Seven TAL Education AI Research Papers Accepted at Top International Conferences in 2020

Recently, the Machine Learning team of TAL Education's AI Engineering Institute had seven academic papers consecutively accepted by top international conferences, including the International Conference on Artificial Intelligence in Education (AIED 2020), the Educational Data Mining conference (EDM 2020), the International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2020), and the World Wide Web conference (WWW 2020), demonstrating the potential of AI+Education in China.

The selected papers focus on AI+Education scenarios and cover various AI research branches such as speech recognition, data mining, and machine learning. Three of the papers were accepted at AIED 2020.

AIED is a premier international conference in the education application field, known for providing high‑quality research on intelligent systems and cognitive science methods for educational computing. Papers accepted by AIED represent the latest developments and standards of AI applications in education.

The three papers accepted at AIED 2020 are: Siamese Neural Networks For Class Activity Detection , which identifies and separates teachers' voices from single‑track classroom recordings in both online one‑on‑one and offline small‑class settings, achieving AUCs of 94.2% and 85.5%; Neural Multi‑Task Learning for Automatic Detection of Teacher Questions in Online Classrooms , which proposes a novel framework for automatically detecting different types of teacher questions (open, knowledge‑seeking, dialogue‑management, procedural) to quantify teacher behavior more finely; and Automatic Dialogic Instruction Detection for K‑12 Online One‑on‑One Classes , which adapts to different subjects and grades to help teachers improve instructional techniques and quality.

At EDM 2020, a top conference on educational big data, TAL Education presented the paper Identifying At‑Risk K‑12 Students In Multimodal Online Environments: A Machine Learning Approach . This work is the first industry‑academic effort to predict student dropout behavior in K‑12 online education by analyzing multimodal data from classroom activities and after‑class services, enabling timely insight into students' learning status and informing personalized learning plan adjustments.

At ICASSP, a leading conference on signal processing, two TAL papers attracted strong attention. In Multimodal Learning For Classroom Activity Detection , the authors introduced a multimodal speaker‑recognition model based on a voice‑print attention architecture, surpassing state‑of‑the‑art models by roughly 10% in accuracy for speaker separation in classroom scenarios. The second paper, UPGRADING CRFS TO JRFS AND ITS BENEFITS TO SEQUENCE MODELING AND LABELING , upgrades the classic Conditional Random Field (CRF) to a Joint Random Field (JRF), consistently outperforming CRF across various algorithmic metrics and offering broader possibilities for sequence modeling and labeling tasks.

Another paper, Dolphin: A Spoken Language Proficiency Assessment System for Elementary Education , was accepted at WWW 2020. It presents innovative algorithms for assessing elementary students' spoken language proficiency, addressing the challenge of fast, large‑scale, and standardized oral assessment.

More than 70% of TAL Education's AI Engineering Institute's technical staff have contributed to papers and patents. Recently, the institute also had multiple works accepted at AAAI 2020, NCME 2020, and won the CVPR 2020 EmotioNet facial expression recognition competition, underscoring international recognition of its AI research strength and its practical value in educational scenarios.

TAL Education’s AI research achievements stem from its deep understanding of AI+Education. The company has invested heavily in AI R&D, developing over 100 AI capabilities across eight types (image, speech, data mining, NLP, etc.) and more than ten AI‑driven educational solutions covering teaching, learning, testing, practice, and assessment. Many of these capabilities have been productized and deployed internally.

Building on internal success, TAL is actively sharing its AI capabilities with the broader industry through the TAL AI Open Platform, which now offers over 100 education‑focused AI services, with half being industry‑leading or unique models. In August, the Ministry of Science and Technology approved the construction of a national next‑generation AI open innovation platform for smart education based on TAL’s technology.

Looking ahead, TAL Education will continue to uphold principles of source‑level innovation, application‑driven development, and open sharing, committing to advancing AI+Education integration and collaborating with industry partners to accelerate intelligent transformation of the education sector.

About TAL Education AI Open Platform

TAL AI Open Platform (https://ai.100tal.com/) leverages years of education industry experience to deliver AI technologies that have been validated in real‑world scenarios, providing high concurrency, high availability, and comprehensive services for the entire education industry, and seeks to co‑build a new AI ecosystem for education with partners.

Recruitment Information

TAL’s technology team is actively hiring senior engineers for testing, backend, operations, client development and other roles. Interested candidates can click the “Technical Recruitment” section of this public account for details.

Additional recommended readings include topics such as GPU computing basics, WebRTC source analysis, deep knowledge tracing for intelligent education, TV remote interaction libraries, and more.