Prof. Yuqi Sheng, Macau University of Science and Technology, China
Research Area: Philology, Lexicology, Information Processing, Modern Educational Technology, Knowledge mining, Language Resources
Title: The Bottleneck and breakthrough of Chinese intelligent Knowledge
Alphago's victory has brought unprecedented energy and imagination to the realm of multi-dimensional space. In the critical period of innovation and development, as Academician Zhang Cymbals reminds us, "Deep learning cannot extract features at the semantic level, but can only extract features at the bottom, which is the fundamental reason for its fragility and vulnerability." Even "will lead to the artificial intelligence system built on big data facing the challenge of being unreliable, unreliable, insecure and difficult to promote". The knowledge bottleneck of Chinese intelligentization has become a cross-century problem that contemporary linguistics and Chinese information processing must face together.
Faced with modern Chinese where "ancient, modern, popular, traditional, elegant, vulgar, literary, and white" coexist simultaneously, this study draws on the theory of generative grammar and construction grammar, and puts forward a "chunk" standard view, according to "rules + statistics" "Principle, discuss the feasibility of realizing language knowledge acquisition and semantic intelligent processing.
Prof. Thomas Canhao Xu, BNU-HKBU United International College, China
Research Area: Artificial intelligence, Algorithm analysis and design, Internet-of-things, Medical imaging, Hardware/software co-design
Title: Hardware/Software Codesign of Efficient Deep Learning Algorithms
The efficiency of machine learning and deep learning algorithms is more and more important nowadays. Improving accuracy without considering model efficiency is undesirable. Deep learning algorithms on embedded devices, such as educational devices and/or educational robots, often have demanding real-time requirements. For example, object recognition systems based on cameras usually require a latency of hundreds of milliseconds to respond to events in a timely manner. Commercial embedded devices sometimes offload the machine learning algorithms to the cloud. However, network connection quality and speed are becoming another challenging constraint for these devices. Another choice is to implement a high efficient deep learning algorithm on the embedded device, which isn’t affected by the internet connection. Enabling deep learning on the embedded device is difficult. The main characteristic of embedded devices is low power, which usually means the limited computational capability of the processor and limited size of the memory. From the perspective of software/hardware codesign, in order to speed up the processing speed of deep learning and image recognition algorithms, optimizations at both the algorithmic and hardware-level are required.
A. Prof. Olga Predushchenko, Jiujiang University, China
Research Area: Education, Educational Psychology, Language Education
Title: Innovations in Teaching Practice as a Result of Temporary Independent Distance Learning: Experience in China
Abstract：The hypothesis of this research is based on the statement that upgrading the teacher’s pedagogical activity by implementing the recommendations for teaching staff, obtained during Temporary Independent Distance Learning, will improve the academic results of University students. The areas of teacher’s focus are: discipline, basic general self-learning skills in students, learning environment, phycological comfort, and health condition. The experimental training is performed in “English Writing” as an academic course. The practical activity of a teacher is arranged in accordance with the model of psychological structure and content of activity from the position of System-Activity and Competency Approaches. Comparative analysis of academic results of students in 2019 and 2020 years demonstrates the effectiveness of the experimental training.