jointly held by Alibaba Tianchi Platform, Meteorological Bureau of Shenzhen Municipality and Hong Kong Observatory, aiming at rainstorm prevention in the Guangdong, Hong Kong, and Macau Bay Area by AI technology.  

       The competition consisted of preliminaries and finals, lasted for 5 months, and had over 1,700 teams worldwide involved. In addition to the awards, the top five teams would be invited to attend the IEEE ICDM 2018 Singapore Conference for presentations and awards. Every team should download the SRAD2018 training data set and The team downloaded the SRAD2018 training data set and the preliminary test set through the Tianchi platform. Each training sample has 61 radar images, with each radar image taken every 6 minutes. Every team needs to design their own algorithms to predict the next six images (3h, 30 mins/image) based on the previous 31 images (3h, 6 mins/image). Once the model is determined, run the test set and submit the predicted results.

Challenge Significance

       Ranging from the travel of the public to the route of the typhoon, Accuracy of weather forecasts stands significantly to the society. By holding this challenge, the Meteorological Bureau called AI teams to contribute more accurate weather forecasts by their innovative technologies.

Competition Process

       At the beginning of the preliminary round, students were divided into two subgroups no sooner than they got the data. The instructor teachers wished them to try attempting in various directions. After determining the optimal model, the students started data selection and categorization. They kept trying new ideas and model optimization, successfully entering the finals. With the help of teachers, they finished the transformation from the regression model to the classification model, which raise their rank to the second place. Since then, they became more careful, and finally won the championship.

Students Testimonials
Shuliang Ning:Computer Science, SSE, Senior, Shaw College

       As the captain of the team, I am very happy to win the championship with my friends. Everyone didn't understand the image of weather forecasts at first, neither heard about the video prediction. I really appreciate the professors who brought us into the domain of science. After this challenge, I learned a lot about network construction, model optimization, and more importantly, how a project was finished. It will be fine even if our model doesn't work well. So long as we still have ideas, we will take our time to try - try to get a better result. We were honored to be invited to attend the ICDM meeting where we meet lots of reputations and learned a lot. However, we still have plenty of problems and deficiencies. For example, we are still not familiar enough with meteorological images, so that our data processing wasn't professional. The experience of this meeting has strengthened my determination to explore and further study. I'm looking forward to the day in the future when I am qualified to attend the meeting with my own papers.

Yige Hong:Applied Mathematics, SSE, Junior, Diligentia College

       I never expected myself to be the core of the team, neither win the championship together. From May to October, the competition process in the past six months brought us a rough time. We encounter a new field and plenty of new knowledge. In this rough time, we quoted over 30 papers, adjusted parameters and algorithms repeatedly, and modified our codes lines by lines. I am proud of the five weeks stated from July, we made progress every week. I really appreciate the professors for their guidance, as well as teammates for their close cooperation and kind encouragement. This competition helped me realize that I need a full understanding of issues characteristics to solve the problem. In that case, I could find the key point and promote my efficiency. Wish we could continue our research and further study.

Tianxing Yang:Computer Science, SSE, Junior, Diligentia College

       The greatest experience for me in this competition is flexibility. I need to keep trying new methods. Among the three methods we tried at the beginning, we retained only one at the end. Without various forms of changes and attempts, it won't become our current model. During the meeting, we communicated a lot with a team of meteorological practitioners and learned a lot about their practical experience which could possibly improve our test results. We have been trying and changing, but also missing a lot of possibilities. I learned that you will have some regrets if you miss it, but if you don't even try or change it in time, the ideal result will be even less likely.

Haoyu Chen:Computer Science, SSE, Sophomore, Shaw College

       This competition helped me find a lot of deficiencies I never realized before. Meanwhile, I expanded a lot of knowledge and mastered some methods, which help myself a lot in the future. Besides, I am also happy about knowing my teammates, and very grateful to our three mentors. Everyone in this team is very good and brought me a lot of help.

Instructor Professors Testimonials
Prof. Han Xiaoguang :
       The issue of meteorological forecasting has crucial social significance. It is also a popular research topic in the field of data science for many years. Six months ago, our students did not have any professional knowledge in this area, neither know about meteorological image processing and machine learning algorithms. However, I am proud to see them very skilled in data processing and be capable to master the most advanced deep learning techniques. They even improved the existing methods to suit the weather forecasting task and won the championship. I have witnessed their growth, and I hope that they could continue to finish their challenges in the future.


       Dr. Han Xiaoguang    Research Assistant Professor at SSE

       Prof. Han received his Ph.D. in Computer Science from the University of Hong Kong in September 2017. Prior to this, he graduated from the Department of Mathematics of Nanjing University of Aeronautics and Astronautics in 2009. He obtained his master's degree in applied mathematics from Zhejiang University in 2011 and served a research assistant at the City University of Hong Kong from 2011 to 2013. Dr. Han joined the Chinese University of Hong Kong (Shenzhen) and the Shenzhen Research Institute of Big Data in September 2017, being engaged in teaching and research in the field of computer science. His main research interests include computer vision, computer graphics, human-computer interaction, and medical image processing.


Prof. Li Zhen:
       After entering the finals, our students accurately analyzed and summarized the problems they encountered in the preliminary competition, and boldly switched the game strategy from image regression to coarse-grained classification in the final. No arrogance, no fear, and no laziness. They also had close cooperation with the teammates and timely communication with the mentors. The championship is their best award and encouragement.


       Dr. Li Zhen Research Assistant Professor at SSE

       Prof. Li received his bachelor's and master's degrees from Sun Yat-Sen University in 2011 and 2014. Afterward, he received his Ph.D. from the University of Hong Kong in 2018. Dr. Li Zhen also conducted research on visiting scholars at the University of Chicago and the Toyota Institute of Chicago in 2016 and 2018. In September 2018, he joined the Hong Kong Chinese University Shenzhen and Shenzhen Research Institute of Big Data. His research focuses on data mining and deep learning algorithms for protein structure prediction, from sequence level to folding level. He is a key member of the protein structure prediction competition CASP12 champion and has received the latest breakthrough and innovation award from PLOS CB 2018. He is engaged in machine learning algorithms and three-dimensional computer vision problems, such as RGB-D semantic segmentation and shape completion.


Prof. Huang Rui:
       I hope that the students could fully digest the knowledge and skills they learned in this competition. I also wish they could also apply these techniques to their study and practices, obtaining further reputations.


        Dr. Huang    Rui Associate Professor at SSE.

        He graduated from Peking University (Bachelor of Science, 1999), Institute of Automation, Chinese Academy of Sciences (Master of Engineering, 2002), and Rutgers University (Ph.D., 2008). After continuing his post-doctoral research at Rutgers University for 2 years, he returned to China and joined Huazhong University of Science and Technology in 2010. From 2012 to 2016, he worked as a researcher at NEC China Research Institute. He is currently an associate professor at the Chinese University of Hong Kong (Shenzhen) Institute of Technology.。

       Prof. Huang has plenty of research works on data dimensionality reduction and subspace analysis, deformable models, probability map models and their applications in computer vision, pattern recognition, and medical image processing. He was mainly engaged in the researches related to intelligent video surveillance, including pedestrian detection, tracking, and identification. His current research interests focus on the application of computer vision in the field of robotics. Prof. Huang has published more than 50 academic papers in related fields. He has also hosted many research projects including the National Natural Science Foundation.