Jian Li

I am a doctoral student at the College of computer science in Jilin University, China. As an experienced and dedicated algorithm developer proficient in various deep neural networks and long-tailed visual recognition, I have long been involved in the interdisciplinary research of artificial intelligence and education, which was committed to utilizing computer vision technologies to perceive the behaviors of teachers and students. Participating in serveral scientific researches and corporate projects, I am able to effectively self-manage during independent projects as well as collaborate as part of a productive team.

[Feb. 2023] Our paper about long-tailed visual recognition is accepted to CVPR 2023.

Education

Publications

• Jian Li, Ziyao Meng, Daqian Shi, Rui Song, Xiaolei Diao, Jingwen Wang, Hao Xu. FCC: Feature Clusters Compression for Long-Tailed Visual Recognition. CVPR 2023.
• Jian Li, Yu Feng, Jingwen Wang, Tongyu Zhao, Yuxuan Huang, Yatong Zu, Hao Xu. Automatically Identifying Teachers' Autonomy Support using Text Classification on Imbalanced Data. CogSCI 2023. (Abstract)
• Jian Li, Daqian Shi, Piyaporn Tumnark, Hao Xu. A System for Real-time Intervention in Negative Emotional Contagion in A Smart Classroom Deployed under Edge Computing Service Infrastructure. Peer-to-Peer Networking and Applications. 2020.
• Xiaolei Diao, Daqian Shi, Jian Li, Mingzhe Yue, Ruihua Qi, Chuntao Li, Hao Xu. Toward Zero-shot Character Recognition: A Gold Standard Dataset with Radical-level Annotations. ACM MM 2023.
• Yu Feng, Tongyu Zhao, Jian Li, Yatong Zu, Adriano Tavares, Tiago Gomes, Hao Xu. Social Behavioral Sensing: An Exploratory Study to Assess Learning Motivation and Perceived Relatedness of University Students using Mobile Sensing. CogSCI 2023.
• Tongyu Zhao, Jiaying Gao, Yu Feng, Jian Li, Yatong Zu, Sandro Pinto, Adriano Tavares, Hao Xu. Feature Visualization and Attribution Analysis of Confusion for Massive Open Online Course. CogSCI 2023.
• Zijing Yi, Rui Song, Jian Li, Hao Xu. Neighbor-enhanced graph transition network for session-based recommendation. International Journal of Machine Learning and Cybernetics 2023.
• Jiaxiong Zhou, Jian Li, Yubo Yan, Hao Xu. Mixing Global and Local Features for Long-Tailed Expression Recognition. Information 2023.

• Jian Li, jingwen Wang, Ziyao Meng, Daqian Shi, Xiaolei Diao, Hao Xu. BSDC: Biased Shortest Distance Criterion for Long-Tailed Recognition. Submitted to AAAI 2024.
• Jian Li, jingwen Wang, Han Wu, Xue Gu, Rui Song, Qi Zhou, Hao Xu. F3C: Feature Cluster Center Compression based on Knowledge Distillation for Long-Tailed Recognition. Submitted to AAAI 2024.
• Jian Li, Yu Feng, Jingwen Wang, Tongyu Zhao, Tao Yu, Yatong Zu, Hao Xu. Automatically Identifying Teachers' Autonomy Support using Text Classification on Imbalanced Data. Submitted to Applied Intelligence.

Research Experiences

• I led a team to design and develop an intelligent voice note-taking platform for teaching scenarios based on Huawei’s Ascend 310 chip, in which emotion and speech recognition algorithms are migrated to the Atlas 200DK terminal, achieving real-time recognition of academic emotions of students and lecture speech of teachers. This platform has been applied to the actual teaching process at Jilin University and has been included in the Huawei Ascend Developer Community.
• Based on Huawei’s Ascend 910 chip, we migrated advanced deep learning algorithm such as BCE from traditional GPUs to Huawei NPU chips, achieving full-stack migration tasks from model training to accuracy tuning.
• During the migration process, we often encountered situations where the underlying operators were not supported, such as the next operator being incompatible during data iteration. We helped Huawei personnel adjust and develop chip underlying operators.

• Led the team to design and develop academic emotion recognition algorithms for real teaching scenarios.
• Led the team to develop a teacher teaching behavior recognition algorithm that integrates psychological domain knowledge.
• led the team to study long-tail recognition and optimization algorithms at the feature level. Related papers were published in CVPR 2023.

• For more than 600,000 policy text data in Jilin Province, automatic classification is performed from multiple dimensions such as subject and genre. The words in the policy text data are vectorized, and the LSTM model is used to extract text features to achieve multi-label classification. The classification accuracy reaches 98%, and a policy knowledge map is constructed based on labeled policy documents to support policy query applications.
• Based on government legal documents, a matter map is constructed. Based on Bert, LSTM network and CRF module, entity recognition is performed on labeled legal documents, and the recognition accuracy can reach more than 80%.
• Based on the entities such as matters, laws, and departments in the policy documents, a government affairs map is constructed, using the Neo4j graph database to store entities and relationships, and supporting upper-level government intelligent question answering applications.

• Participated in the development of the mobile sensing platform architecture, responsible for the development of the collection and preprocessing module for multimodal sensor time series data, and applied to social behavior data collection and analysis experiments in 13 universities in 10 countries. Led the team to develop a localized mobile sensing platform iSense for Chinese student social behavior analysis, and conducted two data analysis experiments with more than 50 people at Jilin University.

Skills

• Proficient in using deep learning libraries such as PyTorch, Tensorflow, and Keras, familiar with classic long-tail visual recognition methodss, and has rich algorithm application experience in images, speech, text, etc.
• Has good expression ability, has served as a teaching assistant at Jilin University to teach courses such as machine learning and Python, and often shares scientific research papers and technical guidance for students in the experimental group.
• English (CET-6), proficient in using English for paper writing, academic exchanges and daily work. Has good logical thinking and design ability, often assists the tutor in writing project applications/final reports, and produces related report slides.

Awards & Certifications

• Jilin University Postgraduate Academic Scholarship-First Prize. 2023.
• Title of Ascend Developer, Huawei, China. 2021.
• Outstanding master's degree paper. 2020
• China Business Federation Business Science and Technology Progress Award. 2020.
• 2017 Annual Cross-Strait Competition Award-First Prize. 2017.
• National Undergraduate Computer Application Ability and Information Literacy Competition-First Prize. 2017.
• China University Computer Design Competition-Third Prize. 2017.