Scholar
Haokun Lin
Google Scholar ID: 7DnpUlIAAAAJ
City University of Hong Kong & CASIA
Citations & Impact
Citations
660
H-index
7
i10-index
6
Publications
15
Co-authors
0
Publications
Resume (English only)
Academic Achievements
- NeurIPS 2024 Oral paper: "DuQuant: Distributing Outliers via Dual Transformation Makes Stronger Quantized LLMs" (co-first author)
- CVPR 2024 paper: "MoPE-CLIP: Structured Pruning for Efficient Vision-Language Models with Module-wise Pruning Error Metric"
- ICLR 2024 paper: "Plug-and-Play: An Efficient Post-training Pruning Method for Large Language Models"
- ACL 2024 Findings paper: "IntactKV: Improving Large Language Model Quantization by Keeping Pivot Tokens Intact"
- ECCV 2024 paper: "MATHVERSE: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?"
- Accepted to ICCV 2025: "DOGR: Towards Versatile Visual Document Grounding and Referring"
- IEEE TMM 2025 paper: "Scale Up Composed Image Retrieval Learning via Modification Text Generation"
- ICLR 2025 paper: "Image-level Memorization Detection via Inversion-based Inference Perturbation"
- Multiple preprints including "TokLIP", "LRQ-DiT", and "Quantization Meets dLLMs"
- First Prize, 2024 Graduate Academic Forum at UCAS
- Top Reviewer at NeurIPS 2024
Co-authors