Scholar
Zeliang Zhang
Google Scholar ID: 7nLfsSgAAAAJ
PhD Candidate @ University of Rochester; BEng @ HUST
Citations & Impact
Citations
582
H-index
12
i10-index
15
Publications
20
Co-authors
17
list available
Publications
Resume (English only)
Academic Achievements
- NeurIPS 2025: Proposed methods to boost adversarial transferability in ViTs (first author, project leader)
- ICCV 2025: Introduced π-AVAS, a physics-integrated audio-visual modeling approach
- CVPR 2025: Proposed targeted forgetting of image subgroups in CLIP (first author)
- CVPR 2025: Co-developed VidComposition, a new benchmark for evaluating video composition understanding in MLLMs
- ICLR 2025: Developed a powerful audio-visual adversarial attack and defense method (first author, project leader)
- ICLR 2025: Proposed FLOPS for optimal query allocation in forward-only training
- Preprint 2024: Pruned visual-related computation in MLLMs to accelerate inference (co-first author)
- ICML 2025: Provided theoretical insights into model ensemble in transferable adversarial attacks (co-first author)
- Findings of EMNLP 2024: Empirically studied quantity bias in CLIP
- Preprint 2024: Proposed Differentiated Beam Decoding (DBD) to mitigate hallucinations in LVLM-based image captioning
Research Experience
- Organizing a workshop at ICCV 2025 in Hawaii (starting April 2025)
- Research intern at Microsoft Research (Redmond), Deep Learning Group (starting May 2024)
- Research intern at Microsoft Research Asia (Beijing), Machine Learning for Sustainability Group (starting October 2021)
- Collaborating with Prof. Yijie Peng (Peking University) on gradient estimation (Zeroth-Order optimization)
- Collaborating with Prof. Xiao-Yang Liu (RPI/Columbia) on high-performance quantum and tensor computation
