Language Models
Understanding shared LLM decisions
DecodeShare identifies a compact subspace shared across tasks and tests how it influences language-model decisions during decoding.
ICML'26, Spotlight
Kangning Cui
Assistant Professor · City University of Hong Kong (Dongguan)
I develop efficient and reliable machine learning methods for language models, geospatial data, and medical images and signals. My work focuses on model compression, interpretability, and learning with limited labels.
PhD and research assistant opportunities: join the group.
Recent News
Earlier news
Selected Research
All researchGeospatial AI
Mapping palms in tropical forests
PRISM turns large aerial images into geographic maps of individual palms, combining detection, segmentation, and confidence calibration.
IJCAI'25, Oral
Medical Imaging
Measuring heart motion without labels
A self-supervised motion model detects cardiac phases in adult and fetal ultrasound videos without manual phase annotations.
MICCAI'25, Oral
Join the Group
PhD opportunities at CityU and research assistant positions at CityU-DG. For CityU-DG projects, feel free to email me to discuss your interests.
Application detailsTeaching · Fall 2026
Lecture slides and tutorials with solutions are available on each course page.
Appointments
2026–Present
City University of Hong Kong (Dongguan)
2025–Present
Andrew Sabin Family Center for Environment and Sustainability
2025–2026
Wake Forest University
Education
2021–2025
Ph.D. in Mathematics
Research in image and signal processing for medicine and remote sensing, advised by Prof. Raymond H. Chan and Prof. Jean-Michel Morel.
2024–2025
Visiting DPhil Student in Biomedical Engineering
Research in ultrasound video analysis and biomedical imaging, advised by Prof. Alison Noble.
2017–2021
B.S. in Computing Mathematics
Training in computing mathematics with First Class Honours and a minor in Computer Science.
2020
Exchange Student, Applied Mathematics
Exchange study in applied mathematics in the Netherlands.