Prof.Yu Kai Attended the 2nd World Laureate Forum

Released Time: 2019-11-28

On October 29th, the 2nd World Laureate Forum was inaugurated at Lingang New Town, Shanghai. Nearly 70 laureates of Nobel Prize, Turing Award, Wolf Prize, Lasker Prize, and Fields Medal, more than 10 members of the CAS and CAE, and more than 100 outstanding young scientists from around the world discussed the possibilities of science and technology. Prof. Yu Kai of the Speech Lab of the CS Department was invited to the forum as a representative of young scientists for the second time.

How to treat basic research and applied research? How to cultivate the scientific spirit of youth? Is human being the highest form of life on earth? Will humans create creatures smarter than us? Does science and technology develop too fast and pose a threat to society? Does science will definitely bring happiness? At the 2nd World Laureates Forum, scientists discussed many propositions.The forum has closed, but the scientific spirit remained. At the closing ceremony, "Science and Technology, for the Common Destiny of Mankind - The 2nd World Laureates Shanghai Initiative" was released. The Initiative reaffirmed the concern and emphasis on the basic science, a transformative source of society; emphasized that the concept of openness, cooperation and the community of science and technology is still the decisive foundation for supporting scientific development; called for the support for young scientists worldwide; and building a healthy global scientific ecosystem. The initiative proposes that science and technology and scientists benefiting the future of mankind should take on more responsibilities, which can be used as a cornerstone to create a better technology management model.

In this forum, Prof. Yu presented a semantic parsing framework based on dual learning in an academic poster. Semantic parsing is the task of transforming natural language queries into structured logic. A fundamental challenge in this area is the lack of annotation in training samples. In this work, he developed a semantic parsing framework based on dual learning algorithm, which enables semantic parsing models to make full use of the game between labeled and unlabeled data and the game between the main model (semantic parsing) and the dual model (question generation) through dual learning and force them to learn from each other. By using the prior knowledge of logical structure, he proposed a formal and semantic reward signal, which makes the model tend to produce a complete and reasonable logical form. Experiment results show that the method performs at the highest level in ATIS data set, and achieves a competitive effect in OVERNIGTH data set. Article Link:



Copyright @ 2013 SJTU Computer Science & Engineering All Rights Reserved