Hi! I am Xiaohui Wang, a final-year undergraduate student majoring in Electronic Information Science and Technology at Fudan University. I am fortunate to be advised by Prof. Tao Chen at the Fudan EDL Lab. I also work closely with Dr. Peng Ye (Postdoc at MMLab@CUHK. My research interests lie in Efficient AI and Embodied AI.
I am proud to have received the National Scholarship (2022-2023) for my academic achievements. I am passionate about exploring ways to enhance AI systems for more efficient and versatile real-world applications.
I have worked as a research intern at MMLab@HKU under the supervision of Prof. Ping Luo. I am excited to work alongside talented researchers.
In addition, I am a member of the EGA robotics team at Fudan University, representing the university in the RoboMaster competition.
If you are interested in connecting, collaborating, or discussing ideas, feel free to reach out to me via email. You can also access my CV HERE.
π₯ News
- 2025.09: Β ππ Our work UltraDelta has been accepted as a poster at NeurIPS 2025οΌ
- 2025.05: Β ππ Our work UltraDelta is now on arxivοΌ
- 2025.05: Β ππ I am proud to be named one of the Top 10 Students in the School of Information Science and Technology!
- 2025.05: Β ππ I have received the Fumei Summer Research Scholarship. Many thanks to the Fumei Foundation!
- 2025.03: Β ππ Our work Delta-DCT is now on arxivοΌ
- 2023.09: Β ππ I have received the National Scholarship. Many thanks for the recognition!
π Honors and Awards
- 2022-2023 National Scholarship
- 2024-2025 Top 10 Student Award at School of Information Science and Technology
- 2025.05 Fumei Summer Research Scholarship (awarded to only 4 students at Fudan University)
- 2023-2024 SCSK Corporation Scholarship
- 2024.09 China Undergraduate Mathematical Contest in Modeling (Second Prize, Shanghai)
π Publications
Breaking the Compression Ceiling: Data-Free Pipeline for Ultra-Efficient Delta Compression
Xiaohui Wang*, Peng Ye*, Chenyu Huang, Shenghe Zheng, Bo Zhang, Lei Bai, Wanli Ouyang, Tao Chenβ
(Accepted by NeurIPS 2025)
- To break the compression ceiling of delta weights, we analyze the limitations of existing methods in information preservation and model stability, and propose UltraDelta, the first data-free pipeline enabling ultra-efficient delta compression, achieving both ultra-high compression ratios and strong performance without relying on any data.
Seeing Delta Parameters as JPEG Images: Data-Free Delta Compression with Discrete Cosine Transform
Chenyu Huang*, Peng Ye*, Xiaohui Wang, Shenghe Zheng, Biqing Qi, Lei Bai, Wanli Ouyang, Tao Chenβ
- Motivated by the classic JPEG compression, we explore the delta compression from the DCT domain for the first time. We first realize data-free delta compression and further reduce the performance degradation. We propose a framework based on compression in the DCT domain, named DELTA-DCT.
π Education
- 2022.06 - present, Electronic Information Science and Technology, Fudan University. (GPA:93/100, rank: 5/95 in class, 5/206 in school)
- 2025.06 - 2025.09, School of Computing and Data Science, The University of Hong Kong (Summer Intern at MMLab@HKU)
Last update in September 2025