ProSMA-UNet: Decoder Conditioning for Proximal-Sparse Skip Feature Selection
MICCAI · 2026
Oral presentation · Early accept
Decoder-conditioned sparse feature selection in U-Net skip connections.
Chun Wun (Sam) Cheng
PhD Student · University of Cambridge · SNSF Scholar
I am a PhD student in Mathematics of Information at the University of Cambridge, in the Department of Applied Mathematics and Theoretical Physics. I am supervised by Carola-Bibiane Schönlieb and Angelica I. Aviles-Rivero, and supported by a four-year Swiss National Science Foundation scholarship.
My research lies at the intersection of machine learning, scientific computing, and computational physics. I develop scalable and robust methods for solving partial differential equations, with a particular interest in neural operators and continuous-depth models.
I also work on medical imaging, diffusion models, and vision-language learning. Previously, I was a research intern in the BBC Computer Vision Team and a visiting student at Tsinghua University’s Yau Mathematical Sciences Center.
For research discussions and collaborations, please get in touch at cwc56@cam.ac.uk.
news
- 2026
- ProSMA-UNet and MAP-Diff accepted for oral presentations at MICCAI 2026.
- Jul 2026
- Co-organised Data, Models, and Gradients: Learning to Optimise — and Beyond at SIAM Optimization, Edinburgh.
- Jun 2026
- Co-organised the AI4Science Beijing Meetup: Learned Partial Differential Equations at Tsinghua University.
- 2025
- Mamba Neural Operator accepted by the Journal of Computational Physics.
- 2025
- Cross-Modal Few-Shot Learning with Second-Order Neural ODEs accepted for an oral presentation at AAAI 2025.
selected publications
all publications →
MAP-Diff: Multi-Anchor Guided Diffusion for Progressive 3D Whole-Body Low-Dose PET Denoising
MICCAI · 2026
Oral presentation
Multi-anchor guided diffusion for progressive 3D low-dose PET denoising.
Mamba Neural Operator: Who Wins? Transformers vs. State-Space Models for PDEs
Journal of Computational Physics · 2025
State-space models and transformers for learning PDE solution operators.
Implicit U-KAN2.0: Dynamic, Efficient and Interpretable Medical Image Segmentation
MICCAI · 2025
Dynamic and interpretable medical image segmentation with implicit U-KAN architectures.
Cross-Modal Few-Shot Learning with Second-Order Neural Ordinary Differential Equations
AAAI · 2025
Oral presentation
Second-order neural ODEs for cross-modal learning from few examples.
Continuous U-Net: Faster, Greater and Noiseless
Transactions on Machine Learning Research · 2024
Continuous U-Net architectures for efficient, noise-robust image analysis.
* Equal contribution.
beyond research
I enjoy teaching and mentoring students, and helping bring the research community together through workshops and tutorials. Outside research, I play volleyball and previously competed with Wah Ching in Hong Kong.