Recently, the 2021 World Artificial Intelligence Conference (WAIC 2021) was successfully held in Shanghai. Through various forms such as conference forums, exhibitions, competitions, and application experiences, it built an international cooperation and exchange platform. Among them, the WAIC·AI Developer Forum, which attracted much attention from developers, was themed "The AI Era After Deep Learning", inviting experts from both academia and industry to share cutting-edge technologies.

Wu Zhongqin, Vice President of Technology at TAL Education Group, was invited to attend the forum and delivered a keynote speech titled "Multi-modal Machine Learning and Large-scale Automatic Generation Technology: Algorithm Framework and Industry Practice", providing an in-depth interpretation of the application of multi-modal machine learning and large-scale automatic generation technology in the education field to the audience on site, and introducing the latest research breakthroughs and successful practice cases of TAL Education Group's AI Research Institute.

Wu Zhongqin, Vice President of Technology at the TAL Education Group, delivers a keynote speech

The 2021 World Artificial Intelligence Conference was themed "Intelligent Connectivity, Building a Smart World Together", co-hosted by the National Development and Reform Commission, the Ministry of Industry and Information Technology, the Ministry of Science and Technology, the Cyberspace Administration of China, the Chinese Academy of Sciences, the Chinese Academy of Engineering, the China Association for Science and Technology, and the Shanghai Municipal People's Government. The conference gathered multiple Turing Award winners, dozens of business leaders and renowned scholars, as well as hundreds of leading enterprises such as Huawei, Tencent, and TAL Education Group, to engage in dialogue and exchange on AI innovation ideas, technologies, applications, and talents.

The multi-modal machine learning (Multi-modal Machine Learning，MMML) technology that TAL Education Group highlighted at the AI Developer Forum of this conference aims to establish a model from multiple modalities, capable of processing and correlating information from multiple modalities to solve problems that single-modal models cannot. For example, in daily life, when we understand a person by combining their appearance, lip movements, and body language, it is often more comprehensive and accurate than just listening to their voice. The maturity of this technology brings new possibilities to smart education.

As a technology education company, TAL Education Group generates a large amount of teaching content every day. How to standardize the positive interaction between teachers and students in the classroom and ensure the quality of the classroom is one of the key focuses of TAL Education Group. The "GodEye Classroom Quality Monitoring System" was developed by TAL Education Group's AI Research Institute based on the concept of multi-modal deep learning.

This system can intelligently identify and analyze various classroom behaviors between teachers and students, such as explanations, interactions, and incentives, throughout the class. It can promptly intervene and remind teachers of unreasonable teaching behaviors during the class, and provide prompts for key teaching interactions based on the classroom rhythm and students' states, thereby achieving real-time teaching management to ensure the high quality of each class. At the same time, students can receive personalized learning reports.

On this basis, TAL Education Group's AI Research Institute will conduct a more comprehensive analysis of teachers' teaching processes. The system will label teachers with multi-dimensional tags, including teaching style, fluency of expression, emotional changes during teaching, and specific teaching methods for certain knowledge points. By deconstructing the classroom, the system can help teachers accurately identify areas for improvement in their teaching processes, provide professional suggestions and detailed guidance based on actual analysis, which is very helpful for teachers' classes, practice sessions, and refinement sessions, thereby helping to improve the overall teaching quality.

Currently, the "GodEye Classroom Quality Monitoring System" has been widely applied within TAL Education Group. This project has also been selected for the UNICEF "AI for Children" application report, CSDN's 2019 National Excellent AI Application Cases, and BAAI's 2020 Beijing Zhiguan AI Source Conference's outstanding AI achievements. During the COVID-19 pandemic in 2020, the system analyzed a large amount of classroom data, providing strong underlying technical support for improving teaching quality.

Besides "multimodal machine learning", another major trend in the deep integration of AI and education is "large-scale automatic generation technology". This technology can help achieve personalized intelligent teaching content production, thereby efficiently assisting teaching. For example, the GPT model in the educational scenario of TAL Education automatically generates questions. As long as the teaching and research teachers input the knowledge outline, and then input the knowledge points that are prone to errors or confusion, the teaching content can be quickly generated through the pre-trained model and specific models, generating a large amount of content at a speed far exceeding that of manual work, greatly liberating the teachers' energy. At the same time, students can also achieve efficient breakthroughs in their weak points through targeted learning.

"AI can accept, learn, interpret and reason about information in multiple modalities like humans do, and through large-scale automatic generation algorithms, it can produce intelligent content and interactions. This set of algorithms and frameworks is of great significance to the education industry," Wu Zhongqin said in his speech.

As a pioneer in AI exploration in the education field, the AI Research Institute of TAL Education has achieved remarkable results in its cutting-edge research. In the first half of 2021 alone, TAL Education achieved multiple important breakthroughs in AI technology, with five papers selected for the top international conference on AI in education, AIED 2021. It also won four championships in the CVPR 2021 challenge with its leading visual algorithm technology, and claimed the double championship in the "Non-native Children Recognition" category at INTERSPEECH 2021.

Adhering to the mission of "Love and technology make education better", TAL Education has always been committed to promoting educational equity and high-quality development with technology. As the construction unit of the National New Generation Artificial Intelligence Open Innovation Platform for Smart Education, TAL Education firmly believes in the concept of openness and win-win, continuously open-sourcing its research results, empowering industry partners, and working with renowned universities and research institutions to continuously break through in educational technology, helping to develop more equitable and higher-quality education, and jointly achieving the beauty of education through open cooperation.