Multimodal Foundation Models · Agentic AI · Video Understanding
About Me
I am a master’s student in Computer Science and Technology at Beijing Normal University (GPA: 4.0/4.0, Rank: 1/71). My research focuses on multimodal foundation models, agentic systems, and structured video understanding. I enjoy turning concrete scientific or product bottlenecks into scalable data, modeling, and post-training pipelines.
Research Showcase · Google Scholar · GitHub
Experience
- Ant Group, AI Safety — Multimodal foundation-model co-training and capability distillation; dynamic-policy generation agent for SingGuard. (Jun. 2026–Present)
- Kuaishou, Main Site Technology — Long-video chaptering for short-drama understanding and downstream trend analysis. (Dec. 2025–May 2026)
- ByteDance CQC — Automated multimodal clustering and group-wise post-training for Douyin risk-control models. (Oct.–Dec. 2025)
- ModelBest, AI4Science — Recursive agent for converting papers and books into hierarchical knowledge trees. (May–Oct. 2025)
Selected Research
- CausalChapter — First author, EMNLP 2026 Main. Intervention-inspired dependency modeling improved CIDEr from 95.36 to 116.88 (+21.52) with the same Qwen backbone. [Paper]
- CurvSeg — Co-first author, ICLR 2026. Curvature-guided task synergy for skeleton-based temporal action segmentation. [Paper]
- SingGuard — Core contributor to dynamic policy generation and the Teacher-Verifier data loop for a policy-adaptive multimodal guardrail. [Technical Report]
- ST-4DGS — Co-author, SIGGRAPH 2024. Spatial-temporally consistent 4D Gaussian Splatting for dynamic scene rendering. [Paper]
- RemapVR — Co-author, CHI 2025. Immersive authoring for rapid prototyping of remapped interaction in VR. [Paper]
Education & Honors
- M.Sc., Computer Science and Technology, Beijing Normal University, 2024–2027
- B.Sc., Computer Science and Technology, Beijing Normal University, 2020–2024
- Jingshi First-Class Scholarship ×4; CCPC Regional Bronze Medal; iGEM Silver Medal; Challenge Cup Silver Medal