Xubing Ye (叶栩冰)

Now, I am a 3rd year master at Shenzhen International Graduate School, Tsinghua University (M.Eng.@THU’2026), supervised by Prof. Yansong Tang. I obtained my bachelor's degree from the School of Software Engineering at Tongji University in 2023. I collaborated with Dr. Yukang Gan, Dr. Yixiao Ge, Dr. Ying Shan and Dr. Zhao Yang.

My current research interest lies at Agentic LLMs and MLLMs.

Email: [email protected]  /  Github  /  Scholar  /  X

* Indicates Equal Contribution

Proposed VoCo-LLaMA, an attention-distilled video token compression method enabling video-LLMs to train and inference million-token (1+ hour) videos within a 4k-context LLM.

Proposed ATP-LLaVA, an efficient MLLM that performs adaptive instance-wise and decoder-layer-wise token pruning with nearly no performance degradation.

Proposed LAVT, a Transformer-based universal referring image and video segmentation (RIS and RVOS) framework that performs language-aware visual encoding in place of cross-modal fusion post feature extraction.

Industrial Experience