Recently, 7 academic papers from the AI Engineering Institute of Youdao, Inc. were consecutively selected for several top international academic conferences, including the International Conference on Artificial Intelligence in Education (AIED 2020), the International Conference on Educational Data Mining (EDM 2020), the International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2020), and the World Wide Web Conference (WWW 2020). These papers showcased the development potential of AI+Education in China to the world.

It is reported that the 7 selected academic papers are mainly based on research in the application of AI + education, covering multiple branches of artificial intelligence research, such as speech recognition, data mining, and machine learning. Among them, three papers were selected for the AIED 2020 conference.

According to the information, AIED is a top international conference in the field of educational applications, renowned for "providing intelligent systems and cognitive science methods for high-quality research in the field of educational computing applications." The papers included in AIED represent the latest development direction and level of application of artificial intelligence in the field of education.

Specifically, the three papers selected for the AIED 2020 conference are: "Siamese Neural Networks For Class Activity Detection," which focuses on teacher voice recognition and separation, achieving AUC results of 94.2% and 85.5% for online one-on-one and offline small-class teaching scenarios, respectively; "Neural Multi-Task Learning for Automatic Detection of Teacher Questions in Online Classrooms," which introduces a novel framework for automatically detecting teacher questions in online classrooms, providing a more granular quantification of teacher behavior through the detection of question types; and "Automatic Dialogic Instruction Detection for K-12 Online One-on-one Classes," which adjusts the requirements for teacher classroom behavior based on different subjects and grades, helping teachers improve their teaching skills and enhance teaching quality.

During EDM 2020, TAL's paper "Identifying At-Risk K-12 Students in Multimodal Online Environments: A Machine Learning Approach," which models student dropout behavior in the online one-on-one teaching mode, was selected. This represents the first attempt by both industry and academia to predict student dropout behavior in the K-12 online education scenario. By analyzing multi-dimensional data such as classroom behavior and after-class services, the study aims to timely understand students' learning status and knowledge acquisition, thereby helping to optimize and adjust students' learning plans.

Additionally, at the ICASSP conference, in the paper "Multimodal Learning For Classroom Activity Detection," TAL's multimodal speaker recognition model method based on voiceprint attention structure demonstrated an accuracy that exceeded the SOTA model by about 10%, showcasing the model's superiority in speaker separation results in teaching scenarios. In another paper "UPGRADING CRFS TO JRFS AND ITS BENEFITS TO SEQUENCE MODELING AND LABELING," TAL upgraded the classic sequence model CRF to a joint generative model - JRF. The new model consistently outperformed CRF in various algorithm metrics, offering greater potential for improvements in sequence modeling and labeling tasks across a broader range of fields.

At the same time, TAL's paper on assessing spoken language proficiency in free scenarios, "Dolphin: A Spoken Language Proficiency Assessment System for Elementary Education," was also selected for the top international internet conference WWW2020 and presented at the conference. The paper, based on the solution and algorithm innovations developed by TAL AI Lab, addressed the problem of rapidly, scalably, and standardly assessing students' spoken language proficiency.

It is worth mentioning that over 70% of the technical staff in the TAL AI Lab machine learning team have participated in paper and patent publications. Recently, TAL has had multiple academic achievements selected for top international academic conferences such as AAAI2020 and NCME2020. Furthermore, the TAL AI Lab won the championship in the EmotioNet Facial Expression Recognition Competition at CVPR2020, one of the top conferences in the field of computer vision. A series of AI research achievements have been consecutively accepted by top international academic conferences, signifying the recognition of TAL's research capabilities by the international academic community and indicating that TAL is providing more practical value in educational scenarios through AI technology.

According to reports, in recent years, TAL has continuously increased investment in AI research and development. At present, TAL has developed over 100 AI capabilities across 8 major types, including image, voice, data mining, natural language processing, etc., tailored to educational scenario needs. This has resulted in more than 10 AI solutions for educational scenarios, covering various teaching aspects such as teaching, learning, testing, practicing, and assessment. Currently, TAL is also commercializing multiple AI capabilities and applying them extensively in various internal business operations.

Source: Global Network:  https://baijiahao.baidu.com/s?id=1667291307423879485&wfr=spider&for=pc