Diversity-Driven Knowledge Distillation of Large-scale Models
2025 4th International Conference on Image Processing, Object Detection and Tracking (IPODT 2025)
Keynote Speakers
 | Prof. Jianping Gou, Southwest University, China Jianping Gou received the Ph.D. degree in computer science from University of Electronic Science and Technology of China, Chengdu, China, in 2012. He was a Post-Doctoral Research Fellow with The University of Sydney. Now, he is currently a professor and doctoral Supervisor in College of Computer and Information Science, College of Software, Southwest University, Chongqing, China. His current research interests are artificial intelligence and machine learning. Dr. Gou has been intensively working on developing novel machine learning theories and efficient deep learning algorithms for more than ten years. His research has resulted in more than 150 publications on top-tier journals and conferences. He served as a Section Editor of Recent Advances in Electrical & Electronic Engineering, a Guest Editor of Mathematics, a Guest Editor of Frontiers in Physics, Editorial Board of Mathematics, and Associate Editor of Cognitive Robotics. He served as a reviewer of many international journals such as IJCV, TPAMI, TNNLS, TIP, TKDE, TMM, TKDD, and TCYB, and a program committee member for several international conferences such as IJCAI, ECCV, and ICME. He is also a Senior member of IEEE, a Senior member of both CCF (China Computer Federation), and a Senior member of CSIG (China Society of Image and Graphics). |
Speech Title: Abstract: Large-scale model distillation is a core technology of empowering large-scale models for various downstream applications with low-cost and high-efficiency. On the basis of briefly introducing the development of large-scale models and summarizing the relevant technologies of large model compression, the theory, algorithms, and applications of model distillation are reviewed, the series of works on diversity-driven knowledge distillation are further presented, and the latest large language model distillation is reported. Finally, the prospects of large-scale model distillation are given. | |
 | Prof. Shiqian Wu, Wuhan University of Science and Technology, China Dr. Shiqian Wu is a Professor at the School of Information Science and Engineering, Deputy Director of the Institute of Robotics and Intelligent Systems at Wuhan University of Science and Technology (WUST), and Director of the Hubei Provincial Key Laboratory of Intelligent Information Processing and Real-time Industrial Systems. Dr. Wu received his Bachelor's (1985) and Master's (1988) degrees from Huazhong University of Science and Technology (HUST), China, and his Ph.D. (2001) from Nanyang Technological University (NTU), Singapore. Before joining WUST, he served as an Assistant Professor, Lecturer, and Associate Professor at HUST from 1988 to 1997. From 2000 to 2014, he worked as a Researcher or Research Scientist at the Agency for Science, Technology, and Research (A*STAR), Singapore. Prof. Wu has co-authored two books and over 280 scientific publications (including book chapters and journal/conference papers). He has been recognized as one of China's Most Cited Researchers (2015–2022, Elsevier) and received the 2015 ICIP Best 10% Paper Award and the 2020 ICIEA Best Paper Finalist Award. His research interests include computer vision, pattern recognition, robotics, and artificial intelligence. |
| Conference Proceedings... | |
 | Prof. Mingguang Shan, Harbin Engineering University, China Mingguang Shan, Ph.D., Professor, and Doctoral Supervisor, is the leader of the Optoelectronic Information Intelligent Processing Technology Team and the director of the provincial "Advanced Intelligent Perception Technology Collaborative Innovation Center." He is a council member of the Mechanical Measurement Instruments Branch of the China Instrument and Control Society, a member of the Holography and Optical Information Processing Committee of the Chinese Optical Society, a member of the Optical Testing Professional Committee of the Chinese Optical Society, a member of the Youth Working Committee of the China Instrument and Control Society, and a member of the First Young Experts Committee on Advanced Optical Manufacturing of the Chinese Society for Optical Engineering. His main research areas include optoelectronic detection and instrumentation, computational imaging technology, and high-speed image/video processing technology. He has received two second-class provincial natural science awards (ranked 1st and 4th in contribution), holds over 40 authorized invention patents, and has published more than 110 academic papers in SCI-indexed journals. |
Speech Title: Research Progress on Optical-flow-based Video Vibration Measurement Technology Abstract: Optical-flow-based video vibration measurement transforms each camera pixel into a high-precision virtual vibration sensor. This technique enables intuitive visualization of a structure’s true vibration maps in physical space and achieves seamless integration of temporal dynamics with spatial distribution. It offers distinct advantages, including full-field and non-contact measurement, long-range capability, markerless operation, high spatial resolution, and direct vibration visualization. These features have attracted significant attention and demonstrated strong potential for applications in aerospace engineering, industrial manufacturing, and civil engineering. This report systematically reviews the recent research progress of the authors’ group in optical-flow-based video vibration measurement. The work emphasizes four key aspects: vibration optical-field correction, accurate vibration signal extraction, multi-source noise suppression, and structural modal identification. These advances aim to overcome persistent challenges in existing methods, such as strict requirements on frame rate and illumination, limited data processing efficiency, poor adaptability to complex environments, and insufficient capability for high-frequency vibration measurement. The developed techniques have been validated through vibration measurements on wind tunnel models, critical ship structures, and nuclear power components. Experimental results confirm their strong engineering practicality and reliability. | |
 | Prof. Yang Zhao, Harbin Institute of Technology, Weihai, China Dr. Yang Zhao is a Professor and Doctoral Supervisor, currently serving as the Director of the Weihai Key Laboratory of Intelligent Photoacoustic Detection and Sensing Technology. His primary research focuses on cross-medium (air-sea) detection and communication technology, as well as ultrasonic testing and signal processing technology. He has led or participated in numerous significant research projects, including the National Natural Science Foundation of China (NSFC) Youth Science Fund, NSFC General Program, International Science & Technology Cooperation Program (Ministry of Science and Technology), Key Technology Project for Major Accident Prevention in National Work Safety, Shandong Provincial Natural Science Youth Fund, Shandong Provincial Natural Science Foundation General Program, Shandong Outstanding Young Scientist Award Fund, Shandong Key R&D Program, Shandong Innovation Industrial Cluster Special Project, and Shandong Major Innovation Engineering Project. |
Speech Title: Applicaiton of Photoacoustic Technique on the Detection and Communication Abstract: Laser generated sound (LGS) technology is widely used in material testing and cross-media communication and other fields due to its characteristic of transmitting sound waves in different media by taking advantage of the photoacoustic conversion effect. This presentation reviews the investigation of LGS technology used for the detection titanium alloy microstructure firstly , and then its application on air-water cross-medium communication (AWCMC) was highlighted. The pulsed laser is modulated by the way of on-off keying (OOK) modulation，pulse position modulation (PPM) and Frequency-shift keying（FSK）modulation, respectively. Then, thermal expansion was suggested to generate the sound signal at the air-water interface. A novel hydro-acoustic sensor with high sensitivity was developed by utilizing the distributed feedback fiber laser (DFB-FL). The hydro-acoustic sensor uses an enhanced single-path differential divide and differential self-multiplication (SDD-DSM) algorithm in conjunction with a phase-generate carrier (PGC) type demodulation technique to detect LGS signals with great sensitivity. After filtering, envelope extraction, and digital shaping of LGS signals, the modulated information was successfully decoded, demonstrating the potential of OOK-LGS, PPM-LGS and FSK-LGS based on the given photoacoustic communication system applied to the AWCMC. | |