Westlake ENCODE Lab

ENCODE Lab Family

About Us:
The ENCODE Lab (Efficient Neural network COmputation and DEsign) is led by Dr. Huan Wang, a Tenure-Track Assistant Professor in the School of Engineering at Westlake University. We focus on building efficient and reliable AI systems that are scalable and self-improving, driving both theoretical innovation and practical impact.

Research Focus:
Our research centers on Efficient AI, spanning multiple domains in vision and language:

LLMs/MLLMs: [DyCoke, HoliTom, OmniZip, MergeMix, StreamingTOM, RLKV, VidKV, ResSVD, SparseSSM, ReasonMap, RewardMap, Survey]
Image Generation (diffusion/AR models): [OBS-Diff, VideoPruning, ARPG, FreqExit]
3D Reconstruction: [Niagara]
Snapshot Compressive Imaging: [QuantizedSCI, MobileSCI]

Join Us:
Our lab is actively recruiting Ph.D. students (2026 Fall, 1 position), Research Assistants (1 position), and Visiting Students to work on Efficient AI, Multimodal Systems, and Generative AI. We foster a professional, equal, chill, and creative environment with competitive compensation, sufficient computing resources, and opportunities for academic collaboration and industry internships. Check out the details and fill out this form, or send your CV to [email protected] to apply!

News

2025/11	`[Funds]` We received a research grant from CAAI-Ant Group (CAAI-蚂蚁科研基金). Thank you, CAAI and Ant Group!
2025/10	`[Preprint]` OBS-Diff: The first paper led by a Westlake undergraduate in our lab! Junhan Zhu (Class of 2023) led this work on one-shot pruning for diffusion models. Congrats! [arxiv]
2025/09	`[NeurIPS'25]` Four papers accepted to NeurIPS 2025 in the field of efficient and reliable AI. Congrats to our students and collaborators! Two papers from our lab: HoliTom: A top-performing video LLM token compression method that maintains 99.1% performance while reducing FLOPs to just 6.9%—and it’s training-free! [arxiv] [code] [webpage] FreqExit: A dynamic inference framework for visual autoregressive (VAR) models via early exit with novel frequency-aware guidance. [openreview] [code] [webpage]
2025/09	`[Services]` Prof. Huan Wang will serve as an Area Chair for AAAI 2026, ICLR 2026, and CVPR 2026.
2025/07	`[Preprint]` We are excited to present the first systematic survey on multimodal long-context token compression methods. [arxiv] [code]
2025/07	`[MM'25]` Our paper on efficient video diffusion models via network pruning has been accepted to MM’25. Congrats to Yiming! [arxiv]
2025/06	`[Award]` Congrats to our 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	`[ICCV'25]` Our paper on efficient robot manipulation has been accepted to ICCV’25. Congrats to Yiming! [arxiv]
2025/02	`[CVPR'25]` DyCoke has been accepted to CVPR’25! Congrats to Keda! DyCoke is a training-free, plug-and-play token compression method for fast video LLMs, achieving 1.5x inference speedup and 1.4x memory reduction with no performance loss. [arxiv] [code]
2024/07	`[MM'24]` We present the first real-time, on-device video SCI (Snapshot Compressive Imaging) framework through dedicated network design and distillation-based training. Congrats to Miao! [arxiv] [code]
2024/07	`[ECCV'24]` Our paper on efficient video SCI (Snapshot Compressive Imaging) via network quantization has been accepted to ECCV’24 as an oral presentation. Congrats to Miao! [arxiv] [code]
2024/06	`[New Start]` Our lab is established! Prof. Huan Wang joins Westlake University as a tenure-track assistant professor.

Selected Publications

NeurIPS

HoliTom: Holistic Token Merging for Fast Video Large Language Models

Kele Shao, Keda Tao, Can Qin, Haoxuan You, Yang Sui, and Huan Wang

NeurIPS, 2025

arXiv Bib Code Website

@inproceedings{shao2025holitom,
  title = {HoliTom: Holistic Token Merging for Fast Video Large Language Models},
  author = {Shao, Kele and Tao, Keda and Qin, Can and You, Haoxuan and Sui, Yang and Wang, Huan},
  booktitle = {NeurIPS},
  year = {2025},
}

NeurIPS

FreqExit: Enabling Early-Exit Inference for Visual Autoregressive Models via Frequency-Aware Guidance

Ying Li, Chengfei Lv, and Huan Wang

NeurIPS, 2025

Bib PDF Code Website

@inproceedings{li2025freqexit,
  title = {FreqExit: Enabling Early-Exit Inference for Visual Autoregressive Models via Frequency-Aware Guidance},
  author = {Li, Ying and Lv, Chengfei and Wang, Huan},
  booktitle = {NeurIPS},
  year = {2025},
}

CVPR

DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models

Keda Tao, Can Qin, Haoxuan You, Yang Sui, and Huan Wang

CVPR, 2025

arXiv Bib Code

@inproceedings{tao2025dycoke,
  title = {DyCoke: Dynamic Compression of Tokens for Fast Video Large Language Models},
  author = {Tao, Keda and Qin, Can and You, Haoxuan and Sui, Yang and Wang, Huan},
  booktitle = {CVPR},
  year = {2025},
}

News

Selected Publications

Supported By