|
Ruiyang Zhang I am an second-year Ph.D. student at University of Macau, working with Prof. Zhedong Zheng. Previously, I obtained bachelor degree of computer science at Fudan Unversity in 2021. I am instereted in computer vision and related topics. I believe the ultimate goal of computer vision is to make machine see like human. Currenty, my reseach focuses on enhancing the reasoning and tool use capabilities of multimodal agents through reinforcement learning. |
|
News
|
Research |
|
VSearcher: Long-Horizon Multimodal Search Agent via Reinforcement Learning
Ruiyang Zhang, Qianguo Sun, Chao Song, Yiyan Qi Zhedong Zheng Preprint, 2026 [Paper] / [Code] (1) Build the RL infrastructure required to develop multi-turn tool-using multimodal search agent, including tool implementation, caching mechanism, and RL framework adaptation. (2) Build long-horizon, multi-turn multimodal search agent in real-world web environment via reinforcement learning. Propose a systematic post-training framework for multimodal search agents, including Iterative Injection-based Data Synthesis, Rejection Sampling Fine-tuning, and RL. |
|
SketchThinker-R1: Towards Efficient Sketch-Style Reasoning in Large Multimodal Models
Ruiyang Zhang*, Dongzhan Zhou*, Zhedong Zheng ICLR, 2026 [Paper] / [Code] We propose a three-stage framework to incentivize sketch-style reasoning in large multimodal models, including Sketch-Mode Cold Start, SketchJudge Reward Model training, and Sketch-Thinking Reinforcement Learning. |
|
Uncertainty-o: One Model-agnostic Framework for Unveiling Epistemic Uncertainty in Large Multimodal Models
Ruiyang Zhang, Hu Zhang, Fei Hao, Zhedong Zheng Preprint, 2025 [Website] / [Code] We propose a unified framework for uncertainty estimation of Large Multimodal Models, and harness uncertainty for hallucination detection, hallucination mitigation, and uncertainty-aware CoT. |
|
VL-Uncertainty: Detecting Hallucination in Large Vision-Language Model via Uncertainty Estimation
Ruiyang Zhang, Hu Zhang, Zhedong Zheng Preprint, 2024 [Website] / [Paper] / [Code] We leverage semantic-equivalent prompt perturbation for refined LVLM uncertainty estimation, thereby faciliating more accurate hallucination detection. |
|
Harnessing Uncertainty-aware Bounding Boxes for Unsupervised 3D Object Detection
Ruiyang Zhang, Hu Zhang, Hang Yu, Zhedong Zheng ICCV, 2025 [Paper] / [Code] We harness uncertainty learning to mitigate the negative influence of inaccurate pseudo labels in unsupervised 3D object detection. Specifically, we estimate coordinate level uncerainty and then utilize the learned uncertainty to regularize the self-learning process. |
|
Approaching Outside: Scaling Unsupervised 3D Object Detection from 2D Scene
Ruiyang Zhang, Hu Zhang, Hang Yu, Zhedong Zheng ECCV, 2024 [Paper] / [Code] / [Blog] We enhance Unsupervised 3D Object Detection via LiDAR and 2D fusion to improve detecting ability of far and small objects. |
Internship |
|
Research Intern @ IDEA
2025.10 - Shenzhen, China Topics: Developing long-horizon multimodal search agent via reinforcement learning (RL) Project: VSearcher |
|
Research Intern @ Shanghai AI Lab
2025.5 - 2025.9 Shanghai, China Topics: Leveraging RL to incentivize sketch-style reasoning in large multimodal models Project: SketchThinker-R1 |
Work Experience
|
Education
|
Academic Service
|
Teaching
|
Honor and Award
|
Interest
|
|
Template from Jon Barron's website. |