The hard work algorithm is smarter under scientific research (self-reliance and self-improvement in science and technology, young scientists)
Core reading
Yan Chenggang, a 37-year-old professor at Hangzhou Dianzi University, focuses on intelligent information processing. The algorithm designed by him can detect bad video information in the network and get large-scale application. He often said that scientific research shares worries for the country, and innovation benefits the people. He encouraged students to choose their scientific research direction according to their own plans and interests, and do more research that solves practical problems and is closely related to public life.
In the intelligent information processing laboratory of Hangzhou Dianzi University, with the sound of keyboard tapping, a string of codes jumped on the computer screen.
Yan Chenggang, a professor and doctoral supervisor at the School of Automation, Hangzhou Dianzi University, is writing the code. Yan Chenggang is 37 years old and has served as the director of the intelligent information processing laboratory. Since 2008, Yan Chenggang, as the second finisher, has participated in the project "Theory and Method of Qualcomm Calculation of Internet Video Streams". The algorithm designed by him can detect bad video information in the network and effectively deal with complex video scenes. Now, this achievement has been widely used.
Upgrade algorithm to identify network video and image information.
"Now there are a large number of users uploading data to the platform every day, most of which are pictures and video data, which may contain some vulgar content." Yan Chenggang said that this phenomenon has attracted the attention of him and his team.
Yan Chenggang said that the traditional filtering algorithm mainly focuses on some fixed features in single-mode content, which can play a certain role when the media form is relatively single, but it is difficult to meet the actual needs at present. So, how to innovate algorithms to complete complex information screening?
"We first try to make the algorithm understand the general idea of the whole multimodal content and give a qualitative judgment." Yan Chenggang introduced, but the actual application scenario is very complicated, and it is difficult to meet the actual application requirements by simply giving judgments. Therefore, the team further tried to make the algorithm identify various details in multimodal content, and then connected the details in series to realize the descriptive understanding of multimodal content.
With the help of cross-modal retrieval technology, the team transformed pictures and video data into specific words, and on this basis, built a corpus retrieval database for secondary association and reasoning, so as to obtain more accurate semantic information.
"When the content is complex and it is difficult to give a detailed and comprehensive description directly, you can find the closest content in the database through retrieval, thus indirectly realizing understanding. In this way, as long as our database is good enough and large enough, we can solve this problem. " He added, "For example, if you enter a Chinese character’ Ma’, the understanding of the content is to draw a horse, while the search only requires finding a horse, which is obviously simpler and time-saving." Focusing on single-mode information processing methods such as images and texts, Yan Chenggang’s team has made many breakthroughs in content saliency detection, complex text recognition, target detection and target retrieval.
Innovate to tackle key problems and solve practical problems through scientific research.
Distinguished professor of Changjiang Scholar Award Program, young scholars of Changjiang Scholar Award Program, outstanding scientific and technological worker of Chinese Institute of Electronics, winner of Zhejiang Outstanding Youth Fund … Yan Chenggang has won many honors, but he prefers to call himself "information screener". In his view, the goal of team members is to screen out bad information and clear up cyberspace.
In the interview, Yan Chenggang always talked about "scientific research sharing worries for the country, innovation benefiting the people". "Do scientific research to indomitable spirit, researchers should have the spirit of responsibility. On the one hand, we must really solve practical problems for the country. On the other hand, we must serve the people. " Yan Chenggang said.
Yan Chenggang’s team has successively won the first prize of Science and Technology Award of China Institute of Electronics, the second prize of National Natural Science Award and the first prize of Wu Wenjun Artificial Intelligence Science and Technology Progress Award … Behind the high yield of research results, it is inseparable from his self-discipline and diligence.
In 2019, Yan Chenggang followed up an important project in Beijing. It happened that another national key project that the team was responsible for entered a critical stage. How to balance the two projects? Yan Chenggang recalled: "I concentrate on doing projects in Beijing during the day, and the time after 9 pm belongs to team projects. I try my best to solve and demonstrate the problems raised by members during the day." Yan Chenggang said that every Sunday, he would fly back to Hangzhou early in the morning to discuss and study with his team members, and then fly back to Beijing that evening. "That time was simply too busy. Time waits for no one, and opportunity waits for no one. I can’t delay the research progress because of me. I would rather sleep less and make the project progress faster. "
Teach by example and support students to explore areas of interest.
In the summer of 2021, a news that "using gait algorithm to optimize identity recognition to help criminal investigation solve cases" attracted attention, and the algorithm can play an active role in finding lost children and other fields.
"This is the scientific research achievement of the doctoral students I directed. The paper was published at an international academic conference and won the best paper nomination award." Yan Chenggang’s words are full of pride. The practical application of students’ academic achievements makes him very happy: "It is meaningful to combine academic research with public life."
Yan Chenggang’s laboratory research scope includes image processing, computer vision, computer graphics, medical image processing, biological information processing and many other aspects. "The research field of the laboratory is relatively broad, and it is also to provide students with more room to play. As long as students are interested in and have potential fields, I support them to explore." Yan Chenggang said that students should be encouraged to choose their research direction and work direction according to their own plans and interests.
Meng Xiangsong, a graduate student majoring in control engineering in Tsinghua University, is a student of Yan Chenggang at the undergraduate stage. "I used to send emails to ask Teacher Yan. When he replied, it was often one or two o’clock in the middle of the night. Every time I wrote back, I made a detailed problem analysis and was never vague. " Thanks to Yan Chenggang’s example, Meng Xiangsong has developed good research habits and is currently committed to the research direction of artificial intelligence.
Yan Chenggang said: "University teachers must first teach and educate people, preach and teach. This is what Academician Dai Qionghai, my postdoctoral co-tutor, said to me, and it has also become my lifelong pursuit. " He said that he should pass on this spirit of being willing to be a ladder and cultivate more useful talents for the country.
(Wang Liwei participated in the writing)
People’s Daily (15th edition, February 10th, 2022)