No 128, Academia Road, Section 2
Nankang, Taipei 11529, Taiwan
Tel: +886-2-2788-3799#1808 Fax: +886-2-2782-4814 Email: lwku
I study natural language processing (NLP) and multimodal AI, with a focus on sovereign AI and local narratives. As principal investigator of Taiwan’s Public Media NewsLLM project, I work on grounding AI in Taiwan’s public memory and diverse perspectives. I also explore physical AI through biology, agriculture, and sports science, connecting human motion understanding with physical features and simulation as a foundation for future robot motion technologies.
Ongoing Research and Collaborations
Taiwan Public Media NewsLLM: Sovereign AI and Local Narratives As principal investigator, I work with seven public media organizations to build a traceable public memory infrastructure from news and audiovisual archives. Our research combines retrieval-augmented generation (RAG) and multimodal AI to ground responses in local context and diverse Taiwanese perspectives, while supporting autonomy over data use and how local stories are represented.Knowledge Distillation with Hon Hai Research Institute (Foxconn) We study how knowledge transfers from a larger language model to a smaller one. I previously led FoxBrain LLMSR, which focused on multi-step reasoning. Can a smaller model carry what it learned over to a new problem that requires several steps?Agriculture and Robotics I lead the Taiwan side of AgriLoop, a collaboration with the University of Bonn. We study how greenhouse robots can connect a grower’s instruction to a particular plant and action, then adjust when the grower corrects them.Biology and Genomic AI I lead a project on structure-aware genomic AI for cell therapy. We study which genomic features might help predict suitable donors and whether features learned in one cell type are informative in another.Sports Science: Human Motion Understanding and Physics-Based Simulation Building on CoachMe’s motion analysis and coaching feedback, this research direction combines human motion understanding, physical feature modeling, and simulation to study motion correction and generation. It aims to lay the groundwork for future robot motion learning and control.