Rui Wang

wr_szu@163.com2400101058@mails.szu.edu.cn+86 15938593166 (WeChat)wangrui2025.github.io

Current Status

Applying for Fall 2027 PhD admission; available to join a research group early; currently based in Shenzhen.
Research Interests: Experience-Driven Agent Learning and Continual Adaptation; Previous Work on Efficient Temporal Memory and Stable State Updates.

Education

Shenzhen University—College of Computer Science and Software Engineering
Computer Science and Technology·Master of EngineeringGuangdong, Shenzhen·2024-2027
M.S. Research: AI, Computer Vision and Video Segmentation; Advisor: Prof. Huisi Wu
GPA: 3.19/4
Henan University—School of Computer and Information Engineering
Data Science and Big Data Technology·Bachelor of EngineeringHenan, Kaifeng·2020-2024
GPA: 3.45/4

Publications

OSA: Echocardiography Video Segmentation via Orthogonalized State Update and Anatomical Prior-aware Feature Enhancement
[CVPR 2026 Highlight][CVPR Compute Transparency Champion Award]Computer Vision Top Conference, CCF-A
Rui Wang, Huisi Wu*, Jing Qin
Core Contributions: Developed Stiefel-manifold orthogonal state updates to mitigate memory rank collapse under temporal degradation and noise in echocardiography videos；Integrated anatomical-prior-aware feature enhancement to improve segmentation accuracy and temporal stability under complex cardiac motion。
[arXiv][Project]
GDKVM: Echocardiography Video Segmentation via Spatiotemporal Key-Value Memory with Gated Delta Rule
[ICCV 2025]Computer Vision Top Conference, CCF-A
Rui Wang, Yimu Sun, Jingxing Guo, Huisi Wu*, Jing Qin
Core Contributions: Long-range temporal modeling makes computational efficiency and fine-grained segmentation accuracy difficult to balance; built a linear key-value memory network with gated Delta updates for efficient cross-frame memory updates；Designed a key-pixel feature fusion module to improve robustness to blurred boundaries and noise。
[arXiv][Project]

Projects

StateEvo (Ongoing)
Studying how Qwen agents learn from successful and failed WebShop tasks under limited parameter memory. I introduced Bounded State into SD-LoRA, consolidating past and new updates through a fixed-capacity recurrence to balance learning efficiency and knowledge retention. Multi-round local behavior checks passed; matched trainer comparisons averaged about 37x speedup (trainer only). Planned submission: ICLR 2028.
how2research
An open-source guide to conference submissions, LaTeX writing, and research tooling, maintained as miyuki11.

Honors & Awards

National Graduate Scholarship2025, 2026
Shenzhen University Graduate Academic Scholarship — Special Prize2024
Shenzhen University Graduate Academic Scholarship — First Prize2025, 2026
Shenzhen University Academic Launch Scholarship2026
Henan University Merit Student, Henan University Scholarship2022
CCF-CSP (170 pts, Top 28%), Lanqiao Cup (Provincial Award), Computer Design Competition (Provincial Award), Challenge Cup (Provincial Award), Internet+ (Provincial Award)2022 - 2023

Skills

English: CET-4: 548, CET-6: 435

Hobbies

Volleyball, Baseball, Calligraphy
Last updated: 2026-11-07