|
Jingchen Sun (孙精辰)
Ph.D. Student Department of Computer Science and Engineering University at Buffalo, State University of New York Email: jsun39@buffalo.edu |
I am a CS Ph.D. student at University at Buffalo (SUNY), advised by Prof. Changyou Chen. I received my M.S. from Zhejiang University and B.S. from North China Electric Power University.
My research broadly focuses on Transfer Learning for Multimodal Large Language Models (LLMs), particularly on employing techniques such as knowledge distillation,prompt tuning, train-free adaptation, and to enhance their application in various downstream tasks, such as visual perception, video understanding, and multimodal retrieval.
I am exploring how to use reinforcement learning methods, such as GRPO, and on-policy distillation method to enhance reasoning capabilities of multimodal LLMs. If you are interested in these topics and would like to collaborate, feel free to drop me an email!
I am on the job market and looking for a full-time industry opportunity start in 2027 !
|
Future-OPSD: Learning Video Foresight Reasoning from Unlabeled Videos via On-Policy Self-Distillation Jingchen Sun*, Jiaxing Zhang*, Qingqiao Hu, Dingqiang Ye, Yuqi Jia, Yijun Liang, Di Fu, Yunjie Tian, Dongdi Zhao, Changyou Chen. We propose an on-policy self-distillation framework that learns video foresight reasoning from unlabeled past–future video pairs, focusing the model on the most valuable future information. Preprint 2026 Webpage Code (Coming Soon) |
|
Beta-KD: Uncertainty-Aware Knowledge Distillation for Multimodal Large Language
Models Jingchen Sun, Shaobo Han, Deep Patel, Wataru Kohno, Can Jin, Changyou Chen. We propose a novel uncertainty-aware knowledge distillation method, which can improve the performance of the student model by leveraging the uncertainty of the teacher model. CVPR 2026 Code PDF Webpage |
|
CLAP-S: Support Set-Based Adaption for Downstream Fiber-Optic Acoustic Recognition Jingchen Sun, Shaobo Han, Wataru Kohno, Changyou Chen. We introduce a support set–based adaptation approach to enhance domain adaptation performance in audio-language models. ICASSP 2025 Code PDF Webpage |
|
Craft: Cross-modal Aligned Features Improve Robustness of Prompt Tuning Jingchen Sun, Rohan Sharma, Vishnu Suresh Lokhande, Changyou Chen We propose a novel cross-modal feature alignment method that mitigates the overfitting issue in visual-language model prompt tuning across different domains. WACV 2025 Code PDF Webpage |
|
PIDNet: An Efficient Network for Dynamic Pedestrian Intrusion Detection Jingchen Sun, Jiming Chen, Tao Chen, Jiayuan Fan, Shibo He We propose a novel dynamic pedestrian intrusion detection network, which can efficiently detect the pedestrian intrusion in the video. ACM MM 2020 Code PDF |