Research / Tianjin University
Generative modeling.
Controllable synthesis. Useful data.
I am a PhD student in Information and Communication Engineering at Tianjin University. I study how generative models can produce controllable, structured synthetic data that supports learning when real data is limited.
My work spans geometry-constrained image editing, optimal transport and Schrödinger bridges, and model information discrepancy. Medical imaging provides a demanding setting for much of my work so far.
01 / Selected work
Geometry-constrained image editing through entropic optimal transport and a guided Schrödinger bridge. Synthetic augmentation improves detection on held-out real angiograms.
Controllable device synthesis with anatomical constraints and paired endpoint labels. Synthetic pre-training supports downstream endpoint localization.
Auto-annotated synthetic image pairs connect generative modeling with matching and multi-view geometric correction.
Dataset-independent comparison of model information using Deep Dream, with applications to model evaluation and interpretability.
02 / Updates
OT-Bridge Editor accepted to ICML 2026.
VDSB-GWSyn received early acceptance to MICCAI 2026.
Our work on auto-annotation data generation and multi-view correction appeared in AAAI 2026.
Our invention patent on tuple-loss constrained GAN inversion was granted.
03 / Background
2024 — Present
PhD · Information and Communication Engineering
Research in controllable generation, synthetic data augmentation, and data-efficient medical imaging.
2020 — 2024
B.Eng. · Communication Engineering
Engineering foundations in signal processing and software development, followed by research in deep learning and image generation.
Undergraduate
Mobile Team Lead
iOS and Flutter development for WePeiyang, including feature delivery, release maintenance, and technical training.
04 / Beyond papers
Granted invention patent · 2025
A tuple-loss constraint for generative model inversion, supporting reconstruction accuracy and identity-specific feature preservation.
CN 118470193 B Patent record
Python, PyTorch, MATLAB, Swift, Flutter, C/C++, and LaTeX. Experience connecting model development with reproducible experiments and usable software.
IEEE Student Member
