I am currently pursuing a master’s degree in the HCP Laboratory 
at Sun Yat-sen University in Guangzhou, under the supervision of Professor Liang Lin (林倞), cooperating with Associate Professor Pengxu Wei (魏朋旭).
I graduated from Software Engineering College, Northeast University(东北大学软件学院) with a bachelor’s degree.
During my undergraduate studies, I served as a research intern in Associate Professor Shi Feng (冯时)’s NEU Data Mining group, where I conducted research on Image-Text Multimodal Sentiment Analysis.
Following my undergraduate studies, I joined DISC Lab 
led by Associate Professor Zhongyu Wei (魏忠钰) at Fudan University as a long-term research intern, focusing on the systematic evaluation and interpretability analysis of Large Vision-Language Models (e.g., LLaVA).
I won the National Scholarship (国家奖学金), the First Prize Scholarship of the College (院级优秀学生一等奖), the First Prize of the Yonyou Network Technology Scholarship (用友奖学金一等奖), etc.
My research interest includes Video/Image Generation and Large Vision Language Model.
🏛️ Publications
ZiYi Dong, Chengxing Zhou, Weijian Deng, Pengxu Wei, Xiangyang Ji, Liang Lin
TL;DR Through analysis of Self-Attention in models like SDXL and PixArt, we find localized instead of global attention patterns, similar to CNNs. Thus, we propose Δ-ConvBlocks, a CNN-based module that distills Self-Attention, achieving 57× lower computational cost (from $O(n^2)$ to $O(n)$) with comparable 1K image generation quality.
Siyuan Wang, Dianyi Wang, Chengxing Zhou(共一), Zejun Li, Zhihao Fan, Xuanjing Huang, Zhongyu Wei
TL;DR Inspired by the visual cortex of the human brain, we investigate whether a functional “visual cortex” exists within large language models (LLMs) as part of their cognitive core. We propose an efficient training strategy for vision-language models, fine-tuning only 25% of sparsely distributed parameters (acting as the “visual cortex”), reducing training time by 23% while preserving 99% of visual performance and slightly enhancing text processing.
Zejun Li, Ye Wang, Mengfei Du, Qingwen Liu, Binhao Wu, Jiwen Zhang, Chengxing Zhou Zhihao Fan, Jie Fu, Jingjing Chen, Zhongyu Wei, Xuanjing Huang
TL;DR This work proposes to re-formulate existing benchmarks into unified LVLM-compatible formats, and presents the ReForm-Eval benchmark, an open-sourced benchmark and evaluation framework offering substantial data for evaluating various capabilities of LVLMs.
📝 Projects
底层视觉下游任务通用大模型 Denoise Anything
TL;DR This project aims to develop a general pre-trained model for low-level vision tasks, based on the image restoration paradigm, encompassing dataset construction, architecture design, and transferability evaluation.
I was primarily responsible for the training and optimization of a multi-scale pre-training framework. Our multi-scale diffusion framework integrates hierarchical feature extraction, structural guidance via ControlNet, and semantic enhancement through a fine-tuned Swin-Transformer. It outperforms single-scale models on eight downstream tasks with lower model complexity.
🎖 Honors and Awards
- 2023.03 National Gold Medal ‘Challenge Cup’ National College Student Business Plan Competition (第十三届“挑战杯”中国大学生创业计划竞赛).
- 2022.08 National Second Prize WeChat Mini Program & Mini Game Development Competition, a national-tier segment of the China Collegiate Computing Contest (2022年中国高校计算机大赛微信小程序、小游戏应用开发赛).
- 2022.10 Provincial First Prize with subsequent nomination for National First Prize evaluation China Undergraduate Mathematical Contest in Modeling (MCM, 2022年高教社杯全国大学生数学建模竞赛).
- 2022.09 Provincial Gold Medal China College Students’ Internet+ Innovation and Entrepreneurship Competition (“建行杯”辽宁省第八届“互联网+”大学生创新创业大赛)
🎓 Educations
- 2024.09 - now, I am currently pursuing a Master of Science degree in Computer Science at Sun Yat-sen University (中山大学计算机学院)
, where I conduct research in the Human Cyber Physical Intelligence Integration Lab (HCP Lab). - 2020.09 - 2024.06, I earned my Bachelor’s degree in Software Engineering College at Northeastern University (东北大学软件学院)
before being admitted through recommendation exemption to Sun Yat-sen University’s Computer Science College.
💬 More About Me
🤖 I’m deeply passionate about cutting-edge AI technologies and committed to developing solid, practical, and interesting work that creates real-world impact. 🔬 My research interests span across multiple domains of artificial intelligence, with particular enthusiasm for implementable solutions.
Beyond my professional pursuits, I maintain diverse hobbies including ✈️ traveling, 🏃 running, 🎮 PC gaming, and 🎲 board games - all of which continuously inspire my technical work with fresh perspectives.
😆 I’m always open to meaningful discussions about technology or casual hangouts. 📬 Feel free to reach out via email or connect on WeChat (xxxiiinnggg) - whether you’d like to 💡 exchange ideas about AI or 🎉 organize a game night!