Cenwei Zhang

Shanghai Jiao Tong University

B.Eng. in Biomedical Engineering

Incoming IE MSc at CUHK

I am a senior undergraduate at Shanghai Jiao Tong University. My current research focuses on efficient and interpretable foundation models, with a particular interest in medical image understanding, generative modeling, and multimodal model ecosystems.

For my graduation research, supervised by Asst. Prof. Suncheng Xiang, I work on structured pruning strategies for medical segmentation foundation models such as SAM/MedSAM, aiming to identify and preserve the adaptation-sensitive components that are crucial for medical image segmentation. I am also interested in generative modeling and explainable AI. In particular, I study how generative on-manifold flows can provide a principled explanatory paradigm for attribution and causal interpretation. This line of work has been guided in detail and pleasantly by Asst. Prof. Lei You, Dr. Manxi Lin, and Dr. Lin Zhu.

After SJTU, I will pursue an MSc in Information Engineering at The Chinese University of Hong Kong. With a humble and open attitude, I am delving into generative models, LLM/VLM engineering, and AI4Med, and I am always open to communication and collaboration！

Outside of research and study, I enjoy citywalking, hiking, and Genshin Impact!

news

Apr 01, 2026	Joined IQuest Research as a LLM algorithm intern (Medical Track).
Dec 13, 2025	Received an offer for an MSc in Information Engineering at CUHK (starting Sep 2026).
Apr 01, 2025	Joined Dolphin AI as a research algorithm intern working on AI for Medical Imaging.

selected publications

arXiv

MedCore: Boundary-Preserving Medical Core Pruning for MedSAM

Cenwei Zhang, Suncheng Xiang^†, and Lei You^†

arXiv preprint arXiv:2605.13688, 2026

Abs HTML PDF Code

Medical segmentation foundation models such as SAM and MedSAM provide strong prompt-driven segmentation, but their image encoders are still too large for many clinical settings. Compression is also risky in medicine because a model can keep high Dice while losing boundary fidelity. We propose MedCore, a structured pruning framework for MedSAM. The main idea is to preserve two kinds of structures: structures that became important during SAM-to-MedSAM adaptation, and structures that have high boundary leverage. We identify the first type by a dual-intervention score that compares zeroing a group with resetting it to its original SAM weight. We identify the second type by boundary-aware Fisher estimation. We also introduce a boundary leverage principle, which shows that compression-induced boundary displacement is controlled by logit perturbation on the boundary divided by the logit spatial gradient. This principle explains why boundary metrics can degrade even when Dice remains high. On polyp segmentation benchmarks, MedCore reduces parameters by 60.0% and FLOPs by 58.4% while achieving Dice 0.9549, Boundary F1 0.6388, and HD95 5.14 after recovery fine-tuning. It also reaches 86.6% parameter reduction and 90.4G FLOPs with strong boundary quality. Our analysis further shows that MedSAM lies in a head-fragile boundary regime: head-pruning steps have 2.887 times larger 95th-percentile boundary leverage than MLP-pruning steps, and this logit-level effect is consistent with BF1 and HD95 degradation.
arXiv

From Baselines to Transport Geodesics: Axiomatic Attribution via Optimal Generative Flows

Cenwei Zhang^*, Lin Zhu^*, Manxi Lin, and 1 more author

arXiv preprint arXiv:2603.05093, 2026

Abs HTML PDF Code

Feature attributions often hide a critical modeling choice: they explain a prediction along a counterfactual path from a reference state to an input. Different baselines, interpolations, and generative trajectories define different paths and can therefore produce different explanations. We study this path ambiguity as a modeling problem. Our central question is whether the path can be chosen by the data-generating transport process, rather than by a hand-designed interpolation or by the sensitivity geometry of the model being explained. We separate attribution into fixed-path credit allocation and path selection. For a fixed path, we prove that the Aumann-Shapley line integral is the unique attribution rule under standard fixed-path axioms and explicit coordinate-trace regularity. For path selection, we minimize kinetic action over flows that transport a reference distribution to the data distribution, yielding a transport-geodesic attribution principle. We approximate this ideal with Rectified Flow and Reflow and derive stability bounds linking vector-field error to attribution error. Experiments show that lower-action, transport-consistent paths produce more stable and structured explanations, preserving competitive deletion faithfulness, without claiming data-manifold membership.