CV
A visual overview is available in my interactive research showcase.
Education
- Beijing Normal University, M.Sc. in Computer Science and Technology, 2024–2027
- GPA: 4.0/4.0 · Rank: 1/71
- Beijing Normal University, B.Sc. in Computer Science and Technology, 2020–2024
- GPA: 3.8/4.0 · Rank: 8/56
Industry Research Experience
- Ant Group · AI Safety, Algorithm Research Intern, Jun. 2026–Present
- Multimodal foundation-model co-training and capability distillation through large-scale data synthesis and MOPD.
- Built the dynamic-policy generation agent and Teacher-Verifier data loop for SingGuard; dynamic-policy accuracy reached 74.15% (+9.5 pp) over Qwen3-VL-8B.
- Kuaishou · Main Site Technology, Algorithm Research Intern, Dec. 2025–May 2026
- Developed temporally grounded, high-level chapter generation for short-drama understanding and downstream trend analysis.
- First-authored CausalChapter, accepted to EMNLP 2026 Main; improved CIDEr from 95.36 to 116.88 (+21.52) with the same Qwen backbone.
- ByteDance CQC · Douyin Risk Control, Algorithm Intern, Oct.–Dec. 2025
- Added automated multimodal clustering and group-wise training to the post-training loop, without manual categorization; improved the owned risk domain by 4 points.
- ModelBest · AI4Science, Algorithm Intern, May–Oct. 2025
- Built a recursive knowledge-structuring agent with bottom-up abstraction and outline-guided expansion; improved retrieval by approximately 6 points on average over vanilla RAG.
Selected Research
- CausalChapter, First Author — Long-video chaptering with intervention-inspired dependency modeling. EMNLP 2026 Main. [Paper]
- CurvSeg, Co-first Author — Curvature-guided task synergy for skeleton-based temporal action segmentation. ICLR 2026. [Paper]
- Self-Purifying Caching, Student Lead — Training-free audio-visual video event perception under noisy reference samples. Under review.
- SingGuard, Core Contributor — Policy-adaptive multimodal LLM guardrail with dynamic reasoning. [Technical Report]
Honors and Awards
- Jingshi First-Class Scholarship ×4
- Merit Student & Competition Scholarship
- Bronze Medal, China Collegiate Programming Contest (CCPC)
- Silver Medal, International Genetically Engineered Machine Competition (iGEM)
- Silver Medal, Challenge Cup
Skills
- Research: Multimodal LLMs, agentic workflows, video understanding, temporal modeling, model safety
- Post-training: SFT, RL, knowledge distillation, MOPD, scalable data synthesis and verifier loops
- Engineering: Python, C++, PyTorch, Linux, Git
Publications
Curvature-Guided Task Synergy for Skeleton based Temporal Action Segmentation
Guozhang Li*, Xinran Duan*, Mei Wang, Lizhi Wang, and Hua Huang. Curvature-Guided Task Synergy for Skeleton based Temporal Action Segmentation. ICLR 2026. *Equal contribution.
RemapVR: An Immersive Authoring Tool for Rapid Prototyping of Remapped Interaction in VR
it is not available yet.