MammoFlow
Generating anatomically consistent paired mammograms from complementary views with flow matching.
Hi, I’m Yuexi.
I teach machines to
see, listen & reason.
Ph.D. candidate at Yale, building multimodal models for medical imaging and MLLM autoraters for video generation at Google Cloud.
multimodal reasoning + clinical evidence
views + scales + anatomy → alignment
./build_better_models.sh
Current quest
I build models that connect images, language, and evidence—from mammography synthesis and clinical reasoning to audiovisual generation.
- Grounding multimodal reasoning in clinical evidence
- Aligning language across views, scales, and anatomy
- Learning representations that transfer across rotations, tasks, and datasets
-
Google Cloud
Ph.D. Software Engineering Intern in GenMedia, developing MLLM autoraters for video generation.
-
Yale
Ph.D. candidate in Biomedical Engineering, advised by Prof. Nicha C. Dvornek.
-
Michigan + SJTU
Before YaleBachelor’s degrees in Computer Science and Electrical & Computer Engineering. Two campuses, one long adventure.
Lab log
CARE
Orchestrating proposal, segmentation, reasoning, and answer review around explicit clinical evidence.
GLAM
Using mammography geometry to align local visual details with language across views.
MaMA
Combining multi-view supervision and local alignment for mammography vision-language pre-training.
Inventory
Builds
- Mammography synthesis & vision-language pre-training
- Evidence-grounded medical reasoning agents
- Open-source code for reproducible research
Teaches
- Biomedical image processing & analysis at Yale
- Computer vision at Michigan
- Comics, visual art & programming at SJTU
Reviews
- Medical imaging, vision & language
- CVPR, MICCAI, ICLR, ISBI & more
- Machine learning and medical imaging journals
Open to good science & great side quests.