Bohan Zhuang
Research Professor
Works on efficient deep learning, including model compression, efficient transformers and large-model training and inference.
Bohan Zhuang's lab at Zhejiang works on efficient training & inference. 54 papers since 2023, 41 at top venues, mostly NeurIPS and ICLR. 11 PhD students and 11 master's students. 7 alumni, most often to ByteDance and Alibaba. Applications close .
At a glance
- papers, 2023–2026
- 14/ year
- at T1 venues
- 76%
- PhD students
- 11
- MS students
- 11
- alumni listed
- 7
Papers by yearfrom the lab homepage, 2023–2026
- 13 papers2026 · T1 10 · collecting
- 12 papers2025 · T1 9
- 18 papers2024 · T1 11
- 11 papers2023 · T1 11
T1 venuesOtherVenues: NeurIPS, ICLR, CVPR, ICCV, ICML
Research focuspaper titles that use each topic's words
- Efficient training & inference24 papers
- efficient20 papers
- transformers1 paper
- model compression1 paper
- large-model training0 papers
Paper list
From the lab homepage.
202613 papers
- CVPR (T1)An Empirical Study on How Video-LLMs Answer Videos Questions
- SIGGRAPH Asia Technical CommunicationsBIFE: Better Interaction, Fewer Errors for Minute-Long Video Generation
- ICLR (T1)BLADE: Block-Sparse Attention Meets Step Distillation for Efficient Video Generation
- Findings of ACLCOV: Chain-of-View Prompting for Spatial Reasoning
- ICML (T1)Flash-GRPO: Efficient Alignment for Video Diffusion via One-Step Policy Optimization
- ICML (T1)FlashBlock: Attention Caching for Efficient Long-Context Block Diffusion
- CVPR (T1)Geometrically-Constrained Agent for Spatial Reasoning
- SIGGRAPH Asia Technical CommunicationsMotion Anything: Any to Motion Generation
- ICLR (T1)RAPID^3: Tri-Level Reinforced Acceleration Policies for Diffusion Transformer
- ICLR (T1)Sparsity Forcing: Reinforcing Token Sparsity of MLLMs
- ICML (T1)TriAttention: Efficient Long Reasoning with Trigonometric KV Compression
- ECCV (T1)VolSplat: Rethinking Feed-Forward 3D Gaussian Splatting with Voxel-Aligned Prediction
202512 papers
- ICLR (T1)Are Large Vision Language Models Good Game Players?
- AAAI (T1)Channel Merging: Preserving Specialization for Merged Experts
- —Few-Step Distillation for Text-to-Image Generation: A Practical Guide
- NeurIPS (T1)FPSAttention: Training-Aware FP8 and Sparsity Co-Design for Fast Video Diffusion
- ICCV (T1)Frequency-Aware Autoregressive Modeling for Efficient High-Resolution Image Synthesis
- ICCV (T1)Neighboring Autoregressive Modeling for Efficient Visual Generation
- —PSA: Pyramid Sparse Attention for Efficient Video Understanding and Generation
- —R-Stitch: Dynamic Trajectory Stitching for Efficient Reasoning
- ICLR (T1)T-Stitch: Accelerating Sampling in Pre-Trained Diffusion Models with Trajectory Stitching
- ICML (T1)ZipAR: Parallel Autoregressive Image Generation through Spatial Locality
- ICCV (T1)ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification
- NeurIPS (T1)ZPressor: Bottleneck-Aware Compression for Scalable Feed-Forward 3DGS
202418 papers
- —[Paper][Project Page] ECCV 2024
- —[Paper][Project Page] ICLR 2024
- CVPR (T1)Efficient Stitchable Task Adaptation
- ICLR (T1)EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models
- —Enhancing Perception Capabilities of Multimodal LLMs with Training-Free Fusion
- —Evaluating and Advancing Multimodal Large Language Models in Ability Lens
- —GMAI-MMBench: A Comprehensive Multimodal Evaluation Benchmark Towards General Medical AI
- ECCV (T1)LongVLM: Efficient Long Video Understanding via Large Language Models
- Findings of ACLLoRAPrune: Structured Pruning Meets Low-Rank Parameter-Efficient Fine-Tuning
- —ME-Switch: A Memory-Efficient Expert Switching Framework for Large Language Models
- NeurIPS (T1)MiniCache: KV Cache Compression in Depth Dimension for Large Language Models
- CVPR (T1)ModaVerse: Efficiently Transforming Modalities with LLMs
202311 papers
- IJCAI (T1)A Survey on Efficient Training of Transformers
- ICCV (T1)BiViT: Extremely Compressed Binary Vision Transformer
- CVPR (T1)Dynamic Focus-aware Positional Queries for Semantic Segmentation
- TPAMI (T1)End-to-end One-shot Human Parsing
- NeurIPS (T1)Mask Propagation for Efficient Video Semantic Segmentation
- TPAMI (T1)Pruning Self-attentions into Convolutional Layers in Single Path
- NeurIPS (T1)PTQD: Accurate Post-Training Quantization for Diffusion Models
- NeurIPS (T1)Second-Order Degradation and Reconstruction for Test-Time Image Super-Resolution
- ICCV (T1)Sensitivity-Aware Visual Parameter-Efficient Fine-Tuning
- TPAMI (T1)Single-path Bit Sharing for Automatic Loss-aware Model Compression
- CVPR (T1)Stitchable Neural Networks
Where alumni went7 listed, 86% to industry
Industry 6Academia 1
Most often to
- ByteDance
- Alibaba
Also to
- CUHK
Apply to ZhejiangFall 2027
Computer Science PhD (international applicants)
- Round
- Self-funded
- Opens
- —
- GRE
- See program page
- English
- Per ZJU language requirements
- Fee
- RMB 800
Not yet confirmed for this intake.
Similar labsnearest by research summary
- CNTsinghua
Yanda Li
- Signal processing & information theory
- CNSJTU
Xiaoying Gan
- Time series, anomaly & spatio-temporal data
- CNSJTU
Guanjie Zheng
- Time series, anomaly & spatio-temporal data
- USStanford
Jure Leskovec
- Graph mining & networks
- USCMU
Alan Montgomery
1 T1 - USUIUC
Tarek Abdelzaher
- Time series, anomaly & spatio-temporal data
Your notes
Save this lab to keep notes.
Data
Checked against the lab homepage.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
Zhejiangcloses