Jinhao Chen

Multimodal reasoning & video understanding

Jinhao Chen

Beihang University Beijing, China

I work on multimodal reasoning and video understanding, with interests in self-evolving agents and reliable temporal grounding.

Google Scholar GitHub Email
LANGUAGE → VIDEO → EVIDENCEFIG. 01
From a language query to a grounded video interval An illustrative research schematic. A natural-language query points to a sequence of video frames. A selected interval is marked on a timeline and linked to an interval-level confidence score. It is not an experimental result. A natural-language query LOCALIZE VIDEO SEQUENCETIME → t_startt_end Interval-level confidence Rank · select · accept or reject Where, and how sure?
Grounding language in time.Illustrative schematic

Research interests

Reasoning across modalities

Connecting visual evidence with mathematical reasoning, reflection, and iterative learning.

Explore MathSE

Grounding with confidence

Finding the right moment in a video, with explicit confidence to support controllable selection.

Explore temporal grounding

Agents that evolve

Adapting video agents through evidence-guided feedback and targeted improvements to their design.

Explore MetaVideoAgent

Writing & research

Selected publications

All publications on Scholar
2026arXiv preprintVideo grounding

Grounding with Confidence: Controllable Generative Video Temporal Grounding

Jinhao Chen, Benlei Cui, Ruijian Jia, Ziheng Wang, Tianyu Wo, Pengfei Sun, Longtao Huang, Hui Xue, Yitong Yang, Haiwen Hong

Interval-level confidence for ranking and selecting generated temporal candidates, without an external verifier at inference.

Paper Code

arxiv.org/abs/2609.39883

BibTeX
@misc{chen2026grounding,
  title = {Grounding with Confidence: Controllable Generative Video Temporal Grounding},
  author = {Chen, Jinhao and Cui, Benlei and Jia, Ruijian and Wang, Ziheng and Wo, Tianyu and Sun, Pengfei and Huang, Longtao and Xue, Hui and Yang, Yitong and Hong, Haiwen},
  year = {2026},
  eprint = {2609.39883},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url = {https://arxiv.org/abs/2609.39883}
}
2026arXiv preprintVideo agents

MetaVideoAgent: Automated Video-Agent Evolution for Long-Form Video Understanding

Benlei Cui, Ruize Wang, Junjie Li, Jinhao Chen, Longtao Huang, Yinghao Chen, Yuwen Zhai, Jingqun Tang, Ruijian Jia, Weiwei Wu, Pengfei Sun, Haiwen Hong

Automatically evolving modular video agents through evidence-grounded failure analysis and localized validation tasks.

Paper Code

arxiv.org/abs/2608.04587

BibTeX
@misc{cui2026metavideoagent,
  title = {MetaVideoAgent: Automated Video-Agent Evolution for Long-Form Video Understanding},
  author = {Cui, Benlei and Wang, Ruize and Li, Junjie and Chen, Jinhao and Huang, Longtao and Chen, Yinghao and Zhai, Yuwen and Tang, Jingqun and Jia, Ruijian and Wu, Weiwei and Sun, Pengfei and Hong, Haiwen},
  year = {2026},
  eprint = {2608.04587},
  archivePrefix = {arXiv},
  primaryClass = {cs.CV},
  url = {https://arxiv.org/abs/2608.04587}
}
2026AAAI 2026Multimodal reasoning

MathSE: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning

Jinhao Chen, Zhen Yang, Jianxin Shi, Tianyu Wo, Jie Tang

A self-evolving framework that develops multimodal mathematical reasoning through repeated inference, reflection, and reward-guided fine-tuning.

Paper Project

doi.org/10.1609/aaai.v40i24.39100

BibTeX
@article{chen2026mathse,
  title = {{MathSE}: Improving Multimodal Mathematical Reasoning via Self-Evolving Iterative Reflection and Reward-Guided Fine-Tuning},
  author = {Chen, Jinhao and Yang, Zhen and Shi, Jianxin and Wo, Tianyu and Tang, Jie},
  journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
  volume = {40},
  number = {24},
  pages = {20136--20144},
  year = {2026},
  doi = {10.1609/aaai.v40i24.39100},
  url = {https://ojs.aaai.org/index.php/AAAI/article/view/39100}
}
2024arXiv preprintMultimodal reasoning

MathGLM-Vision: Solving Mathematical Problems with Multi-Modal Large Language Model

Zhen Yang, Jinhao Chen, Zhengxiao Du, Wenmeng Yu, Weihan Wang, Wenyi Hong, Zhihuan Jiang, Bin Xu, Jie Tang

Specialized multimodal models trained on MathVL, exploring how diverse visual information supports mathematical problem-solving.

Paper

arxiv.org/abs/2409.13729

BibTeX
@misc{yang2024mathglmvision,
  title = {{MathGLM-Vision}: Solving Mathematical Problems with Multi-Modal Large Language Model},
  author = {Yang, Zhen and Chen, Jinhao and Du, Zhengxiao and Yu, Wenmeng and Wang, Weihan and Hong, Wenyi and Jiang, Zhihuan and Xu, Bin and Tang, Jie},
  year = {2024},
  eprint = {2409.13729},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  url = {https://arxiv.org/abs/2409.13729}
}

From ideas to implementations

Open-source work

More on GitHub

Get in touch

Let's connect.

For conversations about multimodal reasoning, video understanding, and research.

Scholar GitHub Blog