Gao Huang
Associate Professor
Works on neural network architectures such as DenseNet, efficient inference, reinforcement learning, multimodal perception and generation, and embodied foundation models.
Gao Huang's lab at Tsinghua works on reinforcement learning, generative & foundation models and efficient training & inference. 77 papers since 2023, 2 at top venues, mostly Pattern Recognition and Nature Communications. 13 PhD students and 1 master's student. 18 alumni, most often to Alibaba and Amazon. Applications close .
At a glance
- papers, 2023–2026
- 20/ year
- at T1 venues
- 3%
- PhD students
- 13
- MS students
- 1
- alumni listed
- 18
Papers by yearfrom the lab homepage, 2023–2026
- 9 papers2026 · T1 0 · collecting
- 20 papers2025 · T1 1
- 27 papers2024 · T1 0
- 21 papers2023 · T1 1
T1 venuesOtherVenues: Pattern Recognition, Nature Communications, TCSVT, TNNLS
Research focuspaper titles that use each topic's words
- Reinforcement learning22 papers
- Efficient training & inference14 papers
- Generative & foundation models11 papers
- network architectures0 papers
- multimodal perception0 papers
Paper list
From the lab homepage.
20269 papers
- TCSVTAligning Vision-Language Model with Fine-grained Semantics for Open-Vocabulary Segmentation (opens in a new tab)
- European Chapter of the Association for Computational Linguistics (EACL)Are My Optimized Prompts Compromised? Exploring Vulnerabilities of LLM-based Optimizers (opens in a new tab)
- International Conference on Learning Representations (ICLR)MemoryVLA: Perceptual-Cognitive Memory in Vision-Language-Action Models for Robotic Manipulation (opens in a new tab)
- IEEE International Conference on Robotics and Automation (ICRA)MOVE: A Simple Motion-Based Data Collection Paradigm for Spatial Generalization in Robotic Manipulation (opens in a new tab)
- AAAI Conference on Artificial Intelligence (AAAI)SpatialActor: Exploring Disentangled Spatial Representations for Robust Robotic Manipulation (opens in a new tab)
- International Conference on Machine Learning (ICML Oral)The Flexibility Trap: Why Arbitrary Order Limits Reasoning Potential in Diffusion Language Models (opens in a new tab)
- IEEE International Conference on Robotics and Automation (ICRA)UltraHiT: A Hierarchical Transformer Architecture for Generalizable Internal Carotid Artery Robotic Ultrasonography (opens in a new tab)
- Pattern RecognitionUltraSeP: Sequence-aware pre-training for echocardiography probe movement guidance (opens in a new tab)
- AAAI Conference on Artificial Intelligence (AAAI)Vision Transformers are Circulant Attention Learners (opens in a new tab)
202520 papers
- Neural Information Processing Systems (NeurIPS Spotlight)Absolute Zero: Reinforced Self-play Reasoning with Zero Data (opens in a new tab)
- IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)AdaGen: Learning Adaptive Policy for Image Synthesis (opens in a new tab)
- Conference on Computer Vision and Pattern Recognition (CVPR)ART: Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation (opens in a new tab)
- Neural Information Processing Systems (NeurIPS)Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning (opens in a new tab)
- Conference on Computer Vision and Pattern Recognition (CVPR)CheXWorld: Exploring Image World Modeling for Radiograph Representation Learning (opens in a new tab)
- International Conference on Computer Vision (ICCV)CODA: Repurposing Continuous VAEs for Discrete Tokenization (opens in a new tab)
- Pattern RecognitionCross-modal Adapter for Vision–language Retrieval (opens in a new tab)
- International Conference on Learning Representations (ICLR)DenseGrounding: Improving Dense Language-Vision Semantics for Ego-centric 3D Visual Grounding (opens in a new tab)
- Proceedings of the AAAI Conference on Artificial Intelligence (AAAI Oral)DiveR-CT: Diversity-enhanced Red Teaming Large Language Model Assistants with Relaxing Constraints (opens in a new tab)
- Neural Information Processing Systems (NeurIPS Best Paper Runner-up; Oral)Does Reinforcement Learning Really Incentivize Reasoning Capacity in LLMs Beyond the Base Model? (opens in a new tab)
- Conference on Computer Vision and Pattern Recognition (CVPR)EchoWorld: A Large-scale Chinese-English Multilingual Knowledge Base (opens in a new tab)
- IEEE Conference on Computer Vision and Pattern Recognition (CVPR)Everything to the Synthetic: Diffusion-driven Test-time Adaptation via Synthetic-Domain Alignment (opens in a new tab)
202427 papers
- European Conference on Computer Vision (ECCV)AdaNAT: Exploring Adaptive Policy for Token-Based Image Generation (opens in a new tab)
- International Conference on Learning Representations (ICLR)ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process (opens in a new tab)
- European Conference on Computer Vision (ECCV)Agent Attention: On the Integration of Softmax and Linear Attention (opens in a new tab)
- Proceedings of the Association for Computational Linguistics (ACL)Boosting LLM Agents with Recursive Contemplation for Effective Deception Handling (Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation) (opens in a new tab)
- Neural Information Processing Systems (NeurIPS)Bridging the Divide: Reconsidering Softmax and Linear Attention (opens in a new tab)
- Medical Image Computing and Computer Assisted Intervention (MICCAI)Cardiac Copilot: Automatic Probe Guidance for Echocardiography with World Model (opens in a new tab)
- Neural Information Processing Systems (NeurIPS)DeeR-VLA: Dynamic Inference of Multimodal Large Language Models for Efficient Robot Execution (opens in a new tab)
- Neural Information Processing Systems (NeurIPS)Demystify Mamba in Vision: A Linear Attention Perspective (opens in a new tab)
- European Conference on Computer Vision (ECCV)Efficient Diffusion Transformer with Step-wise Dynamic Attention Mediators (opens in a new tab)
- IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)EfficientTrain++: Generalized Curriculum Learning for Efficient Visual Backbone Training (opens in a new tab)
- Neural Information Processing Systems (NeurIPS)ENAT: Rethinking Spatial-temporal Interactions in Token-based Image Synthesis (opens in a new tab)
- Proceedings of the AAAI Conference on Artificial Intelligence (AAAI Oral)ExpeL: LLM Agents Are Experiential Learners (opens in a new tab)
202321 papers
- International Journal of Computer Vision (IJCV)Adapting Across Domains via Target-Oriented Transferable Semantic Augmentation Under Prototype Constraint (opens in a new tab)
- IEEE/CVF International Conference on Computer Vision (ICCV)Adaptive Rotated Convolution for Rotated Object Detection (opens in a new tab)
- International Conference on Learning Representations (ICLR)Budgeted Training for Vision Transformer (opens in a new tab)
- Proceedings of the AAAI Conference on Artificial Intelligence (AAAI)Causal Intervention for Human Trajectory Prediction with Cross Attention Mechanism (opens in a new tab)
- TNNLS (T1)Decoupled Prioritized Resampling for Offline RL (opens in a new tab)
- IEEE/CVF International Conference on Computer Vision (ICCV)Deep Incubation: Training Large Models by Divide-and-Conquering (opens in a new tab)
- IEEE/CVF International Conference on Computer Vision (ICCV)Dynamic Perceiver for Efficient Visual Recognition (opens in a new tab)
- IEEE/CVF International Conference on Computer Vision (ICCV)EfficientTrain: Exploring Generalized Curriculum Learning for Training Visual Backbones (opens in a new tab)
- IEEE/CVF International Conference on Computer Vision (ICCV)FLatten Transformer: Vision Transformer using Focused Linear Attention (opens in a new tab)
- IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)Glance and Focus Networks for Dynamic Visual Recognition (opens in a new tab)
- Pattern Recognition (PR)Joint Representation Learning for Text and 3D Point Cloud (opens in a new tab)
- Communications in Computational PhysicsLearning Specialized Activation Functions for Physics-Informed Neural Networks (opens in a new tab)
Where alumni went18 listed, 64% to industry
Industry 7Academia 4
Most often to
- Alibaba
- Amazon
- Huawei
Also to
- John Hopkings University
- Princeton University
- University of Hong Kong
Apply to TsinghuaFall 2027
Computer Science PhD (international applicants)
- Round
- Main
- Opens
- GRE
- See program page
- English
- Per school catalog
- Fee
- RMB 800
Not yet confirmed for this intake.
Similar labsnearest by research summary
- USUT Austin
Mingyuan Zhou
- Probabilistic & causal inference
- USMichigan
Satinder Singh Baveja
- Reinforcement learning
- USNYU
Mengye Ren
19 T1 - USNYU
Vasant Dhar
- Reinforcement learning
- USMichigan
Yixin Wang
- Probabilistic & causal inference
- SGNTU
Mengmi Zhang
- Neuroscience & brain-computer interfaces
Your notes
Save this lab to keep notes.
Data
Some numbers could not be verified:
- The lab site and the paper record do not agree.
- Collected
- Verified
- Papers
- The lab homepage
- Members
- Lab homepage
- Alumni
- Lab homepage
Tiers are venue classifications, not lab rankings.
Tsinghuacloses