Biosketch
Jingpei started his PhD at the Chair of Database Systems and Data Mining at LMU to work on the topic of “Large Language Models for Long-horizon Task Planning” in June 2023. At the meanwhile, he became a relAI member. Before that, he obtained his Bachelor’s degree in Mechanical Engineering at Tongji University in China and a Master’s degree in Robotics, Cognition, Intelligence (RCI) at the Technical University of Munich.
relAI Research
Towards Reliable Video Understanding with Large Language Models
My research focuses on reliable video understanding using multimodal and video large language models (VideoLLMs), with an emphasis on spatio-temporal grounding, hallucination mitigation, and surgical video understanding. I investigate whether VideoLLMs can accurately comprehend object interactions across time and space for precise localization, reasoning, and decision-making. At the same time, I study video hallucination by identifying and reducing inconsistencies in generated responses to improve the trustworthiness and explainability of model outputs. In addition, I explore the application of VideoLLMs in surgical domains, including surgical workflow understanding, scene understanding, and skill assessment, aiming to evaluate and enhance the robustness of AI systems in high-stakes real-world environments. Through these efforts, I aim to develop more reliable, interpretable, and human-aligned video AI systems.