Multimodal Learning · Video Understanding · Video Agents

Seeking PhD opportunities for Fall 2027. I am interested in multimodal learning, efficient video understanding, and evidence-grounded agents. I also welcome research collaborations and related research internship opportunities.

I am a master's student in Computer Technology at Beijing University of Chemical Technology, with expected graduation in 2027. My research spans robust vision-language adaptation, efficient temporal representations, and retrieval-based video agents.

A question connecting my work is: How can multimodal systems select, retain, and verify the right temporal evidence under limited computation?

📬 alexlessend@gmail.com · GitHub

Can Zhang, Xiaotian Han, Junyuan Shang, Yuchen Ding, Zhenyu Zhang, Shuohuan Wang, Dianhai Yu, Ruirui Li
arXiv preprint, 2026 · Paper

RESUME models video as an evolving compact state: an anchor I-frame initializes the state, while motion and residual signals update it over time. A shared readout connects the state trajectory to video-language modeling.

Can Zhang, Baofeng Zhang, Xiaotian Han, Junyuan Shang, Yuchen Ding, Shuohuan Wang, Dianhai Yu, Ruirui Li
arXiv preprint, 2026 · Paper

VESTA is a training-free long-video agent that routes questions to focused, recall, or contrastive retrieval. Its acquire–verify–consolidate loop maintains a temporal evidence ledger to track provenance, coverage, and conflicting evidence.

• COSTA: Collaborative Open-Set Test-Time Adaptation Through Robust Prototype Learning
  Can Zhang, Ruirui Li. CVPR 2026 Findings Track (CVPRF). Paper
• AIM: Augmentation-Invariant Memory for Robust Open-set Test-Time Adaptation
  Can Zhang, Yunfeng Liu, Ye Wang, Ruirui Li. IJCNN 2026. Paper
• Multi-Cache Enhanced Prototype Learning for Test-Time Generalization of Vision-Language Models
  Xinyu Chen, Haotian Zhai, Can Zhang, Xiupeng Shi, Ruirui Li. ICCV 2025. Paper
• Mitigating Cache Noise in Test-Time Adaptation for Large Vision-Language Models
  Haotian Zhai, Xinyu Chen, Can Zhang, Tianming Sha, Ruirui Li. ICME 2025. Paper
• Double-Shrink: Enhancing Model Robustness under SDN Noise by Reducing Uncertain Confidence
  Naihao Wang, Can Zhang, Yunfeng Liu, Wentao Chen, Ruirui Li. ICME 2025. Paper

• Memory that knows what is missing. How can a video agent represent uncertainty, detect gaps in temporal coverage, and decide when to retrieve more evidence?
• Video representations under a compute budget. Can motion, residual signals, and event structure guide when to update a compact state and when to revisit visual tokens?
• Evidence-sensitive reasoning. How should an agent revise an answer when retrieved clips conflict, and how can evaluations measure whether its answer is supported by the video?

These are directions I hope to investigate during my PhD and through future collaborations.

My research internships at Baidu and the China Telecom Artificial Intelligence Research Institute have involved video agents, multimodal training, data pipelines, and distributed training systems. At Baidu, I am currently working on streaming-video understanding agents. My next planned direction is agentic data synthesis for long-video tasks.

Tools I have worked with: Python, PyTorch, verl, Megatron, MS-Swift, and distributed training workflows.

If your group works on video-language models, multimodal memory, efficient inference, or reliable agents, I would be happy to discuss potential research fit for Fall 2027.