Hello, I'm
Zihua Wang 王梓骅
Ph.D. Candidate in Software Engineering
Southeast University (东南大学)
Efficient & High-Quality Vision-Language Generation
Hello, I'm
Ph.D. Candidate in Software Engineering
Southeast University (东南大学)
Efficient & High-Quality Vision-Language Generation
I am a Ph.D. candidate at the School of Computer Science and Engineering, Southeast University, advised by Prof. Yu Zhang. Previously, I obtained my bachelor's degree from the School of the Gifted Young, University of Science and Technology of China (USTC).
Currently, I am a research intern at Alibaba Tongyi Lab - Mobile Agent Team, working on GUI Grounding Agent.
My research focuses on efficient and high-quality vision-language generation, including:
= First author | Full list on Google Scholar
Proposes a unified adaptive neighbor-element clustering attention that groups related visual and textual units to improve coherence and fidelity in image-text generation.
Introduces a coreset-based dual retrieval algorithm that efficiently selects compact and informative in-context demonstrations for few-shot learning.
A latent-aware semi-autoregressive speculative decoding framework that compresses redundant visual tokens and generates drafts in parallel, accelerating multimodal generation.
Proposes a speculative verification framework using a heavy VLA as a low-frequency macro-planner and a lightweight verifier for high-frequency closed-loop checking in embodied control.
Introduces an efficient demonstration selection strategy balancing diversity and relevance via cluster-pruning coreset, plus a tailored speculative decoding scheme for acceleration.
Deconstructs non-autoregressive generation into patterns and semantics, proposing label selection and image pre-fusion with joint training to close the quality gap with autoregressive models.
A hierarchical deep clustering framework leveraging spatial-temporal dependencies for interpretable and temporally stable video structure discovery.
Multi-platform GUI agent framework for autonomous mobile device interaction.
Research Intern · GUI Agent & GUI Grounding
Working on multi-agent system, GUI grounding techniques, and RL-based strategies for GUI agents.
Research Intern · Multimodal Image-Text Models
Ph.D. in Software Engineering
School of Computer Science and Engineering
Mar 2023 — Jun 2027 (Expected)
M.S. in Electronic Information (Integrated M.S./Ph.D.)
School of Cyber Science and Engineering
Sep 2020 — Mar 2023
B.E. in Electronic Information Engineering
School of the Gifted Young (少年班学院)
Sep 2016 — Jun 2020
ICLR, ACL, EMNLP, ACM MM, IEEE TMM, IEEE TCSVT, ACM TOMM, ACM MM Asia, FCS
2024, 2025 Nanjing Lishui Half Marathon
Interactive demos of my personal development projects. Click to explore.
