Data Science and Intelligent Perception & Decision-Making Team

I. Team Profile

With the core objective of "Data Driving Intelligence, Perception Empowering Decision-Making," the team focuses on the interdisciplinary integration of data science, machine learning, and intelligent perception and decision-making technologies. It covers the entire innovation chain from fundamental algorithm research to industry application implementation, having developed distinctive achievements particularly in intelligent decision-making, smart healthcare, big data processing, remote sensing monitoring, and human-computer collaboration. The team is committed to providing theoretical support and technical solutions for the national digital economy development and industry intelligent upgrading.

II. Team Members

Team Leader and Profile:

Zhao Tingting, Professor, Master's Supervisor, Vice Dean of the School of Artificial Intelligence, holds a Ph.D. from Tokyo Institute of Technology, Japan. She is selected for the Tianjin "131" Talents Training Project (Second Level) and the "Innovative Talent Cultivation Plan for Middle-aged and Young Key Personnel." She has presided over 6 vertical projects, including the National Natural Science Foundation General Project, National Natural Science Foundation Youth Fund, and the Scientific Research Startup Fund for Returned Overseas Scholars from the Ministry of Education. In recent years, she has authored one monograph and published over 60 papers in the field of machine learning, including publications in top international journals such as Neural Networks, Neural Computation, and top international conferences such as NeurIPS, IJCAI, ACML, and ECML/PKDD. She holds core patents, including 15 authorized invention patents. Her industry-university-research achievements have been awarded the Shanxi Provincial Science and Technology Progress Award (Second Class), the China General Chamber of Commerce Service Industry Technology Innovation Award (First Class), the Tianjin Municipal Science and Technology Progress Award (Third Class), the China Light Industry Federation Technology Progress Award, and the China General Chamber of Commerce Science and Technology Progress Award (Third Class). Her main research areas are machine learning and intelligent information processing, primarily including deep reinforcement learning algorithms, intelligent control, and applications in smart healthcare.

Team Members and Profiles:

Team members include Wang Yuan, Zhao Qing, Liu Jianzheng, Wu Chao, and Pan Xuran. The team members have presided over or played major roles in more than 40 national, provincial/ministerial level projects, and major horizontal scientific research projects; published over 200 papers; authored more than 10 professional books; and obtained over 30 authorized invention patents, demonstrating strong research capabilities. The research contents of team members are interdisciplinary and complementary, covering multiple research directions such as intelligent decision-making theory, distributed intelligent computing, multi-modal perception technology, and smart application transformation, possessing a solid foundation for research and collaboration.

III. Research Directions

1. Intelligent Information Processing and Machine Learning Theory
   Dedicated to fundamental theories of deep reinforcement learning, large model representation learning, and explainable AI. Focuses on making key breakthroughs in core technologies such as continuous action space policy search, latent space modeling, text semantic understanding and sentiment analysis, and knowledge graph applications, while exploring frontier methods like meta-reinforcement learning and probabilistic generative models. Research outcomes provide solid theoretical support for intelligent control, decision optimization, multi-modal information fusion, and intelligent text generation.

2. Big Data and Distributed Intelligent Computing
   Tackles challenges in efficient large-scale data processing, analysis, and storage optimization. Has developed strengths in distributed computing architectures (e.g., Spark), energy-efficient scheduling in cloud platforms, and reconstruction algorithms for massive time-series data (e.g., astronomical catalogs). Constructs end-to-end intelligent solutions for the "data storage-scheduling-analysis" pipeline, robustly supporting scientific computing and industry big data applications like astronomical information processing and social media mining.

3. Intelligent Perception and Multi-modal Image Processing
   Focuses on multi-source information fusion and intelligent visual analysis. Core directions include: intelligent interpretation of remote sensing images, intelligent video surveillance systems, and biometric recognition and affective computing. Research spans the entire chain of innovation from low-level perceptual feature extraction to high-level scene understanding and decision-making, serving fields like smart cities, public security, and human-computer interaction.

4. Smart Domain Applications and Technology Transfer
   Oriented towards solving core industry pain points, promotes the deep application and implementation of AI technology in vertical fields. Key exploration directions include: smart healthcare, smart agriculture, autonomous driving, smart education, information security, cross-media understanding, text analysis, and the industrialization of scientific information. We are committed to bridging the value conversion chain from "cutting-edge technology -> core product -> industry service," realizing the social and economic value of scientific research成果.

IV. Academic Achievements

(I) Representative Achievements

1. Deep Reinforcement Learning: The team focuses on policy search algorithms adept at handling large-scale continuous action spaces within reinforcement learning, analyzing, improving, and applying related technologies and methods by combining various statistical learning approaches. Starting from different RL scenarios, it analyzes challenges faced in practical applications. For different scenarios, specific policy search algorithms are provided, the statistical properties of estimators and learning parameters are analyzed, and application examples and quantitative comparisons are demonstrated. Ultimately, a complete knowledge system integrating basic theory, algorithms, and applications has been formed in the field of policy search algorithms for reinforcement learning, providing theoretical basis and technical guidance for intelligent control problems in practical applications.

2. High-Speed Railway Hazard Monitoring Platform: Addressing major safety risks caused by intrusions of lightweight hazards (dust nets, plastic film, color steel plate houses) along high-speed railway lines, the team developed the "Sky Eye Smooth Journey – High-Speed Railway Hazard Monitoring Platform." Based on innovative deep learning semantic segmentation algorithms, the platform focuses on the intelligent interpretation of satellite and UAV remote sensing data, overcoming the challenge of high-precision identification of transparent, floating, and irregular hazards in complex backgrounds. It features lightweight model advantages, supports efficient inference at the edge, enabling real-time UAV monitoring and early warning. The platform can integrate with operation and maintenance systems, establishing a closed-loop risk handling mechanism, promoting the transformation of HSR safety protection from passive response to active intelligent defense.

3. Vertical Domain Intelligent Analysis and Interaction Based on Large Models: Constructed an audit Q&A dataset, organized audit systems and regulations, and established a structured knowledge base; utilized multi-dimensional retrieval-augmented generation technology and integrated chain-of-thought interaction methods, combined with large models, to holistically enhance the intelligence level of audit work. Furthermore, designed an audit dialogue system based on a financial large model. This system integrates the audit knowledge base with the reasoning capabilities of large models, can accurately identify user questions, analyze them combined with multi-dimensional information such as regulations and historical cases, simulate the reasoning process of human thinking, provide professional answers, and assist auditors in decision-making.

(II) Representative Projects

1. Vertical Project: National Natural Science Foundation General Project, No. 61976156, Research on Efficient Policy Search Methods Based on Latent Space in Reinforcement Learning, 720,000 CNY, 2020/01-2023/12, Ongoing, Host: Zhao Tingting.

2. Vertical Project: National Natural Science Foundation, No. 61502339, Research on Policy Search Reinforcement Learning Methods for Complex Tasks in Large-Scale Environments, 229,000 CNY, 2016/01-2018/12, Completed, Host: Zhao Tingting.

3. Vertical Project: National Natural Science Foundation, No. 61702367, Research on Topic Semantics Modeling of Semi-structured Short Texts Based on Meta-information Association Network, 270,000 CNY, 2018.1-2020.12, Completed, Host: Wang Yuan.

4. Vertical Project: Tianjin Municipal Science and Technology Commission Technology Innovation Guiding Special Project, No. 20YDTPJC00560, Research on Key Technologies for Smart Healthcare Equipment Workstation, 200,000 CNY, 2020.10-2022.9, Completed, Host: Wang Yuan.

5. Vertical Project: National Natural Science Foundation, No. 11803022, Research on Efficient Time Series Reconstruction Algorithms for Large-Scale Star Catalog Data, 250,000 CNY, 2019.1-2021.12, Completed, Host: Zhao Qing.

6. Vertical Project: Tianjin Municipal Science and Technology Commission Natural Science Foundation Youth Project, No. 18JCQNJC69800, Research on Energy-Efficient Scheduling for Data-Intensive Batch Processing Scientific Workflows in Wide-Area Cloud Platforms, 60,000 CNY, 2018.10-2021.9, Completed, Host: Zhao Qing.

7. Horizontal Project: Development of an Intelligent Lymphoma Recognition and Classification System, 1,000,000 CNY, 2025.4.1-2026.10.1, Host: Wang Yuan.

8. Horizontal Project: Core Technology R&D for Integrated Equipment Workstation Data Lake, 1,000,000 CNY, 2021.7.1-2025.7.1, Host: Wang Yuan.

(III) Representative Papers and Monographs

1. Tingting Zhao, Ying Wang, Wei Sun, Yarui Chen*, Gang Niu, Masashi Sugiyama. Representation learning for continuous action spaces is beneficial for efficient policy learning. Neural Networks, 2023, 159: 137–152. (CAS Q1 Top)

2. Tingting Zhao, Guixi Li, Tuo Zhao, Yarui Chen, Ning Xie, Gang Niu, Masashi Sugiyama. Learning explainable task-relevant state representation for model-free deep reinforcement learning. Neural Networks, 2024, 180: 106741. (CAS Q1 Top)

3. Tingting Zhao, Shuai Wu, Guixi Li, Yarui Chen, Gang Niu, Masashi Sugiyama. Learning intention-aware policies in deep reinforcement learning, Neural Computation, 2023(35), 1657–1677. (CCF-B Recommended Journal, IF 2.9)

4. Zhao Tingting. Statistical Policy Search Reinforcement Learning Methods and Applications. Publishing House of Electronics Industry, 2021.08. ISBN 9787121419591.

5. Yuan Wang, Peng Huo, Lingyan Tang, Ning Xiong, Mengting Hu, Qi Yu and Jucheng Yang*, Modeling Category Semantic and Sentiment Knowledge for Aspect-level Sentiment Analysis, IEEE Transactions on Affective Computing, 2024 (CAS Q1 Top)

6. Yuan Wang, Anqi Liu, Jucheng Yang, Lin Wang, Ning Xiong, Yisong Cheng, Qin Wu, Clinical knowledge-guided deep reinforcement learning for sepsis antibiotic dosing recommendations, Artificial Intelligence In Medicine, 2024 (CAS Q2 Top)

7. Yuan Wang, Yuqiao Liu, Yancui Shi, Yanjun Yu, Jucheng Yang*, User Perceptions of Virtual Hospital Apps in China: Systematic Search, JMIR Mhealth and Uhealth, 2020 (SCI Q2)

8. Qing Zhao, Congcong Xiong, Ce Yu, Chuanlei Zhang, Xi Zhao, A new energy-aware task scheduling method for data-intensive applications in the cloud, Journal of Network and Computer Applications, 2016, 59:14-27 (SCI Q2)

9. Qing Zhao, Chengkui Zhang, Hao Li, Tingting Zhao, Chenzhou Cui, and Dongwei Fan, TT-HEALpix: A New Data Indexing Strategy for Efficient Cross-match of Large-scale Astronomical Catalogs, Publications of the Astronomical Society of the Pacific, p.1-14, 2024.3 (SCI Q3)

10. Qing Zhao, Le Sun, Mengxiang Zhang, Chengkui Zhang, Chenzhou Cui & Dongwei Fan, Storage optimisation and distributed architecture for time series reconstruction of massive astronomical catalogues, Experimental Astronomy, P.821-845, 2023.12 (SCI Q3)

(IV) Representative Patents

1. Zhao Tingting, Yang Jucheng, Zhao Xi, Ren Dehua, Chen Yarui, Fang Shanshan. A Deep Policy Learning Method for Complex Tasks in Large-Scale Environments. (Authorized Invention Patent)

2. Zhao Tingting, Song Yajing, Wang Yuan, Ren Dehua, Yang Jucheng. A Text Generation Method Based on Meta-Reinforcement Learning. (Authorized Invention Patent)

3. Wang Yuan, Xu Tao, Hu Wenshuai, Liang Kun, Zhao Qing, Zhao Tingting, Kong Na. A Short Text Question Answering Method and Device Based on Word Vector Technology and Knowledge Graph Retrieval. (Authorized Invention Patent)

4. Wang Yuan, Zhou Yubo, Xu Tao, Liu Yuqiao, Zhao Tingting, Liang Kun, Yang Jucheng. A Self-Attention Based Joint Word and Label Method for Short Text Classification Prediction. (Authorized Invention Patent)

5. Wang Yuan, Xu Tao, Wang Shilong, Zhou Yubo, Wang Huan, Yang Jucheng, Zhao Tingting, Chen Yarui. A Multi-task Assisted Extreme Multi-label Short Text Classification Method Using Co-occurrence Information. (Authorized Invention Patent)

6. Wang Yuan, Wei Yake, Wu Qin, Yang Hao, Li Jingwei, Wang Dong, Kong Na, Xi Chengshuai. A Risk Prediction Method Based on Clinical Laboratory and Medication Intervention Data. (Authorized Invention Patent)

7. Wang Yuan, Zhang Yaogong, Chen Zengguang, Wang Jinghuan, Yang Jucheng, Zhao Qing, Chen Yarui, Kong Na, Wang Jie. A Food-Disease Association Prediction Method Based on Disease Weighting and Food Category Constraints. (Authorized Invention Patent)

8. Wang Yuan, Zhang Yaogong, Chen Zengguang, Yang Jucheng, Li Zheng, Shi Yancui, Zhao Qing. A Food-Disease Association Prediction Method Based on Matrix Factorization. (Authorized Invention Patent)

9. Wang Yuan, Xing Chen, Yang Jucheng. A Mutual Constraint Topic Model Based Method for Discovering Subtopics in Semi-structured Short Text Collections. (Authorized Invention Patent)

10. Zhao Qing, Li Hao, Zhang Chengkui, Zhang Mengxiang, Sun Le. A Composite Stellar Spectrum Classification Method Based on CNN and LSTM. (Authorized Invention Patent)

11. Zhao Qing, Chen Yarui, Yang Jucheng, Zhang Chuanlei, Zhao Tingting, Sun Di, Liu Jianzheng, Wu Chao. A Hybrid Scheduling Method and System for Scientific Workflows with Wide Nodes in Cloud Platforms. (Authorized Invention Patent)

12. Liu Jianzheng, Yang Jucheng, Yang Huayi, Zhao Tingting, Chen Yarui. A Grayscale Change-Based Biometric Video Replay Attack Detection Method. (Authorized Invention Patent)

V. Social Service Intentions

With the rapid development of AI technology and the socio-economy, AI and software engineering technologies are integrating with various industries. The team is dedicated to focusing on the two first-level disciplines of Software Engineering and Intelligent Science & Technology, conducting interdisciplinary research in next-generation AI, and the intersection of AI with healthcare, public health, light industry, food, agriculture, machinery, brain science, astronomy, remote sensing, etc. The team particularly focuses on how to optimize algorithm models in interdisciplinary fields involving multi-modal data fusion, intelligent decision support, and reliable/trustworthy intelligent system development to improve the performance, accuracy, and interpretability of AI systems, and how to design efficient algorithms tailored to specific application scenarios.

VI. Contact Information

Name: Pan Xuran
Phone: +86 13920157326
Email: pxr@tust.edu.cn
Team Address: Room 312, Building 8, West Campus, Teda Campus, Tianjin University of Science and Technology