Huan Wang's Webpage

2026/06 [TIP'26] One paper about ego-centric 3D scene generation is accepted by Transactions on Image Processing (TIP). Congrats to Zhenyu!
2026/06 [Services] I will serve as an Area Chair (AC) for AAAI 2027.
2026/06 [ECCV'26] Three papers accepted by ECCV 2026, about efficient and reliable AI (topics: speculative decoding, hallucination mitigation, and mobile VAR). Congrats to my students and postdoc (Kejia, Zefang, and Ying)! See you in Malmö! 🎉
2026/06 [CVPR'26] ENCODE Lab is attending CVPR 2026 in Denver. See you in altitude!
2026/05 [Recent works] Can LLMs write efficient kernels for mobile devices? Check it in our MobileKernelBench!
2026/05 [ICML'26] 4 papers accepted by ICML 2026! Congrats to my students and collaborators! See you in Seoul! 🎉
2026/02 [CVPR'26] 6 papers accepted by CVPR 2026! Congrats to my students and collaborators! 🎉 - LinkedIn post
2026/01 [ICLR'26] 4 papers accepted by ICLR 2026! Congrats to my students and collaborators! 🎉 - LinkedIn post
2026/01 [TMLR'26, Survey] Our systematic review of the long-context MLLM token compression methods is accepted by TMLR! [arxiv] [paper-repo]
2026/01 [CPAL'26] Two papers on efficient LLMs/MLLMs via sparsity and low-rank decomposition are accepted by CPAL'26. Congrats to Mingluo (Fall'26 incoming PhD student) and Haolei (visiting student)! Codes will be released soon.
2025/11 [Funds] Received a research fund from CAAI-Ant Group (CAAI-蚂蚁科研基金). Thank you CAAI and Ant Group!
2025/09 [Funds] Received a research fund from a Hangzhou Municipal Talent Program (杭州市青年人才项目). Thanks!
2025/09 [NeurIPS'25] 4 papers accepted by NeurIPS 2025 in the field of efficient and reliable AI. Congrats to my students and collaborators from CWRU! 🎉 The three papers from our group:
- HoliTom: As a top-performing video LLM token compression method, HoliTom can maintain 99.1% performance while reducing the FLOPs to only 6.9%. And, it's training-free! [arxiv] [code] [webpage]
- Poison as Cure: Adversarial visual noise is always malicious to our models like "poison"? No, we find it can also be a cure to mitigate the hallucination problem of VLMs. [arxiv] [code] [webpage]
- FreqExit: FreqExit is a dynamic inference framework for visual autoregressive (VAR) models via early exit equiped with a novel proposed frequency-aware guidance. [openreview] [code] [webpage]
2025/09 [Services] I will serve as an Area Chair for ICLR 2026 and CVPR 2026.
2025/07 [Preprint] We are excited to present the first systematic review of multimodal long-context token compression methods. [arxiv] [code]
2025/07 [MM'25] A paper about efficient video diffusion model via network pruning is accepted by MM'25. Congrats again to Yiming! Code will be released soon.
2025/06 [ICCV'25] A paper about efficient robot manipulation is accepted by ICCV'25. Congrats to Yiming! Code will be released soon.
2025/06 [Award-to-Students] 🎉Congrats to my PhD student Keda Tao on receiving the "2025 Westlake University Xinrui Award (西湖大学博士研究生新锐奖)" (only 2 recipients in AI among all the 2025 Fall PhD students in School of Engineering).
2025/06 [Services] I will serve as an Area Chair for AAAI 2026.
2025/02 [CVPR'25] DyCoke is accepted by CVPR'25! Congrats to my PhD student Keda! DyCoke is a training-free, plug-and-play token compression method for fast video LLMs: 1.5x wall-clock inference speedup and 1.4x memory reduction with no performance drop. [arxiv][code]
2025/02 [Preprint] Can diffusion models blend visual concepts that are semantically very unsimilar (e.g., an orange and a teddy bear)? Yes, we introduce FreeBlend, a new method to blend arbitrary concepts. [arxiv] [code] [webpage]
2025/01 [ICLR'25] One paper about distilling large foundation models with low cost "Compressing Vision Foundation Models at ImageNet-level Costs" is accepted by ICLR'25. Thanks to the lead author Yitian!
2024/12 [Preprint] We present empirical evidence to show that oracle pruning, the "ground-truth" pruning paradigm that has been followed for around 35 years in the pruning community, does not hold in practice. [arxiv][webpage]
2024/09 [NeurIPS'24] We introduce a training framework Scala to learn slimmable ViTs. Using Scala, a ViT model is trained once but can inference at different widths, up to the need of devices with different resources. The project is led by Yitian. Congrats!
2024/07 [MM'24] We present the first real-time on-device video SCI (Snapshot Compressive Imaging) framework via dedicated network design and a distillation-based training strategy. Congrats to Miao!
2024/07 [ECCV'24] One paper about efficient video SCI (Snapshot Compressive Imaging) via network quantization is accepted by ECCV'24 as an oral. Congrats to Miao! [code]
2024/06 [New Start] Join the beautiful Westlake University as a tenure-track Assistant Professor.
2024/04 [Graduation] 🎈PhD student→PhD. So many thanks to my Ph.D. committee (Prof. Yun Raymond Fu, Prof. Octavia Camps, Prof. Zhiqiang Tao) and co-authors!